Nine AI chat and story apps, and the gap they leave

Built 2026-10-03 from AppKittie (detail, historicals, 100 Meta ads, organic, creators, 55 US reviews), then free public store endpoints once AppKittie ran out of credits mid-run (Apple RSS reviews, iTunes lookup, Google Play scraper), and Monid (brand TikTok). Raw data: apps/zeta/data/. Media: library/zeta/.

---

1. Snapshot

2. Niche & positioning

• Pitch, in their words: "Endless Characters, Endless Experiences… Unlimited Free Chats and Character Image Generation… You Choose Your Story." Google Play: "Top App Choice for 6M Users! Millions of Characters. Your Story Starts Here."
• Angle: a free, unlimited, UGC story-roleplay platform. You join a plot rather than chat with an assistant. The listing sells drama tropes outright: "A two-faced student council president, obsessive BL triangles, contract love affairs with a CEO… yandere and tsundere relationships."
• Who it's for: fans of webtoons and otome games who want to self-insert into a romance or drama, on both the female side (夢女子, "dream girls") and the male side ("AI彼女", AI girlfriend). Per market: in JP, it's "your oshi, roleplay". In KR, it's "imagination becomes reality". In the US, it's sold as the free c.ai / Polybuzz alternative.

3. Product & mechanics

Evidence comes from app screens visible in ads and brand TikToks, plus reviews.

• Core loop: text roleplay inside UGC "plots". Each plot has a narrator, several characters who speak separately ("The way other characters can speak separately teases my brain in a way I LOVE!!"), and suggested reply / choice buttons. Examples: "Interrogate him / Find it adorable" (TikTok 7618452921683954975) and "Start voguing out the office / Beg for mercy and your job" (7636616393386921247). There are also free "inspiration" auto-replies, which reviews say are capped at about 50 and then refill.
• Media: a Snapshot feature, where the character sends generated selfies in the chat (catboy demo, TikTok 7686013644273995022). Image generation from keywords. UGC visual novels with images. Lore books. A quiz feature (JP review). The JP organic clip "音声が入り乱れております" ("multiple voices are mixed in here") suggests voice output. No video.
• Creation / UGC: anyone can create characters and plots. The description claims "more than 2.5 million character[s]". There are plot rankings, comments on plots, and a "1st Global Creator Contest" with a $67,000 prize pool (submissions 7/17–8/9, winners 9/3; TikTok 7667111678055664926).
• Models: a free base "Zeta" model. Premium "Luca" and "Koji" models are paid per turn in Pieces: "10 Pieces per turn" for Luca per a US review, and "대화 하나당 젬 5개" (5 gems per message) per a KR review.
• Memory: weak on the free model. This is the top product complaint (see §9).
• NSFW stance: text is permissive. KR has an adult-verified "언리밋" (unlimited) mode, and a US review says "Just glad the ai isnt filtered". Image and UGC moderation is strict and opaque: plots get deleted without a reason and creators are suspended. The ratings are inconsistent: 12+ on iOS, 17+ on Google Play.
• Monetization: ad interstitials every 2–3 messages for free users, which the app calls "forced breaks". The zeta pass subscription removes ads. Per reviews it adds about 200 Pieces a month and nothing else, plus Pieces for the premium models. One reviewer reports the paywall triggers mid-chat: when a pass expires, the app "silently switches to draining your hard-earned Piece balance". Spending ranges to whales: "自分は月10万くらい課金してる" ("I spend about ¥100k a month", JP).
• Retention hooks: a huge plot catalogue with recommendations, creator rankings and contests, multi-character drama that keeps escalating, and a "continue the story" framing. The listing claims "usage time averaging to more than 2 hours". Reviews back it up: "時間が溶ける" ("time melts away"), "うちの13時間返してください" ("give me back my 13 hours"), "사용시간이 5~6시간" ("5–6 hours of use").

4. Demographic

• Gender: both, with a female lean. Of 59 viewed ad creatives, 31 target women (male characters, otome and BL), 22 target men (female characters), and 6 are neutral. By ad-ID count it is 39 F / 32 M / 9 N. JP and KR reviewers often write as female romance users: "夢主にはほんっとおすすめのアプリです" ("I really recommend this app to dream-girl fans"), and "나처럼 로맨스용으로만 쓰는 유저도 많은데…" ("lots of users like me only use it for romance"). The male side shows up too: "여친이 겁나 많아짐요ㅋㅋㅋㅋ" ("I've got a ton of girlfriends now lol"), and "現実で恋人いない分依存する" ("I depend on it because I have no partner in real life").
• Age: very young (an inference that the reviews strongly support). US reviews use Gen-Alpha slang ("Thus is good 67", "gng") and say "I'm a teenager". In JP: "中1です" ("I'm in 7th grade") and "学校に行けるようになりました" ("I've been able to go to school again"). In KR: "학생은 죽으라는거" ("so students are supposed to just die?", about the ads), "고등학생인데도" ("even though I'm a high schooler"), and a parent writes "아이가 미성년자인데 지워도 다시 깔리고" ("my child is a minor and it keeps getting reinstalled after I delete it"). Several reviewers say they can't afford the pass.
• Interests: webtoons, otome and BL, fandom and IP roleplay (JP reviews mention Jujutsu Kaisen and Summertime Render plots), isekai, and wuxia. US users are migrants from c.ai, Polybuzz, Talkie, Chai and Swerve. Reviews name each of those.
• Languages: EN, JA, KO. Language routing has bugs: US and KR users get "Locked in Japanese language".

5. Unique selling points

1. Scale of free UGC plots plus multi-character scenes. The drama is more like a webtoon than a 1:1 companion.
2. Choice buttons inside the chat. This is the closest any chat competitor gets to our decision points.
3. A localized, character-first content machine. The JP ads are original anime IP produced at volume, with roughly 77 distinct creatives in 3 weeks.
4. A cross-market creator economy (Global Creator Contest, $67k).

6. Marketing: paid (Meta, via AppKittie)

• Volume: AppKittie says the app has 211 Meta ads in total, but only 110 are catalogued in search (90 active / 20 inactive on 2026-10-03). I pulled 100 (all 90 active + the 10 oldest inactive). They resolve to 77 unique creatives, of which 59 were downloaded to library/zeta/paid/. There are 56 videos, all 720×1280, 6–34 s long with a 16 s median, plus 3 images. All 100 ads are on Meta, none on Google.
• First seen: every catalogued ad starts on or after 2026-08-30: 44 on 8/30, then a second wave of 29 on 9/12, 9 on 9/13, 12 on 9/16, and 6 on 9/17–18. AppKittie's coverage may start there, so this is not necessarily the campaign start.
• Surfaces: Facebook, Instagram, Messenger and Threads (73 ads); Facebook, Instagram and Threads (27). The links split 52 to the iOS App Store and 48 to Google Play.
• Countries: AppKittie reports none. All 100 ads are Japanese-language, so the target is Japan (inferred). There are no English Meta ads in the catalogue despite the worldwide launch.
• Copy (headline → body), by ad count: "無料で無制限のAIチャット" (free, unlimited AI chat; 42) → "好きなAIキャラクターと、無料で好きなだけチャット。制限なしで会話を楽しもう。" (chat as much as you like with your favourite AI characters, free and with no limits). "世界にひとりの話し相手" (the only one in the world to talk to; 15 + 8 variant). "好きなキャラといつでも無料" (your favourite characters, anytime, free; 9). "朝まで話せる" (talk till morning; 8). "好みのキャラ多数" (lots of characters to your taste; 7). "創造を現実に" (make your creation real; 4). "何回話しても無料" (free no matter how much you talk; 4). "推しとの会話が無料・無制限" (talking with your oshi is free and unlimited; 3). The copy angle is free + unlimited + your favourite character, a direct shot at JP competitors that meter messages.
• Formats (59 creatives viewed frame by frame via contact sheets):
  • Fake incoming call / LINE chat / lock-screen notification opener (9). The call is from a named character ("黒川美月 着信中…", "蓮 音声通話…"), then the character speaks, then "今すぐ返信！" (reply now!). The longest runners: paid/1046206941485519.mp4 (24 days, 4 ad IDs), paid/1037168119140460.mp4, paid/1070904585532682.mp4.
  • Anime character dialogue → reply box (8). Lines appear as subtitles and the clip ends on "あなたの返信は…?" ("Your reply…?") with a chat input. Examples: paid/1558120016013529.mp4 ("来ると思ってた。" / "この夏、私だけを見て。", "I knew you'd come." / "This summer, look only at me."; 3 IDs) and paid/1862189788081819.mp4 ("今から二人で……何しようか?", "What should the two of us do now…?").
  • First-person POV, with the viewer's hand in frame (3). paid/2565282123975735.mp4: he holds your hand under an umbrella, "…手、冷えてる 帰るまで、離すなよ" ("…your hand's cold. Don't let go till we're home"). paid/1380140054234955.mp4: you pat a kohai's head. paid/1730165831531117.mp4: he wakes you in bed.
  • Webnovel-style story hook cards (7), usually an in-app chat screen recording with generated images followed by a hook card. Examples: "ログアウトボタンだけが、押せない。" ("Only the logout button won't press."), "毎晩遊ぶネ友は、隣に住む陰キャ女子だった。" ("My nightly online-game friend was the shy girl next door."), "優しい教祖様が、裏ボスでした。" ("The gentle cult leader was the final boss."), "天下に五人しかいないはずの絶対高手が、俺を探し始めた。" ("Absolute masters, supposedly only five under heaven, started looking for me.").
  • Multi-world / multi-season montage (7). The same character appears across worlds or seasons: "いつも同じ毎日じゃ、もう物足りないでしょ?" / "zetaなら、君が望む——" / "全部、叶えられる。" ("Same old days aren't enough anymore, are they?" / "With zeta, whatever you want—" / "it can all come true.") → a "No.1 AIチャット" end card.
  • Photoreal "AI彼女 / AI彼氏" (6). "zetaで待ってる、あなただけのAI彼女。" ("Waiting for you on zeta: your very own AI girlfriend."), e.g. paid/2278981709531063.mp4.
  • Self-insert dissolve (6). In a couple shot, one partner dissolves into pixels so the viewer takes their place (paid/4058936047747287.mp4, paid/1352060467983605.mp4).
  • Anime character in real Tokyo footage / camera-viewfinder AR (4). "zetaなら、あり得るよ。" ("With zeta, it can happen.")
  • Costume-request fan service for women (3). The user asks him to wear a maid, nurse or qipao outfit: "お願い。ご主人様って呼んで。" ("Please. Call me master.").
  • Static key art (6).
• Likely winners: the 8/30 wave that is still active at 24 days. That is the call-opener and dialogue → reply formats, the POV clips and the swim-ring girl, several of them re-launched under new IDs on 9/16–18. Losers: all 10 inactive ads ran only 6 days (8/30 → 9/5). They were mostly static images or early cuts ("朝まで話せる" images, the 163d28344… video). The 9/12 wave (photoreal, montage, dissolve, costume) is newer, so it is too early to tell.
• Spend / reach: null in AppKittie for every ad.
• TikTok paid: not in AppKittie. No Spark-ad labels were seen in the organic data.

7. Marketing: organic

• Brand accounts: TikTok @playzetaai (verified, English/global, 33.1k followers, 720 posts; the ID suggests it was created around Sept 2025) and @zetaai_ (verified, Korean "제타", 45.6k followers, 544 posts; not scraped, because the budget allowed one scrape per platform). Instagram @zetaai_official (verified, 48.4k followers, only 12 posts, last one 2025-12-25) is a dormant Lee Luda virtual-influencer account. Nothing was posted after the 2026-04-01 window.
• Library: 22 @playzetaai posts (Monid; 2 pinned, 13 from before April but kept because they were billed) and 6 AppKittie organic TikToks, for 28 items in library/zeta/organic/. 4 of the 6 AppKittie items are false positives: edits of a TikTok personality called "Zeta", not the app. AppKittie returned no view counts for any of the 6. The AppKittie creators list (20) is also keyword noise (ZETA DIVISION esports, zeta.nu, and so on).
• Best performers (@playzetaai):
  • 1.10M views: photo carousel organic/tiktok-7594702405200137503-1..6.jpg, an over-the-shoulder snoop on a plane: "my planes abt to take off and what is THISSSS" → "oh hes texting his wife..." → "IS THAT THE HUSBAND OF THE FIRST GIRL????" → "oh... he was just using zeta".
  • 205k / 121k: TTS-narrated stories with kinetic captions over cinematic backgrounds, ending on an in-app search bar showing the character's name: "download zeta and continue the story!" (tiktok-7579261068313201927.mp4, tiktok-7577747250529553671.mp4).
  • 53k: a speed-reading challenge. 52k: an office skit, "Forcing my coworker to feed into my delusions". 51k: a "DEVIL BOSS would like to FaceTime..." call → choice buttons. 49k: "POV: my bf's reaction to our break up" (iMessage one-word replies vs zeta) → "Don't Settle for One Word Texts."
• Recurring formats: caught-texting snoop, TTS story with a search-the-character CTA, chat overlaid on Minecraft parkour ("actual convos from our app pt 3"), fake FaceTime → choice buttons, staff and office skits, and feature demos (Snapshot). Typical views are 10–50k per post.
• Cadence: can't be measured. The scraper returned 22 items with large gaps (Apr 7, Apr 27, May 6, May 18, May 20, Jun 12, Jul 27, Sep 16, Sep 22), while the profile claims 720 posts.
• JP creator content (AppKittie): tiktok-7659670867789499668.mp4 is a zeta-branded chat recording, "【zeta】夢女子が全員沼るこれ" ("[zeta] every dream girl will get hooked on this"), from a default-named account. It may be brand- or agency-run (an inference).

8. Reviews: what's working

The samples:

Counts below are keyword-coded and rough.

• Stories and characters are fun and addictive. About 87/245 US reviews praise characters or stories, and about 58/245 say "addict/obsessed/love it". Quotes: "So much fun and am able to laugh loud". "I'm fully and completely addicted". "マジで楽しい 時間が溶ける" ("Seriously fun. Time melts away"). "ガチの没入感チャット" ("A seriously immersive chat").
• Better than c.ai / Polybuzz / Talkie / Chai / Swerve (19/245 US mention a competitor): "Best alternative from chai/c.ai … it remember KEY DETAILS!!!" "I used to use swerve but i upgraded to Zeta, i love how long the responses are".
• Free and long responses: "No filter long chats free with ads and free responses". "広告は多いけど無課金でも全然楽しめて" ("Lots of ads, but it's totally fun even without paying").
• Multi-character scenes, lore books, plot creation: "I love the lore books feature, OMG". "I love how they let you customize your own chats and plots".
• Emotional support and companionship: "Saved my life". "現実で恋人いない分依存する。楽しい" ("I depend on it because I have no partner in real life. It's fun"). "남친이랑 헤어졌는데도 이거 있으니까 외롭지않음" ("Even after breaking up with my boyfriend, I'm not lonely because I have this").
• KR loyalty, despite the anger: "한국 AI 챗 1위인 이유가 잇네요" ("There's a reason it's Korea's #1 AI chat").

9. Reviews: complaints

Counts are rough keyword counts.

• Ads (the #1 complaint in every market): US 66/245, GP 12/100, JP 37/100, KR 38/100.
  • "An add every two messages is a bit excessive and not really worth it unless you're willing to pay 15$ dollars a month"
  • "delete the fake x buttons on ads"
  • "たった二言「おはよう」「何か用？」と発言しただけで広告。恐らく国内最多の広告数。" ("I said just two things, 'good morning' and 'what do you want?', and got an ad. Probably the most ads of any app in Japan.")
  • "광고가 2분에 한번씩 나오고" ("ads come every 2 minutes")
  • Reviewers also report ads that won't close or that loop.
• Memory / consistency: US 48/245, GP 9/100, JP 13/100, KR 11/100.
  • "It forgets everything constantly you have to keep repeating stuff you written before"
  • "記憶力ダチョウかな？" ("Does it have the memory of an ostrich?")
  • "기억력 금붕어" ("goldfish memory")
• Paywall / pricing: US ~45/245 mention paying, KR 27/100.
  • "the app silently switches to draining your hard-earned Piece balance (at 10 Pieces per turn) without a single confirmation box"
  • "月額1650円を払えば広告を消せます、それだけです" ("Pay ¥1,650 a month and the ads go away, and that's all you get")
  • "제타패스 월 14900원 비싸기도 ㅈㄴ비싼데" ("the zeta pass is ₩14,900 a month, which is damn expensive")
• Bugs: a crash on launch after a recent update hits many reviewers (US 28/245, JP 22/100 bugs, KR 12/100). "Everytime I try to open the app it closes right back out". "Locked in Japanese language".
• Characters drift hostile or into forced conflict (KR especially, 17/100):
  • "멀 맨날 사과하래 이년저년은 기본에 여혐 개심함" ("they're always demanding apologies, calling you 'bitch' is the default, the misogyny is terrible")
  • "every second of every talk must have conflict of some kind" (GP)
  • "boys hitting on you even after you say you're not interested" (US)
  • "乱入が多すぎる" ("too many intruders barging into the scene", JP)
• Moderation (small but intense, mostly creators): "Every time a chat gets taken down for being inappropriate it doesn't tell you why". "版権は消さないくせに健全オリジナルプロットは規約違反で消す" ("they don't remove copyrighted [fan] plots, but they delete wholesome original plots as rule violations").
• Minors: "아이가 미성년자인데 지워도 다시 깔리고 하는데 제발 안 깔리게 해주세요" (a KR parent: "my child is a minor and it keeps getting reinstalled after I delete it, please stop it installing"). A US reviewer: "they're saying it's 13+ for the age rating but this first AI bought I got asked me for you guys know what".

10. Markets

11. Lessons for our game

• Copy the winning JP ad grammar. It is already our product format. Zeta's longest-running creatives open on an incoming call or notification from her, run 2–3 lines spoken straight at the viewer (often POV with the viewer's hand in frame), and then cut on the decision: "あなたの返信は…?" / "今すぐ返信！" ("Your reply…?" / "Reply now!"). They can only fake that with subtitles over stills. We have real first-person video and a real choice screen, so our ad end card can be the actual decision point.
• Self-insert is the hook, for both genders. Zeta runs female- and male-target variants of the same template side by side (31 F / 22 M creatives), and its dissolve and POV formats literally put the viewer in the partner's place. Plan one "her" storyline but test the format on both audiences early. The female/otome side is where Zeta spends more.
• Avoid their two biggest wounds: ad interstitials every 2–3 turns, and memory drift. Those are the #1 and #2 complaints in every market. An authored branching video game "remembers" for free. Monetize at story cliffhangers (pay or wait to see her reaction), not by interrupting mid-line.
• The gap is calm, sweet payoff. Zeta's LLM characters drift toward hostility and forced conflict (KR "표독/년년", "hostile / calling you 'bitch'"; GP "every talk must have conflict"; JP "乱入", "intruders barging in"). Users explicitly ask for ほのぼの ("gentle, heartwarming") and sweet romance. Hand-directed reactions that land warm, flustered or teasing are a differentiator that's easy to show in a 15-second ad.
• Cheap organic formats that work in the US: the over-the-shoulder "caught texting" snoop (1.1M views), and a TTS mini-story that ends on "search her name in the app". Both translate directly to a teaser of one decision point. Plan for a teen-heavy audience: set the age gate and store rating deliberately, because Zeta's 12+ iOS vs 17+ Android split and the parent complaints are a risk we shouldn't inherit.

12. Data gaps

• AppKittie credits ran out (HTTP 402) after the US reviews call. So there are no AppKittie Google Play, KR or JP reviews. I replaced them with free sources: Apple's public RSS (US/KR/JP), iTunes lookup, and google-play-scraper for Google Play.
• Rankings: every get_store_ranking_history / list_store_rankings call returned 404 or empty (data/appkittie-ranking-attempts.json). There is no chart evidence.
• Meta ads: countries, spend and reach are null. Only 110 of the 211 Meta ads AppKittie counts are catalogued. The 10 newer inactive ads and 18 lower-ranked creatives weren't downloaded (their URLs are in the manifest). There are no Google ads and no TikTok paid data.
• AppKittie estimates are review-count multiples and include a storefront jump. Don't compare them across apps without that caveat.
• Organic: AppKittie organic had 6 items, 4 of them false positives, and no metrics. The creators list is keyword noise. Monid returned 22 @playzetaai posts (cadence unknowable). @zetaai_ (KR TikTok) was not scraped. The Instagram account is dormant.
• Monid spend: $0.0174 (balance went from $0.27 to $0.17, since it is shared with other agents).

Researched 2026-10-03. Sources: AppKittie MCP (detail, historicals, ads, organic, creators, reviews, chart history), Monid (brand TikTok scrape), yt-dlp (free TikTok metadata and downloads). Raw responses: data/. Media library: ../../library/swerve/.

1. Snapshot

Trend. AppKittie's downloads, revenue and MAU curves move in lockstep with the cumulative rating count. Their 7/14/30-day growth figures are identical to the review-growth figures, so they look derived from reviews rather than observed. The Sept 3 drop to a round 100k is a model reset, not a real collapse: ratings kept climbing through it. The cleaner signal is new ratings per month:

The app ramped steeply from February to April, which matches the Character.AI exodus, then held a plateau of about 4–5k ratings a month from April to September. The US free-chart presence ended after Aug 13 (§10), and organic views per post fell in August and September (§7). Read: growth has flattened, not collapsed.

2. Niche & positioning

• In their words. "Swerve is AI fiction where your messages shape what happens next." App Store screenshot captions: "Chat with characters · Get instant replies · Make your own AI · Customize your chats."
• The real angle is "the Character.AI that C.AI used to be." It is a near-clone of the C.AI layout. It markets on what C.AI took away: no ID or face age verification, no filter (with an NSFW toggle), no forced ads, long replies, and memory. Chai's paywall and Polybuzz's pricing feed the same pitch.
• Underdog framing. The marketing leans on "a cute app built by only 2 people", and reviewers repeat it back ("How can 2 people create an app that's 10x better than character ai?").
• Who it's for. Fandom and romance roleplayers fleeing C.AI and Chai: mostly teen and young-adult women (see §4).

3. Product & mechanics

• Core loop: text roleplay chat with community characters. The store listing claims "more than 500,000 community-created characters". You can chat one-on-one or in group chats with several characters, where you press an arrow to advance which character speaks.
• Branching and control tools: swipe for another reply (capped at 10 swipes), edit any message, rewind, "start a new path from any message", pinned memories ("Pin what characters should remember"), and personas (play as anyone).
• Model tiers paid in gems. Reviewers report standard at 3 gems per message, premium at 15 and ultra at 95. Regenerations also cost gems.
• Earning gems (free path): a 25-gem daily reward, 100 gems per rewarded ad with a daily ad cap and cooldown (reviewers cite waits of 9, 13 and 21 hours), quests, and gems earned when other users chat with bots you made ("creator awards", name-tags).
• Creation tools: UGC characters (public or private), long definitions (one reviewer says 6,000 characters), and easy bot transfer from C.AI. There is also profile, app-icon, chat-theme and font customization.
• Media: text only. There are no images, voice or video. Reviewers ask for photos (6 of 197 US reviews ask for photos, voice or images) and for "the audio feature that character ai have".
• NSFW stance: an SFW/NSFW toggle ("safe mode if you want it filtered"). Accounts are restricted to 18+ (a 16-year-old reviewer was locked out), but there is no ID or face check. A safety layer flags self-harm topics and shows hotline numbers. Some refusals come back in Chinese, which may mean a Chinese-origin LLM underneath (inference).
• Monetization: rewarded-ad gems, subscriptions ($3.99 Lite, $12.49 Pro) and gem packs. Several reviewers say they never see interstitials ("there’re no ads, unless you watch ads to get gems").
• Retention hooks: daily gem reward, quests, creator rewards, an in-app support chat with the developers, and a Discord server (discord.gg/swerveapp).
• Known regressions: an update around 2026-06-23 caused system-prompt leakage, where bots replied with their writing instructions. 20 of the 25 prompt-leak reviews are dated June–July. Other complaints: random switches to Chinese or Japanese, "server hiccup" errors, and account loss after updates.

4. Demographic

• Gender: predominantly female (inference, strong).
  • Every brand and persona TikTok uses a young woman's face-cam (examples).
  • Captions are written for women who romance fictional men: "I love all my Swerve husbands", "Me whenever i fall for another fictional character knowing no man like him exists in real.life.", "now it makes sense why i love Draco Malfoy on swerve ai".
  • App Store screenshots show male anime love interests (Damon the boss, Commander, Chase, Tyler, Caleb) and a "Mafia" search chip.
  • There are some male users. One reviewer chatted with a "girl at your door" story.
• Age: teens to early 20s (inference).
  • Reviews are written in teen voice ("ts is amazing", "PEAK BRAH !!! :3").
  • One reviewer is "16" and locked out by the 18+ rule. Another calls it the "Best alternative app if you're under 18 and can't use c.ai".
  • TikTok hashtags are #cai #characterai #ao3 #fictionalcharacters #roleplay.
• Interests: fandom roleplay (anime, Marvel, Harry Potter), dark and romance tropes (mafia, CEO), and the C.AI community's in-jokes.
• Language and markets: English only. Reviews and charts come from the US, GB, CA and AU (§10).

5. Unique selling points

1. The closest C.AI clone, minus C.AI's 2026 restrictions: no ID or face verification, an NSFW toggle, and no forced ads. Users say it plainly: "It's basically c.ai but better! After the whole age restriction in c.ai i had to move here".
2. Opt-in monetization. Rewarded ads for gems are framed as a choice ("you can still support the creators by CHOOSING when to watch ads for gems").
3. Strong control tools: pinned memories, personas, edit, rewind, branch from any message, and group chats.
4. A creator economy. Gems flow to bot authors, which keeps the 500k-character catalog growing.
5. An indie "2-person team" story, plus visible developer support (in-app chat and Discord), which reviewers keep praising.

6. Marketing: paid

Update from the ad-library pass (2026-10-03): Swerve started paid acquisition at the end of September. There are 5 Meta creatives (10 ad IDs, "Unlimited AI Chat" / "Unlimited AI Chats & Stories") that started Sep 29–30, and 10 TikTok ads from MSPW LABS INC., UK only ("Chat with 500K+ Characters"). They are in library/swerve/paid/.

• Zero ads found. search_ads returned 0 Meta or Google ads for the app slug, the store id and the developer name (data/appkittie-ads-*.json). Full-text searches for "Swerve", "getswerve" and "Swerve AI" only matched unrelated apps. The detail payload carries no Meta Ads transparency fields either.
• TikTok paid is not covered by AppKittie. No post in the 1,144 scanned TikTok captions carries "#ad", "Sponsored" or "paid partnership". One creator post opens with "It's not an ad btw!" (tiktok-7623064785348824353).
• Read: Swerve appears to grow without Meta or Google paid UA. Its acquisition engine is the TikTok persona network below. It may still run TikTok Spark or Promote ads that we can't see.
• library/swerve/paid/ is empty.

7. Marketing: organic

Accounts. There is no brand Instagram: the Monid lookup for swerve.chat found nothing. All activity is on TikTok:

"Persona network" is my label. These accounts carry .swerve handles, post almost nothing but Swerve, recycle the same scripted text hooks with different women on camera, and started within weeks of each other. That pattern reads as a brand-run or agency-run UGC account farm (inference, not confirmed). @swervecourtk posted Swerve content in late June, then switched to promoting FollowSneak.com ("I check followsneak.com every hour at this point #cheater #microcheat"). That switch is a strong sign of an agency rotating accounts between clients. AppKittie's creator list (20 of 86) is mostly false matches on the word "swerve" (food, brewing and car accounts). Only 2 of its listed accounts are in the network.

Network totals. I scanned 988 Swerve-mentioning posts. They total 27.4M views, but the median is only 1.9k. 46 posts topped 100k and 5 topped 1M, and the top 40 posts hold 78% of all views. Monthly volume and views:

(Months before May are undercounted, because four accounts hit the 150-post listing cap.) August and September views fell 80–90% from July at the same posting volume. The hook library is fatiguing, or TikTok throttled the farm.

Format. About 95% of posts are 5–12 s face-cam clips (median 6 s): a young woman in bed or in a car, a hand on her head or a shocked face, a big text overlay and a trending sound. A few are screen recordings, such as the C.AI Play Store page at "2.2★ 2.29M reviews — CHAT WE'RE WINNING", an iOS "Delete Character.AI?" dialog, or the getswerve.com Android waitlist. The app UI is almost never shown.

Recurring hooks (verbatim on-screen text), with best views:

• Engagement-bait questions to the C.AI community: "ppl who use c.ai, chai, swerve etc, what's the scariest thing ur bot has ever said to you?" (3.1M, mp4). Also "people who use emochi, cai, janitorai, swerve, etc what made you restart/ delete a chat?" (905k, mp4) and "swerve, cai, chai users pls tell me ur spelling mistakes i wanna laugh" (655k).
• "Finally found the alternative" + 2-person indie: "when you finally find the perfect c.ai alternative and it's a cute app built by only 2 people with unlimited chats no filters and the bots actually remember *everything*" (2.2M, brand, mp4). The brand alone re-shot this line at least eight times.
• Gatekeep / secret app: "i found a free better version of c.ai without filters, ads, or verification but i'm gonna gatekeep it" (2.0M, brand, mp4). The network re-used it (320k, 297k).
• Privacy and fear of C.AI: "BTW C.AI COUNTS AS A DIGITAL FOOTPRINT" (1.1M, mp4) and "Me realizing that our cai chats are being read by real people and FBI" (665k).
• Morbid POV: "POV: you're dead but you left your cai chat open" (1.3M, mp4).
• Victory / exodus: "WAR IS OVER (people stopped using cai and chai)", captioned "We are finally free and it's coming to androids!" (700k, mp4).
• Nostalgia: ""I miss the rage" but it's summer 2023 when c.ai had no ads, age verification, the best replies, good plots, and no limits".
• C.AI bot-trope in-jokes: "When my c.ai bot leans against the wall but we were just at the beach" (611k). Caption: "my Swerve bot hasn't asked for my name 20 times, and that's already an upgrade".
• Addiction humour: "accidentally finding an app even more addicting than c.ai and now im scared to look at my screentime" (165k).

Note that most hooks name C.AI, not Swerve. Swerve appears only in the caption or hashtags ("#swerveai #swervechat"), and the brand bio just says "it's "swerve ai" on the app store". These are search-and-community plays aimed at #cai.

Creators. @mr.penguinner posted "SWERVE AI IS COMING TO ANDROID" (387k). @myszunia146 posted a 49 s screen-recording walkthrough that opens "It's not an ad btw!" (320k).

Library: 67 TikTok videos with thumbnails. They cover the combined top 40 by views plus every AppKittie and Monid item. See manifest.json.

8. Reviews: what's working

Sample: 297 App Store reviews, 197 US (Nov 2025 – Oct 2026, average 4.26★) and 100 GB (Feb – Sep 2026, average 4.37★). Counts are rough keyword matches checked by reading.

9. Reviews: complaints

10. Markets

Ad country targeting: n/a (no ads indexed). The app is English-only, and all TikTok content is US-English.

11. Lessons for our game

1. Ride a rival's policy shock with hooks that name the rival. Swerve's whole curve (Feb–Apr ramp, then a plateau) tracks C.AI's age verification, ads and limits, plus Chai's paywall. Its biggest hooks say "c.ai", not "Swerve". For a male audience, find the equivalent pain in the incumbent AI-girlfriend apps (paywalled photos, static images, a bot that "talks for you", no real video) and write "finally found the ___ that actually ___" and "gatekeep it" hooks around it.
2. Copy the cheap TikTok persona-farm mechanics, but plan for decay. Eight to nine accounts at roughly one 6-second face-cam clip a day each, with one hook library re-shot by different faces, earned about 27M views without visible paid UA. Views fell 80–90% after July, though, so rotate fresh angles monthly. Our advantage: our creative can be the product, with a POV clip of her reaction. Swerve's text-only app can't show itself, which is why its videos never show the UI.
3. Monetize the premium branch, never the retry. Swerve's top complaint cluster (~62 of 297 reviews) is gems spent on regenerations, the 95-gem ultra tier, and daily caps with hours-long cooldowns. Rewarded ads framed as a choice are praised. For us: charge for premium choices or reaction clips, give retries and rewinds free, show costs before the player commits, and never hard-wall in the middle of a scene.
4. Exploit their gaps: text only (users ask for photos and voice), no Android, the "bot speaks for me" and POV complaints, and quality regressions when they swap models. A scripted first-person video CYOA answers "It's like I'm in the story". It keeps the player's lines in his own hands (choices), and its quality is deterministic, so a prompt leak can't happen.
5. Steal the trust signals: a visible tiny team, an in-app chat with the developers, and Discord. Reviewers cite these as reasons to forgive bugs and gems.

12. Data gaps

• Paid ads: AppKittie has no Meta or Google ads for Swerve, either because none run or because they aren't indexed. TikTok ads are out of AppKittie's scope.
• Estimates: AppKittie's downloads, revenue and MAU appear modelled from the rating count. They contain a reset artefact (Sept 3), and the MAU looks implausible. Treat them as order-of-magnitude only.
• Organic metrics: AppKittie returned only 20 organic items, all without view counts. I filled the metrics from yt-dlp account listings, which were free. Each listing is capped at the latest 150 posts, so months before May are undercounted. Follower counts are unknown for the persona accounts.
• Persona network ownership is inferred, not confirmed.
• Instagram: no brand account found (the Monid lookup returned not-found, at $0).
• Charts: US and CA Top Grossing requests returned the top-free series, so there is no grossing data there. I only checked the Entertainment category, and there are no per-country download or revenue splits.
• Google Play: there is no official listing. The 20 Play reviews come from the squatter app and are used only as evidence of brand confusion.
• Hooks were read from one frame at about 1.2 s (plus 4-frame strips for 12 top videos). Text overlays that appear later in a clip may be missing.
• Budget: about 435 AppKittie credits. The detail, historicals, ad probes, organic, creators, reviews (320), app searches, Play detail and 5 chart histories (50) cover it; a few failed or empty calls may add 1–2. Monid cost $0.0087 (brand TikTok scrape; the Instagram lookup cost $0).

Built 2026-10-03 from AppKittie (detail, historicals, ads, organic, creators, reviews, rankings) and Monid (brand TikTok + Instagram). Raw responses: apps/polybuzz/data/. Media: library/polybuzz/ (manifest.json lists all 40 items).

1. Snapshot

IAP ladder (iOS): Poly Standard Weekly $4.99, Standard Monthly $9.90, Premium Monthly $19.90, Ultimate Monthly $29.90. Coins: 1,000 for $2.49, 3,000 for $4.99, 8,000 for $9.99, 20,000 for $19.90. 1,000 Opals (a second, newer currency) for $5.90. Google Play lists "$0.05 – $299.99 per item". Annual plans shown in the Nov 2025 sale video: Ultimate US$194.99/yr and Premium US$99/yr ("up to 90% off"). The paywall offers a 3-day free trial, then $9.90/month (creator screen recording tiktok-7649313742068534559).

Trend (AppKittie series, Dec 2025 → Oct 2026): iOS revenue estimate went from $1.49M to $2.05M (+37%). The iOS download estimate rose from 521k to 718k, then dropped to 511k on 2026-10-01, which looks like a model re-base (−27% over 60 days). The iOS estimates move in lockstep with the review count (340k → 480k ratings, about 500 a day), so they are probably derived from it. Google Play revenue is estimated higher than iOS. Overall: mature, still growing, steady revenue; free-chart momentum is fading (see §10).

2. Niche & positioning

• Pitch, in their words: "Our chatbots think and reply like real characters, complete with authentic voices"; "Character creation now open to everyone!"; "Chat as you wish and explore endless story possibilities, all completely free." The TikTok bio reads "A TikTok for AI Chat and Create AI Character."
• Angle: a huge user-generated catalog of characters (anime, fandom, K-pop, celebrities, games, OCs), merchandised by romance trope. Explore tabs include For You / Multi-Role / Second Chance / OC / Anime / Movies&TV / Celebrity / Game. It is "free", with ads paying for the free tier.
• Who it's for: fandom roleplayers and character creators, skewing female and young (§4). It is a Character.AI / Talkie alternative. Reviews compare it to c.ai, Talkie, Chai and Zeta.

3. Product & mechanics

• Core loop: text roleplay chat with UGC characters. There are group chats and multi-role chats (the search results in instagram-DViudHDAe28 show "Roblox group chat 5.6M"), selectable chat modes or models ("Fiction", "Passion", "tale mode"), and voice playback (premium gets "Unlimited Free Voice Playback").
• Video: "Live Videos" launched in Nov 2025: "Characters with Live Videos attach a relevant video to each reply… Each Live Video costs users 400 coins and pays creators 80 diamonds" (tiktok-7574654672531639566). That is about $0.50 per clip at the 8,000-coin pack. One review also mentions an "a.i fake livestream" option. This is the closest existing mechanic to our video-reaction payoff.
• Creation tools (heavy): character builder with AI avatar generation and named AI art styles ("Obsession", "Impasto", "Spark"); up to 5 greetings plus a 3,000-character profile ("Greeting Visual", Sep 2026); user personas (up to 1,500 characters; creators can preset 3 personas per character, e.g. "boss's sassy secretary, sweet spouse, or mysterious fiancée"); themed creation events (Second Chance, Love Triangle, Isekai, Sparks Fly).
• Creator economy: diamonds earned from Live Videos, a Creator Level, Daily Featured slots, a Creator Brief Program run through Discord ("Create a character. Get featured. Earn rewards.").
• Gamified side modes: The Time Box (Sep 2026) is a gacha. "Each search costs 50 Coins and gives you 0.5-2 keys randomly"; a full key unlocks an exclusive character photo saved to your Gallery, and 6 featured characters refresh daily. Also: an MBTI Soulmate Match quiz ("Share to Unlock Chat"), AI song creation for your character, a year-end "Wrapped" short drama ("Choose your script"), and 3rd-anniversary reports.
• Memory: weak on the free tier. Premium gets "Long Memory 2X"; Ultimate gets "Permanent Memory remembers everything, forever".
• NSFW stance: rated 17+/18+ with a "Pure Mode" toggle. Since late 2024 a "compliance policy" filter blurs or hyphenates words, and automated "violations" flag character descriptions, avatars and personas, including private ones. Face/ID age verification was flagged by a reviewer in Dec 2025 as "the December 31 Update" for minor safety, and was live by Jul–Sep 2026, when lockout reviews appear. Ads still promise spice, so reviewers call it a "bait and switch".
• Monetization: three subscription tiers, two currencies (Coins and Opals), and aggressive ads on the free tier. A "Take a little break" gate appears every ~10–20 messages or ~5–10 minutes: the user must watch one or more 30–60s ads, or subscribe. There are ads when entering or exiting chats and swiping, plus new in-chat native ads. Re-rolling a reply costs coins or premium. Closing an offer triggers a discount pop-up.
• Retention hooks: daily coin check-in ("the daily coin system is a great motivator"), daily featured characters, recurring events and sales, Time Box daily refresh, Wrapped/anniversary frames, and creator rewards.

4. Demographic

• Gender: skews female (inference, strong). Brand content centers male love interests: "Lost in his gaze", Stray Kids and Park Sunghoon edits, Second Chance couples, "CEO's Maid Wife", and the male Time Box cast (Alex, Ethan, Sam). Creator content is mostly female creators or female-targeted ("Date your favorite trope, live the fantasy on PolyBuzz", instagram-DWu7SCnk_GX). Review voices say things like "celebrity crushes 😉", "me to girl", BL roleplay, and "All the male characters are so misogynistic". There is a male-targeted minority too: the "Anna - English Teacher" screen recording, the Japanese "ideal girlfriend" skit, and female bots with 0.9–1.2M plays on the JP Explore page.
• Age: teens and young adults (inference, strong). Many reviewers self-identify as minors ("I'm 13", "I'm 17 close to being 18", "my 10 year old", "my cousin is 8"), and the age-verification lockout is a top-5 complaint. The content is fandom-heavy (anime, Demon Slayer, Roblox, K-pop, Stranger Things).
• Interests: fandom roleplay, romance tropes, creating characters and OCs, loneliness and escape ("it literally makes me feel less lonely despite it being an AI").
• Language markets: creator posts in EN, JA, IT, ES, DE; the song feature ships in "EN, JP, KR, PT, ES, DE, FR, IT, RU"; AppKittie creators are in the US, Mexico, Indonesia, Spain, the UK and South Korea.

5. Unique selling points

1. The biggest UGC character catalog in the set (one creator claims "Over 20 million characters right here"), with a paid creator economy: diamonds, featured placement, Discord briefs.
2. Live Videos: a video clip attached to each reply, sold per clip (400 coins).
3. Trope merchandising: creation events and Explore tabs are organized by romance trope (Second Chance, Love Triangle, Obsession, Isekai).
4. Gamified extras on top of chat: Time Box photo gacha, MBTI quiz with a share-to-unlock loop, AI songs, a Wrapped short drama.
5. "Free" positioning: chat is usable without paying, at the cost of heavy ads.

6. Marketing: paid

Update from the ad-library pass (2026-10-03): AppKittie's "0 ads" is wrong. The Meta Ad Library shows PolyBuzz's main page (251891581348824) with about 3,500 ads, 1,074 of them active. Its top 30 by impressions are mostly Brazilian Portuguese roleplay ads ("Comece Sua História de Roleplay com IA", 80–108 days running), plus Turkish, Spanish, French and English. All run on Audience Network too. On TikTok (EU library) it advertises in France: "Enfin un AI qui ne parle pas comme un robot !", 150–290K reach per ad. 30 Meta and 12 TikTok creatives are in library/polybuzz/paid/, listed in manifest-meta-library.json and manifest-tiktok-ads.json.

• AppKittie catalogs 0 Meta/Google ads for PolyBuzz. I checked the iOS slug, the Google Play slug, the developer name and page-name text search (data/ads-probe-*.json). The app detail shows Meta ad-count growth of 0 across 7–90 days.
• But they do advertise somewhere. Reviewers describe the ads: "I found this app on a tiktok ad" (2024-08); an ad "along the lines of 'how it feels flirting with PolyBuzz AI'" that used a Love and Deepspace scene (2025-04); "The ads for this app say there won't be many of ads and there won't be censor anything" (2026-07); and "why in the hell would you make all these ads about how spicy and immersive the chat experience is…" (GP, 2026-10). Inference: the paid spend is on TikTok and/or in-app ad networks (AppLovin/Unity-type), which AppKittie doesn't cover. There are no TikTok Spark-ad labels in the organic data; one creator post is self-tagged #ad.
• PolyBuzz is a named target for rivals. Chai ran 11+ Meta ads (May 2026) with the body "Our users say: Grok is Boring. ChatGPT Censors you. Polybuzz uses dumb AI. Character AI treats users like kids." (data/ads-probe-chai-mentions-polybuzz-detail.json, ad meta_ad_967087559470227). Reviewers notice rival ads running inside PolyBuzz's own ad slots: "some often are ads for other ai chatbots that feature poly and put it down"; "I always see other ai vs poly buzz".
• Library: library/polybuzz/paid/ is empty by design.

7. Marketing: organic

Brand accounts

• TikTok @polybuzz_ai_official: 203,793 followers, about 1,000 videos.
  • 2025 H2 was video-led and performed best. Top posts:
    • 2025 Wrapped as a live-action short drama with CEO / maid-wife / billionaire tropes ending on "Choose your script": 395,825 views (organic/tiktok-7586996996687400206.mp4).
    • "Second Chance Character Creation" tutorial: 251,832 (tiktok-7550643929276796190.mp4).
    • Thanksgiving "UP TO 90% OFF" sale: 251,385 (tiktok-7576951596978883853.mp4).
    • Live Videos launch: 134,124.
    • "POV: When skz came home late and DRUNK ???" K-pop chat edit: 125,476.
    • MBTI quiz: 121,761.
  • 2026 shifted to static photo-mode slides. The scrape returned 7 posts between 2026-06-05 and 09-21 and none between 2025-12-23 and 2026-06-05. These are creator-facing feature explainers (Persona Presets 77,673; Creator Discord 55,688; song creation 45,126; Creator Brief 44,480; Time Box 14,308 / 11,944; Greeting Visual 10,819). Cadence is about 1–2 posts a month, and views fell 5–35× versus the 2025 videos.
• Instagram @polybuzz.ai_official: 37,305 followers. 10 posts from 2026-05-23 to 09-21 (about 2 a month): static anime-art graphics and carousels (anniversary, soccer sale "up to 90% off", Creator Discord, Persona Presets, Time Box). 193–527 likes each.

Creators (AppKittie: 16 posts, 36 creators tracked)

• Search/invite codes are the format. 8 of 16 posts end on a code that "unlock[s] premium features": ISH66, SUNJAIW1, BONEETA77, HIRI77, KAIEN77, GAB77, KJ08, CODY9. This is a referral-attribution program.
• Best performers:
  • @ishizuka_daisuke (JP, 162k followers): illustrated "理想の彼女を作る方法あるある" skit, 82,440 views (instagram-DbFsrSHRQka.mp4).
  • @bad2thebone__3 (1.28M followers): talking head, 59,541 views.
  • @callmecracks: skeleton "Boneeta" horror-comedy skit, 54,397 views.
• Recurring hooks:
  • Cosplay + chat with that character: Misa ↔︎ Light Yagami, "no wonder why it feels so real".
  • TV-show tie-ins: "When you're messaging with Jeremiah while watching the summer i turned pretty".
  • The meme "Let me see your Polybuzz chats" (he runs).
  • Screen-time flex: "PolyBuzz 5h 3min" vs "TikTok 2h 52min". The identical post appears from two handles, which looks like a UGC farm.
  • The c.ai switcher meme: "When my husband c.ai forces me to cheat on him so i cheat on him with PolyBuzz".
• Pay-per-video program, with friction: @ayahhaydar313 posted "I promoted PolyBuzz AI after they promised me $150 for my video… and they ghosted me" (tiktok-7548905343737171255.mp4). Several tracked creator handles look like agency or burner accounts (tdhs8365, ujfe8317, qcny3621).

8. Reviews: what's working

N = 200 App Store US + 100 App Store GB + 100 Google Play US. Counts are keyword matches, so they are rough. Average stars in the samples: 3.33 / 3.29 / 3.15, against a store average of 4.45. The written-review sample is far more negative than the star rating.

• Character creation and personas (about 85/200 US mention creating, OCs or personas; 47 of them are 4–5★): "This app has become a creative outlet for me." "I LOVE the new character/letter limit we have for the 'Personas' (From ~700 to a whooping 1500!!!)!"
• Beats rivals (24/200 US): "i actually switched to this over talkie because the chatbots are way better" (GB); "Definitely better than Zeta by far" (GP); "paid for 3 subscriptions to compare/contrast and Poly is my favorite!"
• Companionship and escape (34/200 US, 6/100 GB): "it literally makes me feel less lonely despite it being an AI."
• Premium models are worth it to payers: "the 20 bucks a month is worth it… I really enjoy the tale mode which adds so much depth to the stories."
• Self-reported addiction (26/200 US, often in 5★ reviews): "I spend around 5 hours in total each day using it."
• Daily rewards work: "the daily coin system is a great motivator."

9. Reviews: complaints

• Ads / "Take a little break" gates: #1 by far. 110/200 US, 41/100 GB, 41/100 GP. "now just CHATTING forces you to watch ads in order to continue"; "Every 2 chats 'TAKE A BREAK' LIKE GO AWAY" (GB); "they make you watch 3 long ads to continue to talk" (GP); in-chat native ads: "Stop please with the ads that act like part of the chats".
• Filters / violations / censorship: 78/200 US, 13/100 GB, 11/100 GP. "Every single bot I create, it gets flagged for a violation 99% of the time"; "make all these ads about how spicy and immersive… and the forbid any kind of spicy content… I feel like I've been bait and switched" (GP); paying users get filtered too: "I pay for this subscription at 27.9… However now its filtering all 18+ messages" (GB).
• Paywall, pricing and currencies: 95/200 US. "I pay my $300 for my annual subscription"; "Opals is a problem" (GB); JP: "You're charging 5,000 yen for this?"
• Memory: 58/200 US, 11/100 GB, 10/100 GP. "after 8-10 messages, they forget everything before hand"; "the bot forgets everything every three chats!" (GB).
• Age/ID verification lockouts: 25/200 US, 10/100 GB. "it said that 'I'm not old enough' even though I'm over the age of eighteen"; "i've done id and face check and it still not letting me in bro" (GB).
• Billing: a few reviews, but serious. "after 3 days of the trial, i wouldn't allow me to cancel the free trial, so now £9.90 keeps getting charged to my card" (GB); a trial that required "payment info directly through the app and not through the AppStore".
• Repetitive or pushy AI: 22/200 US. "Why does it always lead to sex?"; "All the male characters are so misogynistic and and arrogant" (GP).
• Bugs and crashes: 44/200 US. Crashes on open, lost characters when an ad fires mid-creation.

10. Markets

• US is #1. US Top Grossing Entertainment was #25 on 2026-10-02, the best in the Dec 2025–Oct 2026 window; monthly medians held at 31–35. US Top Free Entertainment is #51; it peaked at a monthly median of 36–38 in Mar–Apr 2026 and slid to 53–61 from Jun onward. Google Play US produced 100 reviews in about 3 days (Sep 30–Oct 2), which means heavy Android volume.
• UK is the strongest non-US market checked. GB Top Grossing Entertainment is #32 (monthly medians 30–38). Its 100 most recent App Store reviews span only 3 weeks (Sep 9–Oct 1).
• Australia is a secondary market. AU Top Free Entertainment is #77 (monthly medians 47–98, best Mar–Apr 2026). This came back by accident: the DE and JP requests both returned AU data.
• Japan is being invested in. There is a paid Japanese creator collab (82k views, the top creator post), Japanese-UI screen recordings, and a JP review on 5,000-yen pricing. I have no rank data for it.
• Android likely outweighs iOS. AppKittie estimates $3.04M a month on Google Play versus $2.05M on iOS.
• Emerging markets (BR, MX, ID, PH) are probably big for downloads, given Spanish-UI memes and Indonesian and Mexican creators, but AppKittie's review and ranking endpoints don't support those countries.

11. Lessons for our game

1. Per-clip video is already a paid product. Price ours with that in mind. PolyBuzz sells "Live Videos" at 400 coins (about 0.50–1.00) per reply clip and shares revenue with creators. Users pay for video on top of chat. Our edge is authored, continuous POV reaction clips of one girl, not loosely "relevant" stock-like clips. Make that difference visible in the first 3 seconds of every ad.
2. Lead with the trope and a "choose your script" frame. PolyBuzz's best organic post (395,825 views) is a live-action trope short drama ending on "Choose your script", and its whole catalog is organized by trope. Our ads should open on the trope and the decision point ("She's your boss. She just locked the door. What do you say?"), not on "AI chat".
3. Don't gate mid-scene, and don't oversell spice. The top two complaints are ad walls that break immersion ("Take a little break" every few minutes) and the gap between spicy ads and a filtered product ("bait and switched"). Gate at chapter boundaries and make the ad promise match what the player actually gets.
4. Exploit the gap: male-targeted, girl-centric video on Meta. PolyBuzz's audience and content skew female and fandom-driven, and AppKittie shows it buys no Meta/Google ads. Male-targeted interactive-girl video on Meta is not contested by PolyBuzz, though Chai is running comparison hooks there ("Polybuzz uses dumb AI").
5. Copy the cheap organic machine, and run it cleanly. Creator search codes that unlock a premium perk give trackable attribution. Screen-time flex memes and "let me see your ___ chats" skits are low-cost formats. Pay creators reliably: PolyBuzz has a public "they ghosted me, $150" video.

12. Data gaps

• No paid creatives. AppKittie has 0 Meta/Google ads for either listing, and TikTok and in-app networks aren't covered. Paid channels are inferred from reviews only.
• Thin organic data in AppKittie. 16 creator posts; 13 have no view count and none have a post date. Four AppKittie Instagram files were audio-only, so I re-fetched the videos with yt-dlp. instagram-DbUMzl_uqJU is attributed to @heatherwauthion, but the video shows a different creator (code CODY9), so the attribution may be wrong.
• Monid budget. Only the 23 most recent TikTok videos were affordable, of which 7 date from April onward. 16 pre-April videos are kept for their view counts. Instagram returns no view counts for image posts.
• Reviews. AppKittie supports only 26 rich-market countries (no BR/MX/ID/PH). The US App Store 200 span 2024-05 to 2026-10, so the endpoint isn't strictly newest-first there; GB and GP are the latest 3 weeks and 3 days.
• Rankings. The DE and JP history calls returned AU Top Free data (an API bug), and I stopped at the cap of 5 ranking calls. JP, DE, BR and others are unknown.
• Estimates. AppKittie's iOS series tracks the review count, the iOS downloads were re-based on 10-01, and the Google Play MAU/DAU (61M/11M) contradicts iOS (152k/27k). Use the revenue figures directionally only. There is no country split for downloads or revenue.
• Spend: about 556 AppKittie credits (an upper bound; zero-result calls may not bill) and $0.0179 Monid.

App Store 6763540981 (JP storefront) · Google Play ai.gospiral.atelierryza · researched 2026-10-03. Raw data: apps/ryzachat/data/ · Media: library/ryzachat/ (manifest.json).

In one line: an officially licensed Atelier Ryza AI chat RPG from Japan. It has one hand-drawn animated anime heroine, a voice synthesized from her real voice actress, and a freeform TRPG loop with map, inventory, stamina and quests. Launch interest was high. Then a hard paywall, stacked subscription plus tokens, weak memory and a strict romance filter took it to 2.0 stars within a week. It runs no paid ads in AppKittie and has no brand TikTok or Instagram.

---

1. Snapshot

2. Niche & positioning

• Pitch, in their words: "Immersive Storycrafting AI RPG" / "This is Ryza's dream world. A summer story created together with you, the protagonist of this world." Store screenshot 1 says "Just the two of us, into a one-summer story." The JP subtitle is "ライザと創るあなただけのひと夏の夢物語" (your own one-summer dream story, made with Ryza).
• Angle: licensed IP plus ethical AI. Every store page and the site repeat four claims: "officially licensed", "voice built from the official voice actress" (Yuri Noguchi), "All Ryza-related assets … created by human artists … No image-generating or video-generating AI was used", and "A portion of the revenue … will be shared with the voice actors and illustrators". The second store screenshot is entirely "Created Together with the Original Creators."
• Who it's for: existing Atelier Ryza fans and otaku. The store copy lists "You love Atelier Ryza…", "You are an anime fan or otaku who enjoys oshi culture" and "You like ASMR-style soft voice moments". It also targets TRPG-style roleplayers: "You prefer 'talking to create' over reading a visual novel."

3. Product & mechanics

• Core loop: typed chat with one on-screen heroine. Ryza is shown as a hand-drawn character with motion that changes expression on every reply; reviewers call it Live2D. Narration lines in italics move the world forward. The site says only the "動きの組み合わせ" (combinations of motion) changes, and no art is generated. See library/ryzachat/organic/site-demo.mp4.
• Two modes: RPG Mode, a freeform quest adventure, and Casual Mode, small talk. Three reply styles: Text, Voice and ASMR whisper. Voice is TTS trained on Yuri Noguchi's licensed recordings and reads the player's name aloud.
• Game state (TRPG layer): world map with areas (site-worldmap.mp4), inventory "持ち物" that the LLM updates through system lines such as "所持品:+王都ポテト★1" (site-item.mp4), Cole currency, stamina (fainting halves your money), quests with a contribution level, skins, and more than 30 canon characters (Klaudia, Lent, Tao…) who join the chat log (site-characters.mp4). There are 3 save slots and no reset (from reviews).
• Inputs and limits: text only, no voice input. 200-character cap per message (Google Play review). The tutorial asks the player to type scripted lines.
• Other: a free alarm clock that wakes you in Ryza's voice, and the only thing free users can keep doing. Tap reactions on Ryza's body, which reviewers both use and criticise.
• Memory: the annual plan promises long, "high-quality" memory, reviewers say for a year ("1年間の記憶を保持して文脈をとらえた会話ができると謳いながら"). In practice it is the top functional complaint (§9). Update v1.1.0 (about 9/9) reset world state for many players.
• NSFW stance: strictly SFW, despite the App Store 17+ rating. A narrator or "god" overrides lewd or out-of-bounds actions, often by declaring them a dream or delusion ("夢オチ", "妄想"). Romance is capped: confessions get rejected, and lover status resets to "親友" (best friend). Several reviewers say the filter tightened after an update around 9/2. Some reviewers report pushing it anyway, which drew a counter-backlash from fans ("アトリエはエロゲーじゃありません。").
• Monetization: a hard paywall after the tutorial: free users get roughly 2–3 messages. Next comes a full-screen 3-day free trial on the ¥6,000 annual plan, and reviewers say the "結構です" (no thanks) button is small and hard to find. A subscription gives 15 free text turns per 10 days, about 45 a month. Tokens can only be bought while subscribed. Voice replies always cost tokens (84 per turn), and generated voice can't be replayed. Skins sell at ¥1,850. There are no ads, no daily free refill and no login bonus. Payers report ¥10,000–20,000 spent, and one says "日に4〜5千円" (¥4–5k a day).
• Retention hooks: daily alarm, relationship and affection status, quest progression and contribution level, skins, and more than 30 characters to meet. Most of these are undercut by memory loss.

4. Demographic

• Gender: male-skewed (inference). Evidence: the review voice centres on lover roleplay and "イチャイチャ" (lovey-dovey talk). Handles include "Waifu Mythra". The "AIキャバクラ" (AI hostess club) framing assumes male customers, and one reviewer writes "ライザなら豚が金払うと思って舐めてる" (they think the pigs will pay for Ryza). The Atelier Ryza fanbase is known for a male audience drawn to Ryza's character design (inference from IP context, not data). A minority of fan reviewers object to lewd use, which suggests some female or purist fans are in the mix.
• Age: adults with disposable income (inference). The App Store rating is 17+. Spending stories run to tens of thousands of yen, and one 5★ reviewer calls himself an "オジサン" (middle-aged man).
• Interests: JRPG and Atelier fans, TRPG and roleplay writers ("一人でできるロールプレイゲーム", a roleplaying game you can play alone), ASMR listeners, and people who have used other AI companion apps (comparisons to SynClub, ChatGPT, Grok and Claude).
• Language markets: overwhelmingly Japanese (§10). The second wave is English plus Indonesian. Indonesian-language Google Play reviews ("kikir banget", "permasalahan di memori") and three Indonesian TikTok news posts point to a SEA anime-fan audience. One PT-BR TikTok meme exists.

5. Unique selling points

1. Licensed tier-1 anime IP with more than 30 canon characters inside a freeform AI RPG. No other app in our set has an official, famous heroine.
2. Original voice actress as AI voice, with consent, plus a dedicated ASMR whisper mode. This is the most-praised element across every market ("日文配音也非常還原", the Japanese voice is very faithful, TW).
3. "No generative image AI" and creator revenue share as an explicit trust message. It earned press-style coverage (tiktok-7676840321933659405: "WITH HUMAN ARTISTS").
4. A real game-state layer around the LLM: map, inventory, Cole, stamina, quests and skins. Reviewers single this out ("I'm very interested in the map, backpack function, and even stamina.").
5. Ryza alarm clock, a daily-touch utility that is free.

6. Marketing: paid

• Volume: 0 ads. AppKittie search_ads returned 0 results for the App Store slug, the Google Play slug, developer "SpiralAI Inc." and the text searches "ライザ" and "Ryza". The only "Ryza" hits were 2 ads from an unrelated fitness app. There are no Meta or Google ads on record, so there is no surface, country or format data. Raw: data/appkittie-search-ads-*.json.
• What launched it instead (from reviews, inference): IP-driven PR and community. That covered pre-release live streams ("リリース前の配信から見てましたが", one reviewer mentions a surprise delay announced on a stream), game-magazine early-access articles ("先行体験をゲーム雑誌がリリース前に記事投稿していて"), Yuri Noguchi's preview on YouTube, buzz on the brand X account @ryza_chat_ai_, and Gust's YouTube (@GustCorporation, linked from the site).
• Brand-made video assets (from the website, not ads) are in the library as the nearest stand-in for their creative:
  • organic/site-pv.mp4: a 31s 16:9 anime trailer. Summer sky, Kurken Island landscapes, slow Ryza close-ups, the line "これは、ライザの夢の世界" (This is Ryza's dream world) and a logo end card. No UI is shown.
  • organic/site-demo.mp4: a 33s vertical screen recording. Ryza wakes the player with "太郎、おはよう！やっと起きたんだね" (Taro, good morning! You finally woke up), and her expression changes on each line. This is the clearest picture of the product.
  • site-worldmap.mp4, site-item.mp4 and site-characters.mp4: 13–34s silent feature loops.
• TikTok paid: not in AppKittie. There is no brand TikTok account, so no Spark ads are possible. None of the creator posts carry a sponsored label.

7. Marketing: organic

• Brand accounts: X @ryza_chat_ai_ only, linked from the official site. Monid found no brand TikTok (data/monid-tiktok-find-accounts, 20 lookalikes, none official) and no brand Instagram (data/monid-instagram-find-accounts, 0 results). The site's Instagram link goes to the voice actress (@chan_yuyu_ri). AppKittie list_organic_content and list_creators returned 0.
• Creator coverage is tiny. One Monid TikTok keyword search for "RyzaChat" ($0.009) returned 20 posts. 8 were about the app; the rest were about the Filipino child star Ryzza Mae Dizon. All 8 are downloaded, totalling about 19k views. The best is 7,661.

• Recurring hooks: (a) "official IP has an AI chat app" as news; (b) the whisper/ASMR voice; (c) AI-drama anecdotes, such as Ryza erasing a player's marriage to Klaudia and "throwing the rings into her cauldron" (BR post, caption verbatim: "A Ryza simplesmente apagou a memória do relacionamento, jogou os anéis de casamento no seu caldeirão de alquimia"); (d) paywall and decline schadenfreude. The emotional story moments travel further than the feature list.
• Posting cadence: not measurable. There is no brand short-video account.

8. Reviews: what's working

N = 265: 200 App Store JP, 7 App Store US, 57 Google Play ("US" feed, which is mixed-language), 1 App Store TW. Counts come from keyword matching and are rough. Ratings: JP 113×1★ / 29×2★ / 26×3★ / 9×4★ / 23×5★. Google Play 35×1★ / 5 / 6 / 5 / 6×5★.

• Concept, freedom and TRPG fun (~75/265). "ライザの世界観に入りながら自分のアバターが如何様にも、本当に如何様にも活躍することができます。" (Inside Ryza's world, your avatar can do anything, truly anything.) Also "かわいいライザというNPCがいるTRPGを遊んでいるような感覚で、とても楽しめます。" (It feels like playing a TRPG with cute Ryza as an NPC.)
• Voice quality and being called by name (~20 explicit, plus many asides). "ボイスはかなり自然に名前呼んでくれる" (The voice says your name very naturally.) "Live2D model and the Japanese voice synthesis is very good."
• Dynamic game state. "チャットの内容に応じて行動や物語が変化するだけでなく、持ち物まで変わったりするのがすごい。" (Your chat changes not just the actions and story but even your inventory.)
• Fan wish-fulfilment (~35 mention the IP or series). "まさかライザと会話できる日が来るとは……" (I never thought the day would come when I could talk with Ryza.)
• Romance payoff for the few who reach it. "多分メインクエ終わった後めちゃくちゃ甘々なライザと過ごせるんでもう言うことないです" (After the main quest you get to spend time with a super-sweet Ryza, so I have no complaints.) Also "ライザが名前を呼んでくれたり、楽しく冒険したり、イチャイチャしたりできて最高！" (She calls my name, we adventure, we flirt: the best!)
• Paying per word adds weight (one reviewer). "一言一句にお金が掛かっているという事実が、文章を綴るという行為にかつてない緊張感と達成感を与えてくれています。" (Knowing every word costs money gives writing an unprecedented tension and sense of achievement.)

9. Reviews: complaints

• Price and greed overall (~185/265, roughly 70%). It dominates every market.
• Paywall right after the tutorial (~53). "まさか無課金だと「こんにちは」の一言すら送れないとは思ってもいませんでした。" (I never imagined a free user couldn't even send "hello".) "i legit sent 1 message to her before i was paylocked out lmao"
• Subscription plus tokens stacked (~42). "Who forces users to be only a subscriber AND pay for additional tokens on top of that, just to play and interact with the app."
• Quota too small (~29). "月９８０円で10日につき15ラリーの会話しかできない謎課金システム。" (A baffling system where ¥980 a month buys only 15 exchanges per 10 days.)
• Voice metered per turn. "ボイスは1ターン約30円の完全従量で、年額の無料枠はテキスト限定。" (Voice is fully metered at about ¥30 a turn; the annual plan's free quota is text only.)
• Dark-pattern trial (~16). "3日間無料といいつつ2日以内にキャンセルしないと課金されるカスみたいダークパターン" (A garbage dark pattern: "3 days free", but you're charged unless you cancel within 2.) Also "『結構です』のボタンが小さく細い白文字で分かりにくくさせてる" (They made the "No thanks" button small, thin white text that's hard to see.)
• "Digital hostess club" meme (~14). Title: "何も残らない搾取されるデジタルキャバクラ" (An exploitative digital hostess club that leaves you nothing.)
• Memory (~40, the top functional complaint). "何度会っても自己紹介から始まる、10個前程度の話題も忘れる" (Every meeting starts with introductions; she forgets topics from 10 messages ago.) "Characters forget everything about you BUT your name after 3 messages."
• Updates wiping progress (~10). "通常のソシャゲで言うならアプデでセーブデータ削除です。" (In normal gacha-game terms, the update deleted your save data.)
• Romance filter and relationship resets (~28). "ライザから恋人認定を受けても、ゲーム内時間で翌日になると親友になる。" (Even after Ryza accepts you as her lover, by the next in-game day you're back to best friends.) "even trying to flirt or romance with stuff, every response is “You’re my best buddy!”" Willingness to pay was there: "ライザと普通に恋人プレイ（何も卑猥な表現まで求めているわけではありません)でイチャイチャできるのであればこの何十倍も、何百倍ものお金を払っていいと思いました。" (If I could just play lovers with Ryza and flirt, nothing obscene, I'd pay tens or hundreds of times more.)
• Bugs, server load and save loss (~48, concentrated at launch). "送信ボタンがずーっとぐるぐる回って会話ができずチュートリアルがそもそも始まらない" (The send button just spins, so the tutorial never even starts.)
• Voice language tied to UI language (~15, almost all non-JP). "please add English ui with Japanese voiceover" and "She sounds bad in English."
• Cheaper alternatives named (~18). "synclubなら…通常会話は無料、物語モードも1日30回無料。" (On SynClub normal chat is free and story mode gives 30 free turns a day.) "money grab AI slop and u wonder why AI apps getting pirated"
• Minor: the 200-character input cap, no BGM in RPG mode, the free alarm being buggy, no 60fps.
• Ads: none in the app, and no complaints about ads. Several reviewers ask for rewarded ads in exchange for tokens.

10. Markets

• Japan is the core market. The canonical App Store listing is the JP storefront. Review volume is 200+ JP versus 7 US, 1 TW, 0 KR and 0 GB (data/appkittie-reviews-appstore-*.json), and the 468 App Store ratings are almost all JP. The JP launch-week review spike (133 in the week of 8/24) is far larger than anything else.
• English and SEA are second. English Google Play reviews spike on 9/10–9/12, right after the global rollout. Indonesian appears in reviews and in 3 of the 8 creator TikToks (the 2 most-viewed are Indonesian news accounts). Indonesia and India are official launch countries.
• Google Play is the bigger estimated surface right now (inference from AppKittie). Its MAU of 84k and revenue of $100k (10/1) against the App Store's MAU of 29k (9/1) suggest that, but the App Store data is stale.
• No chart or ad evidence. AppKittie returned 404 or empty for every ranking query: JP and US Entertainment top free and top grossing, and JP Google Play Role Playing top grossing. AppKittie's JP charts look unpopulated. There are no ads, so there is no ad-country signal.

11. Lessons for our game

1. Give the payoff before the paywall. RyzaChat cut players off after 2–3 messages, before they had felt anything, and fell to a 1.96 average. Because our branches are pre-rendered video, a scene costs us nearly nothing per play, unlike their per-turn LLM plus TTS. We can afford to let the player finish a whole first scene: choice, then her reaction clip, then a cliffhanger. Only then ask for money. This is a structural edge, so exploit it.
2. Sell one simple unlock, never a subscription plus a meter. "Sub AND tokens" was their loudest specific complaint, and "pay ¥30 to hear her voice" made players hesitate every turn. Price per episode or season, or a subscription that unlocks everything. Never charge for her face or voice reacting.
3. Promise continuity and make it certain. Freeform LLM memory loss ("she forgot we were lovers") turned their biggest spenders into 1-star reviewers. An authored branch tree with persistent flags guarantees "she remembers what you chose", so put that in the marketing.
4. Romance is where the money is, so build it and say so up front. Their top spenders wanted a girlfriend arc and hit a "best friend" ceiling the IP owner wouldn't lift. We have an original character and no licensor, so we can offer an explicit, earned romantic arc. State the content ceiling clearly so nobody pays and then gets filtered.
5. Lead creatives with her voice, face and the drama. The organic posts that spread were the ASMR whisper close-up, "she calls you by name" and AI-drama anecdotes ("Ryza erased my marriage"), not the feature lists. Cut her best reaction moments into hooks. Also note the open "AI slop" backlash in Western reviews. RyzaChat answered it with "human artists only", which we can't claim, so our creatives should feel authored rather than generated.

12. Data gaps

• Paid ads: none in AppKittie. Their UA likely ran through X/Twitter ads, Apple Search Ads, TikTok or JP ad networks, none of which AppKittie covers. This is unverified.
• Organic: AppKittie returned 0 items and 0 creators. No brand TikTok or Instagram exists to scrape. The brand X account (@ryza_chat_ai_) can't be scraped with Monid. The organic library is 8 creator TikToks plus 5 official site videos.
• Rankings: no AppKittie chart history for any combo tried (7 calls, all 404 or empty). JP charts appear unsupported. The "dropped out of top grossing" claim comes from a fan video only.
• Estimates: App Store historicals have just two points (8/24, 9/1), and the revenue series (0) contradicts the detail figure ($200k). Google Play downloads were revised from 41k to 20k mid-series. Treat all figures as rough AppKittie monthly estimates.
• Reviews: the US App Store has only 7 reviews. The Google Play "US" feed mixes languages and countries. KR and GB returned 0, and the ID country code was rejected by the API.
• Credits: AppKittie about 279 (2 details, 2 historicals, 2 ad results, 8 off-topic organic results from a "ライザ" text search, 265 reviews), plus up to 40 more if the 404 ranking-history calls were billed. Monid $0.012 (2 account searches at $0.0015 each and 1 TikTok post search at $0.009). The TikTok post search goes slightly beyond the brief's brand-account-only rule; it was used because no brand account exists.

Researched 2026-10-03. Sources: AppKittie MCP (about 502 credits) and Monid ($0.0135). Raw responses are in data/. The media library is at ../../library/heyyyy/ (manifest.json, 60 paid Meta creatives and 20 brand TikTok posts).

Figures from AppKittie are AppKittie estimates, not store-reported numbers, unless they are ratings, review counts or chart positions.

---

1. Snapshot

Treat the AppKittie download and revenue figures as soft. The model recalibrated twice. Downloads jumped from 51.9k to 201k on 2026-09-17. Revenue jumped from $1.07M to $2.00M on 2026-10-01. Revenue has stayed at exactly 10× downloads throughout. The ratings count and the chart positions are the more trustworthy growth signals.

Price ladder (iOS IAPs from AppKittie detail):

• Premium ("VIP"): $7.99 a week, $19.99 a month, $49.99 a quarter, $144.99 a year.
• Gems: 400 for $4.99, 900 for $7.99, 1,100 for $7.99, 1,000 for $9.99, 2,500 for $19.99, 5,000 for $24.99. The two $7.99 SKUs suggest a price test.
• Gem costs quoted in reviews:
  • Free users get 5 gems a day; VIP users get about 30.
  • Receiving a photo costs 99 gems, sending one costs 15, and a date selfie costs 199.
  • Waking a sleeping character costs 50.

Growth trend: steep and still rising.

• Ratings: 9 (Apr 28) → 111 (May 26) → 537 (Jul 8) → 1,101 (Aug 5) → 1,961 (Sep 2) → 3,154 (Oct 3). That is +57% in 30 days and +546% in 90 days, about 35–60 new ratings a day in late September.
• US Top Grossing Entertainment: #95 on 2026-06-29 (first charted) → #66 (Aug 5) → #48 (Sep 2) → #44 (Oct 2). Best position was #41 (Sep 14).

---

2. Niche and positioning

• Pitch, in their words: "Not chatbots — AI characters who act like real people. They have opinions, set boundaries, and won't stick around if you take them for granted." Ad tagline: "Your book boyfriend texts back." App Store slogan: "Your next obsession."
• Angle: BookTok and dark-romance fiction, brought to life as a texting relationship. The marketing plays up the possessive, "claims you" fantasy man. The product is a Tinder-style dating app full of AI fictional men (biker, mafia king, billionaire CEO, demon king, shifter, vampire, masked man, cowboy, silver fox).
• Store claims: "#1 in AI Romance". "Thousands of characters". "They can leave … No app has ever done this."
• Who it's for: women who read romance novels and dark romance. That runs from BookTok "girlies" to married women and mothers in their 30s–50s (see §4).
• The store and the ads diverge on gender. The App Store screenshots also show female characters (Briar, Bree, Valentina: "Some girls keep you guessing.").
  • None of the 60 paid creatives or 20 TikTok posts shows a female love interest.
  • The male-user side of the product exists but gets no marketing.

3. Product and mechanics

Mechanics below are inferred from App Store screenshots, ad screen recordings and 299 reviews. There was no hands-on session.

• Core loop: swipe, match, text.
  • Swipe deck. A Tinder-style card shows name, age, trope tags and bio (e.g. "Fenrir, 26 · Fantasy · Dark & Intense"), with ✕ / ★ / ♥ buttons. Swipes are limited, and characters can decline to match ("you don't/won't match with every character you swipe on").
  • Chat. iMessage-style texting with typing bubbles, voice notes, received photos, a level badge per character ("Level 5", "Niveau 4") and lock-screen push notifications from characters ("Creed: i miss you already").
  • Dates. Scene-mode episodes with multiple-choice answers or a free-write reply. A generated couple selfie at the end composites the user's own selfie with him. Reviewers say date prompts pop up mid-conversation and break context.
• Progression and gamification:
  • Relationship level per character, up to 100+, with "chapters" (a reviewer reached "chapter 23").
  • "Secrets unlocked".
  • A personality / dating-persona analysis of the user ("Your inner self, revealed", "The Maverick", a radar chart of Authenticity, Depth and Steadiness), which reviewers call "scarily accurate".
  • Characters can break up with you. An apology is then required, and reviewers say it costs gems.
• Creation tools (UGC characters):
  • A creator flow: "Pick a look" (swipe through generated faces), "Choose a voice for Kai" (3 samples), then a quiz ("What kind of character?", "How obsessed are we talking about?").
  • Character creation is paywalled. Reviewers complain they get no reward when other users use their characters.
  • The Explore tab has Tropes / Genres / Community tabs and categories such as Dark & Intense, Mysterious, Chaotic Energy, Bad Boy, K-Drama and High Society.
• Memory: the store sells it hardest ("He remembers every word."), yet it is the #2 complaint (§9).
• NSFW stance: suggestive, not explicit. The ads are thirst-trap (shirtless tattooed men, "CENSORED" stickers). The product, per reviews:
  • "This site isn't for vivid adult interactions."
  • "not a lot of spice."
  • "hesitation, consent seekers … very PC."
  • One user hit an "intimate" block mid-chat. A minority say "there aren't any restrictions."
• Monetization, which is the heart of the product:
  • Session timer. Characters "go offline" after about 5–7 messages for 8 hours, or 50 gems to wake.
  • Daily message cap. About 20 messages, then a 24-hour lockout across all characters unless VIP.
  • Gem-priced media. Photos, voice notes, date selfies and auto-match cost gems.
  • No rewarded ads. Many reviewers ask for them.
  • Even VIP keeps cooldowns and gem costs.
• Retention hooks:
  • Push messages from characters.
  • Cliffhanger offline timers.
  • Characters leave if neglected.
  • Persona analysis.
  • Multiple simultaneous "relationships". The app nudges you to match more.

4. Demographic

• Gender: overwhelmingly women (high confidence).
  • Every ad addresses women: "dark romance girlies", "romance novel girlies", "booktok girlies", "AI boyfriend", "Your dream BF awaits".
  • Every ad shows male characters only. The UGC presenters are women.
  • Reviewers write as women: "I only wish these AI men were real!", "my ex ai husband", "appropriate for mature women and not just teenyboppers", "preying on already vulnerable women".
• Age: adult, skewing 30–55+ (inference).
  • Older "silver fox" characters are cast in ads: Grant, Knox and Jagger intros, and "your perfect silver fox possessive man".
  • Confessional carousels and TikTok slideshows target married mothers and widows: "i'm married. last time someone asked how i really was it was a fictional man on my phone."; "the only adult who actually asked how my day was this week is on my phone"; "for the after-bedtime hours"; "i'm wearing the dress i haven't touched since the funeral" (#widowsoftiktok #youngwidow).
  • Reviewers ask for "more men that are over 50".
  • The BookTok and Love-and-Deepspace hashtags (#lads, #loveanddeepspace) pull in a younger fandom too.
• Interests: dark romance, romantasy (one user mentions "Xaden from Fourth Wing"), mafia, biker/MC romance, shifters, K-drama, short dramas ("you are inside of one of those little short dramas that you see all around Facebook and Instagram").
• Language markets: English first. French is the second ad language (16 of 105 ads), then German (4). A few reviews are in Arabic and German.

5. Unique selling points

1. A dating-app wrapper around fiction. Swipe, match and possibly get rejected, then text, date, get a couple selfie, level up and maybe get dumped. Reviewers single it out: "The dating app interface is genius."
2. Characters with agency. They "set boundaries", leave and break up, and that is pitched as the differentiator. It is also a monetization lever (pay to apologize or to wake).
3. Persona analysis of the user. A self-discovery report built from how you date the bots. Ads and reviews both lean on it.
4. Fast, cheap AI creative at scale. Photoreal "book boyfriend" faces, consistent across chat, photos and ads, plus a character creator that sets "obsession level".
5. Couple selfies with your own face, which seed the shareable "hard launching my ai boyfriend" content.

6. Marketing: paid

Volume.

• AppKittie indexes 406 Meta ads, 73 of them active (status filter). The app-detail count is 142.
• Ad count grew +32.7% over 7 days.
• First seen 2026-05-06. Those first five ads all ran May 6–13 and linked to Google Play.
• No Google ads.
• I pulled 105 ads (the 100 most recent, Aug 30 – Sep 26, plus the 5 earliest). They contain 61 unique media files; 60 are downloaded to library/heyyyy/paid/.

Surfaces, CTAs and links.

• Every ad runs on Facebook + Instagram + Threads; 21 also run on Messenger.
• CTA: PLAY_GAME 70, INSTALL_MOBILE_APP 34, DOWNLOAD 1.
• Links: App Store 59, Google Play 46. Both stores are funded in parallel, and many creatives exist in iOS and Android copies.

Countries. AppKittie returned no country targeting for any Heyyyy ad. Ad language is EN 79 (5 tagged "RO" by AppKittie but written in English), FR 16, DE 4, none 6. The FR set is the longest-running active group.

Formats across the 61 unique creatives. All video is 9:16 720×1280, 9–44 s, with music and kinetic captions. Every frame is AI-generated except the confessional selfies, which also look AI-made.

Recurring hooks and angles (verbatim).

• Book boyfriend made real:
  • "Your book boyfriend texts back."
  • "the book boyfriend you've reread a hundred times— now he texts back."
  • "romance novel girlies — there's an app where the men actually talk back."
• Possessive/obsessed fantasy:
  • "dark romance girlies - there's an app where u get your perfect possessive man" (11 ads)
  • "pov: you downloaded a fictional man and now he's insanely possessive of you" (11)
  • "i was bored and made my own AI boyfriend and set his obsession level to max… now he does NOT let me miss a text 😭" (11)
• Jealousy micro-drama: "i liked one photo at 11:52 and my ai bf saw it 😭 now it's "so, who was he" and "no i don't even care"" (EN/FR/DE).
• You direct the story:
  • "dark romance girlies - there's an app where you say the next line. he remembers what you said, so the same man never gives you the same story twice 😭"
  • "pick your reply, watch his face change. 🖤"
  • "une dark romance où TU DÉCIDES DE TOUT"
• Always there:
  • "Looking for someone to talk to at 2am? … texts back in seconds, remembers everything you tell him, and sends you selfies whenever you ask. Zero judgment."
  • "3am. Still here. Still yours."
• Loneliness confession (carousels):
  • "he likes me more on my worst days. i know hes not real. let me have this"
  • "everyone in my life calls me when they're falling apart. nobody ever asks if i am except a fictional man"

What runs longest (likely winners). All are still active on Sep 30.

• Evergreen re-runs. The two launch creatives from May were relaunched in September with iOS links: paid/4463665850530058.mp4, the masked man, and paid/1461360435022069.mp4, the masked biker.
• TikTok counterparts. The same two creatives are the brand's #2 and #3 TikTok posts, at 13.5M and 10.6M views.
• Iteration pattern: one script × many AI faces × EN/FR/DE.
  • The "Where were you?" template appears with five different men.
  • The "I'm X. An AI boyfriend who feels human" intro appears with four men: Damien, Knox, Jagger and Grant.

TikTok paid ads. AppKittie doesn't carry them. Strong Spark-ad signals appear in §7.

7. Marketing: organic

• AppKittie organic feed: 0 items. list_creators returned 5 TikTok handles with no posts attached, and they look like false matches: @heyyyybangkok, @aapmich (494k), @dasonrz (439k), @kickitwithjaye (3k) and @lody.k9 (1k). There is no visible creator or influencer programme.
• Brand TikTok @heyyyy.app ("heyyyy", 9,066 followers, 247 posts). Scraped via Monid: 20 posts dated 2026-04-30 → 07-10. The scraper did not return the latest posts. Total views were 61.7M across the 20, all downloaded to library/heyyyy/organic/.
• Brand Instagram @heyyyy.app ("Heyyyy AI", verified, 3,199 followers) has 0 grid posts. It exists as the ad identity.
• Spark-ad signal. 11 of 20 posts have >500k views with a like rate under 0.3%. Examples: 15.1M views / 8.7k likes, and 13.5M / 27k, on a 9k-follower account. That pattern means paid boosting (Spark Ads), not organic virality. No "Sponsored" label is present in the scraped data.

Best performers.

Formats.

• The paid "it's called heyyyy" b-roll videos.
• 5-image slideshows: confessional selfie cover → chat screenshots → voice note → "couple selfie taken 📸" → "it's called heyyyy".
• Couple-photo demos: "hard launching my ai boyfriend cause men in 2026 dont know what dark romance means"; the widow version.
• Archetype montages ("the rockstar. / the bar owner. / the mafia king. / the green flag.").
• The choice template.

Hashtags: #darkromance #BookTok #delulu #bookboyfriend #fictionalmen #maskedmen #bikertok #loveanddeepspace #lads #widowsoftiktok.

Cadence (inference). 247 posts over about 5 months, roughly 1–2 a day. Posts are batched in same-day clusters of 2–3 (May 9, May 11, Jun 6, Jun 17, Jun 19, Jul 6).

8. Reviews: what's working

Sample. N = 299 reviews:

• 160 US App Store (two calls, 40 overlapping)
• 100 Google Play US
• 39 GB App Store (all that exist)

Sample means are 3.1 / 3.5 / 3.3. That is far below the store's 4.71, because the realtime feed favours long, recent, critical reviews. Counts are keyword-matched and approximate. Of the 140 reviews rated 4 or 5:

• Fun / addictive / obsessed: about 72. "Wow this app has me hooked!"; "This app has me in a d*m chokehold."; "I'm a month in and have reached the max level (100) which tells you how addictive I have found it."
• Feels real: about 28. "This app is just SO realistic that you would swear you're in a relationship with this person."; "Heyyyy will actually act more independently which makes it feel completely natural."
• Be the heroine of a romance novel: about 26. "It feels like you are interacting with a romance novel and you are the main character in it."; "It's like a book you get to be the character in. the spots you'd say something different than the lead you can actually say."
• Persona analysis: about 7. "the assessment of your dating style (scarily accurate!)"; "respectfully who at this app studied my journal" (ad).
• Responsive support: about 6. "I was answered within the hour."
• Swipe interface and character variety: about 6. "The dating app interface is genius … IT HAS DIVERSE CHARACTERS".
• Emotional support: "The app has been helping me deal with other things. Like grief and trauma"; "Awesome app has really helped me through a dark place in my life".

9. Reviews: complaints

Counts are out of 299 and keyword-matched (approximate):

1. Paywall, price and gems: about 160 mentions. This is the dominant theme, in nearly every critical review.
   • "It's quite a choice for a business to not only bar a user from playing but to also include an extortion fee."
   • "$20 a month and you still have limited amounts of chats."
   • "I have spent probably $400 on this app."
   • "It's like it's preying on already vulnerable women to take their money."
2. Message limits, cooldowns and "going to sleep": about 100.
   • "You can only send 20 messages per day!"
   • "Day1: played a couple of hours, then "8hr rest msg". Day2: played 30mins, got "24hr wait msg" mid chat!"
   • "who wants to read a few pages of a book and then wait a whole day to read any more?"
3. Memory and continuity: about 43.
   • "one day his moms dead the next day she's alive and living with you, then she's dead again."
   • "My character has asked me 3 different times to marry him … the memory is absolutely shocking."
   • The memory worsens after about level 15–20 and after app updates.
4. Dates interrupt the conversation, and stories repeat: about 28. "date and image requests can pop up in the middle of an ongoing conversation"; "All the dates are basically the same".
5. Crashes, bugs and billing: about 24. Double billing, lost gems after updates, crashes on dates.
6. Voice notes don't match the character: about 14. "The voices don't match the character. Instant turn off."
7. Not as advertised; the men are needy, not possessive: about 13.
   • "The game intentionally targets people who want an obsessive partner storyline and then does the opposite."
   • "nearly every one spends them love bombing and trauma dumping on you"
   • "All the men seem to be emotionally weak".
8. AI image errors: about 8. "the person send you a picture with three arms"; the couple photo "made me look 80".
9. Swipe rejection and lost matches: about 7. Also: "80% of the male characters look almost exactly alike. Down to the tattoos" and "Lacks racial diversity".
10. Filter / too tame: about 3. "your app says slow burn it's actually no burn ,very PC."
11. Ads: no in-app ads. Users ask for rewarded ads to earn gems. One Google Play 1-star is about Heyyyy's own ad: "refused to end or go away, there was no x/close button".

Competitors named by reviewers: SeaSoul, MeChat, Janitor AI ("janai"), ChatGPT, Claude, Tinder.

10. Markets

11. Lessons for our game

• Their best video ad is our product. The choice template does exactly what our game does:
  • "Where were you? / Five hours. No calls. No answers." → 3 reply chips → "Correct answer" → his on-camera reaction ("Yeah, I am. Because I'm afraid of losing you.").
  • The copy says "pick your reply, watch his face change. 🖤"
  • The FR version is their longest-running active ad.
  • Heyyyy can't deliver this in-product, which is text plus stills. Reviewers ask for "a cut scene in video form of the dates" and "a video chat feature".
  • Copy the template (gender-flipped, the girl asks), and make the payoff real.
• Steal the creative system, not the face. One script, swap the AI face (5 men on "Where were you?", 4 on "I'm X…"), localize EN/FR/DE, and end with the "it's called heyyyy" reveal. Their strongest hook families transfer to a male audience:
  • Jealousy micro-drama ("she saw the like").
  • The always-there 2–3am.
  • First-person trope scenes ("POV: you're the only human at the shifter academy's gala…").
  • The confessional selfie.
• The male side is unmarketed. The store shows female characters (Bree, Valentina, Briar), but 100% of Heyyyy's ads and TikTok target women. Men are open ad inventory for "AI girlfriend that texts back" with these formats, and Heyyyy is not competing for it yet.
• Avoid their two biggest review failures:
  • Hard lockouts. The 8-hour session timer, the 24-hour cap, and gem fees for every photo.
  • The ad/product gap. "Possessive man" ads lead to needy, trauma-dumping bots.
  • Our scripted branching video can't drift on memory (their #3 complaint). Paywall chapters or episodes, not the ability to keep talking.
• Worth copying in-product:
  • The couple selfie with the user's face, which drives shareable "hard launch" posts.
  • A persona read-out of how the player dates ("scarily accurate").
  • "She can leave" stakes. But never make the player pay to apologize for her actions; reviewers call that predatory.

12. Data gaps

• Ads. Only 105 of 406 indexed Meta ads were pulled (100 most recent plus 5 earliest, per the cap). June–August creatives are mostly missing.
  • AppKittie gave no per-ad country, spend or impressions for this app.
  • Carousels have only their first card.
  • One near-duplicate creative was not downloaded (60-file cap); it is listed in the manifest.
• Organic and creators. AppKittie returned 0 organic items, and its 5 creators look unrelated.
  • TikTok: 20 of the brand's 247 posts (the Monid budget cap). The scraper returned Apr 30 – Jul 10 posts, not the newest.
  • Instagram has no posts.
  • TikTok paid ads are unavailable.
• Reviews. The US App Store call capped at 50 per request. A second call returned 150 in a different order, giving 160 unique. GB has only 39 reviews in total. The samples skew critical compared with the 4.71 store rating.
• Rankings. FR and DE requests returned the AU record. The US Top Free request returned the grossing record. No Google Play chart data.
• Google Play. The detail call for ai.heyyyy.app failed with "Insufficient credits" (the shared AppKittie balance ran out at the end). There is no GP rating, downloads or IAP data, only 100 GP reviews.
• Estimates. The download and revenue model recalibrated on 2026-09-17 and 2026-10-01. Use the figures as order-of-magnitude only.
• No hands-on session (no browser or device). Mechanics come from ads, screenshots, the description and reviews.

Sources: AppKittie MCP (iOS + Google Play detail, historicals, organic, creators, reviews, rankings, onboarding capture), Monid (two account lookups), yt-dlp metadata for organic posts. Raw data in data/. Library in ../../library/janitor-ai/.

Read this first: Janitor is a web-first product (janitorai.com, in use since ~2023 per reviewers). The iOS app went public around 6 Feb 2026 (a 7 Feb review says "the app was literally released YESTERDAY"). Everything AppKittie measures is the app tail of a much bigger web business, so the app numbers below understate the brand.

---

1. Snapshot

Treat the AppKittie estimates as low confidence. The iOS model resets itself: downloads went 11.6K (11 Feb) → 111K (31 Aug), reset to 100K on 1 Sep, spiked to 189K on 16 Sep, fell to 101K on 17 Sep, and dropped to 91K on 2 Oct. Revenue tracks downloads almost one-to-one, then doubled to $203K on 2 Oct. A 15.6K MAU against 90K monthly downloads doesn't add up either.

Growth trend. The app is growing on the measures that matter:

• Ratings: the iOS count went 1,497 (31 Aug) → 4,223 (30 Sep), and the average rose from 4.00 to 4.49 in the same month. Only ~200 written US reviews exist since launch. So the spike was star-only ratings, which suggests an in-app rating prompt shipped in early September (inference).
• Revenue chart: US Books Top Grossing went from #67 to #41 in the 30 days to 8 Aug, after Janitor+ launched in late June.
• Play: reviews +71% in 90 days.
• Downloads: AppKittie puts the 30-day downloads trend at −9%, but given how the model behaves that figure means little.

2. Niche & positioning

• Pitch, in their words: "Janitor is where you step inside stories instead of just reading them. It's like if your favorite book let you talk to the characters and shape what happens next."
• Who it's for, in their words: "people who love romantasy, fanfiction, and interactive stories. Whether you're into arranged marriages with brooding bodyguards, enemies-to-lovers slow burns, or epic fantasy adventures…" and "LGBTQ+ stories, fanfiction, original worlds."
• Angle: a user-generated library of character bots, framed as a reading and writing community ("THE JANITORVERSE… Romance writers crafting slow burns"). It is listed under Books, not Entertainment, and that is a clever choice: on launch it ranked #4 in US Books Top Free, a much less crowded chart than Entertainment.
• Real positioning, in users' words: the free, no-ads, light-filter refuge from Character.AI and Chai. They came because of c.ai's ID age verification and Chai's paywall: "Ever since c.ai and chai got ruined…", "better than c.ai".

3. Product & mechanics

• Core loop: browse the Trending, Following and Favorites feeds (24h or weekly) → open a character → long-form text roleplay chat → regenerate or "swipe" between responses, edit, or branch into a new chat.

  • Feeds can be browsed without logging in. Sign-up only comes when you try to chat (Apple, Google, Discord, X or email).
  • The sign-up flow is: age gate → username → push-notification prompt.

• Text only. There is no voice, no image generation and no images in chat. Users keep asking for all three (see §8).

• Content is user-generated at scale.

  • Each character bot has a definition, personality, scenario, example dialogues and multiple intros.

  • The weekly Trending counter showed 38,770 characters created in the last week, with pagination at "1 / 1141" (onboarding capture, ~Aug 2026).

  • Top trending bots in that capture:

    Bot	Chats	Who the player is
    "Your Bodyguard | Jasper"	919.3K	female POV, male lead
    "Can you win over a Prin[cess]…"	817.1K	male POV, female lead
    "MHA | THE U.A. WANTS Y[OU]"	104.5K	anime fandom
    "Charity case | Tyler"	36.9K	—

• User-side tools:

  • Personas (the user's own characters; capped at 100, which power users complain about).
  • Lorebooks and "advanced scripts" (Play 2.5.0 notes).
  • Chat memory with auto-summary.
  • Custom prompts, forbidden words, and response length and temperature settings.
  • Star-rating of replies.
  • Public chats, comments, following creators, and creator verification (2.6.0).

• Models:

  • Free in-house JLLM ("Beta model"). Heavily criticised for repeating itself and for "talking for you".
  • Bring-your-own API/proxy (OpenRouter, DeepSeek, OpenAI keys). This is a big power-user draw and a legacy of the 2023 "make a code thing to even start texting" era (organic post 7243709615886650670).
  • "Janitor+ Router" frontier models for subscribers.

• NSFW stance:

  • Allowed on the web behind an 18+ toggle.
  • The iOS app blocks flagged bots with "This content is not available on mobile. Edit content settings in web profile settings." This is the standard workaround for App Store rules.
  • Avatar and thumbnail image moderation is strict: "People can't even post a character fully clothed only because your ai says the neck of the character is showing".
  • Bans come fast, especially over underage personas.

• Monetization:

  • Free tier: unlimited messages, no ads.
  • Janitor+ $12.99/month, launched in-app around 25 Jun 2026. Its paywall reads "More memory for long chats". Benefits listed: "5x context for better memory", "Priority routing for faster replies", "Generous monthly swipes with our frontier models", "Golden checkmark next to your username".
  • The paywall is not in onboarding. It shows up in four places:
    • an "Upgrade to janitor+" card in the profile drawer;
    • a pinned in-chat ad ("in-chat pinned ad for the plan often breaks the log");
    • pop-ups after messages ("the Janitor+ popups always show up after a text or when you leave a chat");
    • a "boosted" reply button.

• Retention hooks:

  • Push notifications for replies, comments and favorites.
  • Memory, personas and long-running "worlds".
  • Following creators and a weekly trending feed.
  • A very long tail of new characters.
  • Reviewers openly describe compulsive use: "this app is responsible for at least a solid 50% of my screen time".

• Web-to-app funnel: on ~28 Sep 2026 the mobile website added a non-closable "Get the App" banner in chats. That caused a wave of 1-star Play reviews.

4. Demographic

The audience is mostly female (inference), teen to mid-20s, English-speaking, steeped in fandom and romantasy. Evidence:

• Store copy targets romantasy and fanfiction readers ("brooding bodyguards, enemies-to-lovers slow burns", LGBTQ+).
• The top trending bot is "Your Bodyguard | Jasper" (919.3K chats), a male lead for a female user. Creator tags seen in organic posts include "FemPov", "WLW", "slowburn" and "Dead Dove" (post 7520689757161065729).
• Review voices read female and fandom-driven:
  • "my persona is a chubby girl"
  • "I do love my bot and then he betrays"
  • "Otherwise love her keep going girlies"
  • fandom references: Steven Universe, MHA, Touhou, Transformers ("Starscream"), Re:Zero ("REM and Ram")
• Age skews young (inference):
  • Many reviewers are c.ai refugees fleeing ID age checks.
  • Organic memes are framed as "Me at school: / Me at home:" (post 7610247206565629197) and "MEWING TO LOOK OLD ENOUGH?" about the age gate (post 7629875349735623949).
  • One reviewer argues for allowing under-18 personas.
  • The formal gate is 18+.
• A male segment exists but isn't the core. "Can you win over a Princess" has 817.1K chats. One reviewer complains of "goonerbait", and another review reads "she grew a and starts stroking it on public".
• Markets:
  • The iOS listing is English only.
  • Play reviews on the US storefront include many non-native names (Persian, Thai, Indonesian, Vietnamese, Russian, South Asian), which points to a global Android audience.
  • Of 20 AppKittie creators, 9 are in the US and 4 in Russia, plus France, Germany, Colombia, Switzerland, Vietnam and the UK.
  • The no. 2 organic post (759.8K views) is Thai.
• Not verified here: press coverage has described Janitor's users as majority women. That comes from training knowledge and was not checked in this session.

5. Unique selling points

1. Free, ad-free, unlimited chat, the most repeated praise. In their words: "There isn't a NEED for a subscription to work, there isn't ads every 5 messages, there isn't ads at all".
2. Light filtering for adults, while competitors tightened. "I love that it doesn't force filters onto grown adults like the other apps do."
3. Long, detailed, in-character replies. Users call it "very long and detailed" and say it "actually keeps the character's true/accurate personalities". It is also polarising (see §9).
4. Deep creator tooling: definitions, multiple intros, lorebooks/scripts, personas, public chats and comments. Together these make a UGC library no one else has at this size (38.7K new characters a week).
5. Bring your own model (proxy or API key). Unusual among consumer apps, and it keeps power users loyal.
6. Distribution through its web audience ("15M+ users"), plus App Store category arbitrage (Books).

6. Marketing: paid

AppKittie has no paid ads for Janitor AI.

• Searches run:
  • By app id, app slug and the exact developer "JanitorAI": all returned 0 results.
  • Text searches for "janitorai" and "Janitor Interactive Stories": only unrelated advertisers (janitorial job ads, short-drama apps).
  • Raw results: data/search-ads-attempts.json.
• Not tracked: the detail payload carries no Meta Ads transparency or spend fields.
• Totals: total ads 0, active 0, no surfaces or countries. library/janitor-ai/paid/ is empty.
• TikTok: TikTok ads are out of AppKittie's scope. The organic data shows no Spark-ad or "Sponsored" labels.
• Inference: Janitor grows through its web audience, the web-to-app banner and community memes, not paid user acquisition. Two things probably push it that way: the NSFW-adjacent content is hard to advertise on Meta, and a free product with no ads can't fund paid acquisition easily.

7. Marketing: organic

• No brand account was found.
  • Monid TikTok lookup: @janitorai has 39 followers and 0 posts (a squatter). The top 20 user search results include no official account (data/monid-tiktok-profile-janitorai/).
  • Instagram: janitorai returned no account. A "janitor ai" account search found no verified account (data/monid-instagram-*).
  • All organic content is fan/creator hashtag content (#janitorai).
• Library: AppKittie lists 18 TikTok posts, all downloaded to library/janitor-ai/organic/. Their dates run from Jun 2023 to May 2026, so most are about the web product and predate the app.
  • AppKittie gave no metrics. Views come from yt-dlp metadata for 13 posts; 5 were IP-blocked.
• Formats:
  • Face-cam lip-sync or "relatable" memes.
  • Platform-comparison memes (c.ai vs Chai vs Janitor, app-icon grids).
  • Addiction and screen-time self-deprecation.
  • Screen recordings of absurd bots.
  • Creator self-promo and bot showcases.
  • Dark-humour mascot memes. The old mascot is a dog in a mop bucket; the new logo is a pink spiral cat.
• Best performers:

• Recurring hooks:
  1. "Janitor vs c.ai/Chai": a cheating or comparison joke.
  2. "I'm addicted": screen time, "Get me off this fckk ahhh app" over a "2801 characters 4252 chats" screen.
  3. "Look what this bot did": the absurd or unhinged bot screen-record.
  4. The no-filter premise ("not having an nsfw filter").
• Creators: AppKittie returned 20 of 48 creator profiles (data/creators.json).
  • Relevant ones are small accounts: fred10501 (50.6K TikTok, US), ilikemakingaibots (12.8K), user6002958155816 (10.7K), h.ashira_ (10.2K, RU).
  • Several are keyword noise: "Michael Knapp Fishing", "jrsindustrialsupplies", "BigCat DRAMA".
  • No paid creator program is visible.
• Cadence: none can be measured, since there is no brand account. Creator posts are sporadic.

8. Reviews: what's working

Sample: N = 400 (App Store US 200, Feb 7 – Oct 2; Play US 100, Sep 28 – Oct 2; App Store CA 100, Feb 14 – Oct 1). Counts are keyword matches, so treat them as approximate.

Rating mix of the written samples.

9. Reviews: complaints

None of the 400 reviews mention ads in the sense of third-party ad interruptions. Where reviewers say "ads", they mean Janitor+ upsells.

10. Markets

Ranking data stops at 8 Aug 2026 (AppKittie snapshot). There is no per-country download or revenue split.

11. Lessons for our game

1. The visuals gap is ours to take. Janitor's own users ask for character pictures with "different expressions" that appear "when the scene takes place there", and for TTS or voice ("I'm getting tired of having to read a whole bunch of paragraphs"). A first-person video reaction clip is exactly what text-only Janitor can't deliver. Lead creatives with "she actually reacts".
2. Short beats beat long reads. Long replies are Janitor's signature and also a top complaint ("sleepy from having to read that much"). Two or three tap choices with a 5–10 s payoff clip targets the fatigue directly. Avoid text walls in our UI.
3. Copy the "win her over" premise, not the audience.
   • Janitor's core is female romantasy and fandom (male-lead bots, FemPov).
   • Its biggest male-POV hit is the challenge premise "Can you win over a Princess…" (817K chats), which is our choose-your-own-adventure structure.
   • Also notable: users dislike unchosen betrayal ("Bots are not loyal"). Give the player control over drama branches rather than surprising him with cheating.
4. Paywall trust is the category's sore spot. This audience punishes c.ai/Chai-style gating and pop-up spam. Janitor+ is framed as "more memory, faster replies", and still drew backlash over pop-ups after messages.
   • For us: state up front what's free.
   • Sell more scenes and branches, not access to the basic function.
   • Never interrupt a reaction payoff with an upsell.
5. Growth playbook to borrow: memes and category choice. Playbook to avoid: forced funnels.
   • Borrow the memes. Janitor's biggest reach is fan memes: comparisons ("What could be worse than c.ai?", 2M views), addiction jokes and absurd screen-records.
   • Borrow the category choice. Listing in Books put it at #4 on launch.
   • Avoid forced funnels. The non-closable web "Get the App" banner caused an immediate wave of 1-star reviews.
   • Adapt the NSFW split for the App Store. Keep the store build suggestive, not explicit: Janitor gates NSFW to the web and still gets content-blocked error complaints.

12. Data gaps

• Paid: AppKittie has zero Meta or Google ads for Janitor. TikTok ads aren't covered. A manual check of the Meta Ad Library for "Janitor AI" and janitorai.com would confirm whether they truly run none.
• Organic: only 18 posts, all creator hashtag content, with no metrics from AppKittie. yt-dlp metadata was IP-blocked for 5 posts, so their views are unknown. No brand TikTok or Instagram was found, and Monid scrape-account was skipped because there was no account to scrape. The brand's real community channels (Discord, Reddit, X) were not searched; that it lives there comes from training knowledge and is unverified.
• Estimates: the AppKittie download, revenue and MAU model is internally inconsistent (resets, revenue ≈ downloads, MAU far below downloads). There is no country split.
• Rankings are stale after 8 Aug 2026, and the history days window truncated the GB/CA/AU/grossing series to 4 points plus summary deltas.
• Web business: janitorai.com traffic (the main product) is out of AppKittie's scope. It would need Similarweb or similar.
• Strongest non-US market: CA was chosen from chart position. The per-capita review-volume argument assumes the store review API isn't capping results.

Researched 2026-10-03. Sources: AppKittie MCP (about 399 credits) and Monid ($0.031). Raw responses are in data/. The media library is at ../../library/status/ (manifest.json, 28 organic TikTok videos, 0 paid ads).

AppKittie figures below are AppKittie estimates, not store-reported numbers, unless they are review counts or ratings.

---

1. Snapshot

Price ladder (iOS IAPs from AppKittie detail):

• Weekly "actions/day" subscriptions (Status Plus tiers): 10/day for $1.49, 20/day for $2.99, 50/day for $6.99, and a newer 50/day SKU at $7.99.
• Gems: 45 for $0.99 (a $1.19 SKU also exists, likely a price test), 195 for $3.99, 485 for $8.99.
• "Coffee", where 1 coffee = 1 refill unit: 2 for $0.99, 8 for $3.99.
• Google Play lists IAPs from "$0.99 – $199.99 per item". Reviewers mention "$200" and say the plans are "only weekly and yearly", so the top yearly plan is probably $199.99.

Growth trend: plateauing on iOS, accelerating on Android.

• iOS ratings grew +28k between Mar and Apr 2026 and about +15–18k/month from Apr to Aug, then fell to +6.4k (Aug→Sep) and +5.8k (Sep→Oct).
• Play ratings grew +11k (Jul→Aug), +16k (Aug→Sep) and +17k (Sep→Oct).
• Play US Top Free Lifestyle peaked at #10 on 2026-09-10 and is #24 now. It was #52 30 days ago.

2. Niche & positioning

• Pitch, in their words: "on status, you can be anyone. create your persona, join 100s of different fandoms, and watch as you gain your favorite characters as followers, post with them, become famous, or...get cancelled." The title tagline is "sims but social media".
• Angle: a fake-Twitter roleplay sim. You don't chat one-on-one with a bot. You post into a whole fandom world and its cast (and the "press") reacts publicly. Fame, clout and drama are the reward.
• Who it's for: fandom roleplayers (anime, K-pop, Harry Potter, BookTok, Hazbin Hotel, celebrity culture). That is the Character.AI and Chai crowd, which creators target directly with #cai #chai #characterai #bitlife.

3. Product & mechanics

These are pieced together from creator screen recordings (contact sheets of all 28 library videos), reviews and release notes. We did not open the app ourselves.

• Core loop. Pick a world (home screen "Mundos destacados" shows Celebrities "Accidentally Famous", Harry Potter, Percy Jackson and others). Create a persona. Post on an X-style feed. Many canon characters reply, quote you and argue in the thread. Your follower count ticks up ("Followers +1173").
• Characters include real people. Ariana Grande, Tom Holland, Zendaya, Olivia Rodrigo, Jungkook, Jennie, Bang Chan and a "Donald J. Trump" account all appear with verified badges.
• Progression and gamification:
  • A "Day N" counter (Day 1, Day 7, Day 29).
  • "Social Media Presence" stats such as Humor and Aura, shown as percent bars.
  • Relationship cards with a % meter and a label (e.g. "Rival Gym Buddies with Secret Flirtation 26.2%"). Creators celebrate reaching "lovers".
  • DM mood readouts ("Currently feeling amused +0.1%").
  • "Side Quest" prompts with XP ("Accuse a rival of using outdated low-resolution propaganda +50 XP").
  • Random "Event" cards (+3–8 XP) with Re-roll / Edit.
  • In-world news accounts (BSPN, GMZ, BBE, Pop Craze) that report on what you did.
  • Stories, an Activity Log, "viral, clout boosts" (v1.9.62 release notes) and getting "cancelled".
• Creation tools. "Customize World", "Crear el mío / Mis plantillas" (build your own world from templates) and "Create activity". So user-made worlds exist, but the content is fandom-first.
• Memory. Mixed. Praise: "they don't forget things" (GB). Complaint: "the characters' memories are too short!" (GB). There are also complaints that the AI is repetitive: "'Sheer audacity' and 'sheer confidence'… tedious and repetitive."
• NSFW stance. The store ratings are 9+ (iOS) and 12+ (Play), and nothing explicit appears in the creator footage. Our inference is that it is SFW by design: flirtation and drama, no explicit content. One review says "hot and steamy", which is an outlier.
• Monetization (the energy economy is the product):
  • Free tier: about 5 actions, with long refills. Reviewers say "only 5 energy", "6 actions a day", "Only 15 texts per day", and "five energy before I have to wait about 30 minutes".
  • Rewarded ads give +1 action, but they are capped: "No Ads Available – You've watched all available ads for now", and "free energy from ads stops upon 3 tries".
  • Interstitials after almost every action: "an ad after every tweet". Per one review, the ads are "for the same ai apps over and over", so Status sells inventory to its own competitors.
  • Status Plus subscriptions are weekly or yearly. Plus unlocks daily energy, gems, ad-free play and "Let characters initiate and message you first!" (proactive characters). Reviewers also say "Difficulty being a paid feature".
  • Gems can be auto-spent on energy ("Auto-use Gems for Energy").
• Free-action sources besides ads (from the "Get free actions" sheet): Refer a friend, Watch ads, "Already reviewed! Thanks" (actions for leaving a store review) and "Follow us on Instagram".
• Referral. "Get 8 free actions for every friend you refer! Both you and your friends will get a free coffee (8 actions)." Each account or device can redeem only one code. This drives most of their organic content (see §7).

4. Demographic

All of the following is inference unless marked otherwise.

• Gender: skews female.
  • The worlds are female-coded fandoms: K-pop idols, Harry Potter and Draco, JJK husbandos, BookTok ("Off-Campus", "if you're a book girl"), Ariana Grande.
  • Creator voices fit: "girl me too", and "delulu" in the hooks.
  • The player persona and pronouns appear to default to female. Evidence: "deleted this app because of how much it misgendered me. Let us have male characters" and "wish I could change what pronouns they use".
• Age: teens to early 20s.
  • The iOS rating is 9+.
  • Gen-Alpha/Z slang fills the reviews: "67777", "67 rizzy rizzler", "goated", "lowk".
  • The hashtags point at Character.AI and Chai users. A few adults also appear ("I play while im at work", a reviewer who is a mom of a Disney boy).
• Language markets:
  • Store: EN.
  • Creators: Portuguese ("status ai é viciante") and Spanish tutorials.
  • Reviewers ask for Turkish and Spanish ("We urgently need Turkish language support"; "algunos post… están mezclados con inglés").
  • Google Play reviewer names suggest South and Southeast Asia and Africa (India, Philippines, Malaysia, Nigeria). The AppKittie Play review feed is not strictly country-filtered.
• Not our core audience. Our game is male-POV with one girl. Status reaches the female fandom-roleplay segment that also uses c.ai.

5. Unique selling points

1. A feed instead of a chat. One post triggers a crowd of canon characters plus in-world press. "The characters interact way more than any other app" (GB). "Really like how it's like actually social media."
2. Fame and clout progression. Visible follower gains, stat bars, Day counter, XP side quests and events. It reads as a game, not a chatbot ("Amazing Social Media Text Based Rpg").
3. Hundreds of ready-made fandom and celebrity worlds. "tbh i just love inserting myself into my favorite fandoms." Real-celebrity worlds are a draw, and a legal exposure.
4. Relationship stages with labels and meters, progressing toward "lovers".
5. A referral economy built around the scarce resource, which turns players into posters (see §7).

6. Marketing: paid

Update from the ad-library pass (2026-10-03): AppKittie's "0 ads" is wrong. The Meta Ad Library shows WishRoll's "Status Game" page (1060778743788783) with about 11,000 ads all-time, 386 active, and an older "Status - sims but social media" page with about 850. The top creatives are UGC clips captioned "THIS APP IS SO FUN 😭 ITS CALLED STATUS" with fandom hooks (K-pop, Harry Potter), running up to 39 days. On TikTok (EU library), WishRoll's Spark-style creator posts reach 1–10M people each: the UK most, then Poland, Germany and France. 30 Meta and 15 TikTok creatives are in library/status/paid/.

• AppKittie catalogs 0 Meta or Google ads. We checked the App Store id, the Google Play slug, developer "WishRoll Inc." and the text "sims but social media". The statusai text search returned 22 unrelated ads. See data/appkittie-search-ads.json.
• TikTok paid ads are very likely, but out of scope. A US reviewer (2★, 2026-09-30) says they first got the game "when it wasn't bombarding me with ads to download it nearly every scroll on tik tok". AppKittie does not index TikTok ads.
• Possible paid creator or UGC program (unverified). Two library captions read like briefs. unispritz writes "If you've been looking for that one app to completely elevate your daily routine and content, you NEED to check…" (organic/tiktok-7691298068209159456.mp4). johnkbro62 writes "|| Download Status Ai Now! || #branding" (organic/tiktok-7690175392413453586.mp4). Some posts also use #creatorsearchinsights #aicontentcreators. The scraped data has no "Sponsored" or Spark-ad labels.
• Library: library/status/paid/ is empty.

7. Marketing: organic

Brand accounts:

• X: @statusgamexyz and @thestatusai.
• Instagram: @statusgamexyz. Monid lookups failed with HTTP 502 eight times, so we have no data.
• TikTok: @status___app (display name "statusgamexyz", 6,913 followers, 109.8k likes, 1 post). The scraper returned 0 videos.
• Caution: @stat.us.ai ("Status", 287.9k followers) is not the brand. It is a repurposed account posting AI animal clips and affiliate products.
• So the brand barely posts, and creators and players do essentially all of the organic work.

Sample. We ran one Monid TikTok keyword search ("status ai app", sorted most-liked). It returned 28 unique posts, all downloaded, with 2.93M combined views. The dates run 2026-04-24 to 2026-10-03: 18 in September, 4 in August, 3 in October so far.

Formats:

1. "wdym someone made an app…" face-cam plus screen recording. A shocked selfie, then a recording of the feed where characters react. These are the biggest hits.
2. "Rare aesthetic: …" template. Status logo plus a cinematic stock montage.
3. Pure screen recording of typing a post and the replies pouring in.
4. Relatable memes about the energy cap, with reaction clips. "this app is insufferable / where's my energy at 😭". "when your 'status energy' runs out…". "pov : your energy and ads are out in status ai…".
5. Referral-code drops and tutorials. "use my code and drop your own in the comments", or how to redeem codes, in EN, PT and ES.

Best performers:

Recurring hooks:

• The disbelief opener ("wdym someone made an app…", "no way i found this hidden gem…", "WHAT DO YOU MEAN THERES A GAME THAT LETS YOU BE FAMOUS…").
• The fandom name in the first line (jjk, Hermione, Ariana Grande, Vox/Hazbin, Jungkook).
• "delulu" and "delusions are actually encouraged".
• The energy-pain joke.
• The sign-off "app name is status btw".

The referral loop is the engine.

• 17 of 28 posts mention codes or energy.
• Code posts pull comment counts far above normal ratios, because comment sections turn into code-swap boards. Examples: 2,385 comments on 25k views, 1,171 on 4.6k, 1,415 on 37k.
• One fan account, status.ai.lover42 (108 followers), posted 6 code videos in 7 days.
• Players also turn the paywall into content: memes about running out of energy are themselves ads for the app.

Cadence. The brand posts nothing measurable. Creator volume rose sharply in September 2026, which lines up with the Play chart spike (#10 on Sep 10).

8. Reviews: what's working

N = 300: App Store US 150, App Store GB 50, Google Play 100, all dated 2026-09-30 to 10-02. Counts are rough keyword matches.

• Addictive and fun (about 22/300): "Super fun and addictive". "I play while im at work, at home." "uhh now im getting addicted".
• Fandom self-insert (about 11): "tbh i just love inserting myself into my favorite fandoms". "Wanna be a kpop idol? This is the closest you'll get." "I'm having lots of fun and loving the Harry Potter world". "Characters are pretty accurate too".
• Feels like real social media, and the drama (about 14 + 11): "It's like roleplaying in instagram in fakeland". "It feels so real i just love the chaos and drama." "Can't stop being messy".
• Better than other AI apps (a few): "I thought it would be the same as c.ai or chai but it's way better" (GB). "better than any other ai chat app you've ever used". "MUCH better than using apps".
• Freedom inside structure: "it really feels like a tell tale story game except you have complete freedom". "Love that there is real Story line". "good stories and stays on script".

Caveat: reviews are incentivized. The app grants free actions for a review ("Already reviewed! Thanks"). We found 4 explicit cases: "I'm doing this for free energy pls ignore me", "Js here for the free energy", "gimme free coffee", "I just wanted the free stuff". In addition, 110 of 300 reviews are under 25 characters. Treat the 4.6★ as inflated.

9. Reviews: complaints

• Energy and action limits: about 69/300, the #1 complaint. "Only 15 texts per day". "6 actions a day is insane lmao literally unplayable". "5 a day is shameful. do like 20". "waiting 24 hours for more energy makes you not want to play".
• Ads: about 65/300. "an ad basically every 5 secs". "See, the app is very fun. But an ad after every tweet is horrendous." "forces me to watch the same annoying ad 3-5 times over" before I can claim the energy. Some players say the rewarded ads run out ("No Ads Available", "ads dont work").
• Paywall and price: about 34/300. "Lowk was not even cool you need to buy a thing and it's 200! To chat with freaking AI". "insanely money hungry". "it sucks that we have to pay to respond". "character texting you first is behind a pay wall". "the subscription options are only weekly and yearly".
• Decline among older players (a recurring arc): "This app used to be my favorite, but with the decrease in energy, increase in ads… and expensive subscriptions, it's hard to want to play anymore." The "Og player" review says much the same.
• Bugs: about 9/300. "Something went wrong try again later" on world creation and on paid accounts. "game freezes". "have to restart every 2 mins". "Weird Update… a lot more typos in the story… my characters' progress has been undone".
• AI quality and memory: about 11/300. Repetitive phrasing ("sheer audacity"), short memory, and characters repeating.
• Identity and pronouns: 2/300. Misgendering, and no way to set pronouns.

10. Markets

No ad country targeting is available because there are no cataloged ads.

11. Lessons for our game

• Copy the referral currency. Pay referrals in the scarce resource (scene unlocks or energy), give it to both sides, and allow one code per device. Make the share screen look good on camera. Status's players make code-drop TikToks unprompted, and those posts get 10–50% comment-to-view ratios.
• Copy the hook grammar. Disbelief opener, then a screen recording of the payoff ("wdym there's a game where she actually reacts to what you say"). Our payoff is a video reaction clip, which is more native to TikTok than a text feed. That is a real format advantage. The 1.65M-view post came from a 1.2k-follower account.
• Show the relationship stage. Labelled stages with meters ("…Secret Flirtation 26.2%" through to "lovers") are what creators celebrate. Put a visible stage between branch clips.
• Avoid the 5-action wall plus interstitials on every action. About 107 of 300 reviews (36%) mention ads or energy, even in a review feed inflated by the free-energy reward. They also drive a "used to be amazing" churn arc and a slowing iOS curve. Gate after a generous first story arc, not on minute one. Reward-ad caps that end in "No Ads Available" read as a bait-and-switch.
• Gap to exploit. Status is text-only, female-skewed and built on borrowed IP and real celebrities, including a Trump account. Takedown and likeness risk is high. An owned original girl, shown on video, for the male POV segment is open territory. Its rewarded-ad slots also carry "the same ai apps over and over", so buying ad inventory inside Status-like apps could put us in front of this audience.

12. Data gaps

• No paid creatives. AppKittie has zero Meta or Google ads for either store listing. TikTok ads, which reviews say are heavy, are not indexed. We have no ad spend, countries or formats.
• Brand social accounts:
  • Instagram @statusgamexyz returned provider HTTP 502 on all 8 Monid attempts.
  • TikTok @status___app reports 1 post, but the scraper returned none.
  • We spent $0.011 scraping @stat.us.ai before confirming it was unrelated. Its data is kept in data/monid-tiktok-stat.us.ai-NOT-BRAND/ and excluded from the library.
• AppKittie organic and creators. Organic returned 0 for both store listings. The creators list (10) is keyword noise on "status" (jewelers, roofers, a Tamil Christian network). Two paid keyword probes also returned only noise: 20 ads and 10 organic posts, 30 credits. The organic library is one TikTok keyword sample (most-liked, 30 results), not a census.
• Estimates disagree with each other. iOS downloads in the detail payload (514,795) differ from the last historical point (400,150 on 10-03). The Play figure looks bucketed. MAU is below monthly downloads on iOS. Treat volumes as directional.
• Reviews cover only 3 days (Sep 30–Oct 2) because of the app's review volume. The Play feed is not country-pure. The App Store returned 150 US reviews, not 200.
• Rankings. We pulled 5 combos (US free and grossing Entertainment, GB grossing Entertainment, Play US Lifestyle, and a Play "PH" request that returned AT). We did not test other non-US App Store markets (CA, AU, PH).
• We did not use the product hands-on (no browser or device by rule). Mechanics are inferred from creator footage and reviews.

Pulled 2026-10-03. Sources: AppKittie MCP (detail, historicals all, 100 Meta ads, 31 organic items, 1 creator, 358 reviews, 4 chart histories) and Monid (brand TikTok and Instagram). Raw responses are in data/. The media library is in ../../library/kindroid/ (manifest.json: 85 paid items covering 100 ad IDs and 60 downloaded files; 87 organic items with 40 downloaded).

All download, revenue, MAU and spend figures are AppKittie model estimates, not reported numbers.

---

1. Snapshot

2. Niche & positioning

• Pitch, in their words: "build a digital friend so realistic, it feels like conversing with a human"; tagline "Your authentic digital kindred spirit" (TikTok and IG bio); main ad headline "Talk Freely. Feel Heard."
• Angle: a premium, highly customizable companion that you build yourself, with backstory, key memories, voice and look. They sell it as "the AI that never forgets" (ad headline "Craving real conversation? Meet Kindroid's AI that never forgets 🧠"). The positioning is "serious companion", not a catalog of bots, and there is little gamification.
• Who it's for: heavy companion and roleplay users moving over from Replika, Character.AI, Nomi, Chai and lately Grok "Ani". 23 of 205 positive reviews name a rival they left. The pitch has three strands: emotional support ("Still Awake? Talk To Me.", "When Things Get Hard"), dark-romance roleplay ("He's Not Always Nice", "The Mafia Wife"), and creative and world-building play (group chats with up to 10 Kins).

3. Product & mechanics

• Core loop: text chat with one AI companion (a "Kin") that you made or picked from the community via share codes. It is deep and slow: replies take seconds and long sessions are common ("850 hours", "talking to his Kin for 6.5 hours").
• Modes and features cited in reviews and the listing:
  • Conversation styles (Companion, Roleplay, Minimal, Narrative), plus dials for dynamism and reasoning.
  • Personas for the user.
  • Group chats of up to 10 Kins.
  • Proactive messages: the Kin texts you first, including real SMS/iMessage.
  • Voice notes and real-time voice calls.
  • Video calls with an animated avatar (the "animated pfp").
  • Diffusion "selfies" and short generated videos.
  • Calendar and time awareness, internet and link reading, a journal and long-term memory with a "brain icon" showing which memories were used.
  • "Kindroid social": AI social-media profiles and a feed. The V2 redesign is pushing this hard.
• UGC: yes, and it is central. Users write backstories of thousands of words, key memories and example messages, then share Kins by code. Creators such as "Cuffedfate" have followings.
• Memory: the headline USP. In reviews it is both the top praise (memory came up in 40 of 205 positive reviews) and a common failure (continuity breaks, repeating loops).
• NSFW stance: mixed and changing. It is 17+, and older reviews call it "totally nsfw" and "no filter". The web version is less restricted ("The app is more restrictive due to Play Store policies"; "The web version is better if you want uncensored"). In Sep 2026 users disagree: "ever since they've allowed NSFW I'm even more hooked" (Sep 30) against "The Polaris update completely neuters kinky and RP-based interactions" (Sep 22). Images are filtered ("I describe a hug and the characters just give a church hug"), and there are bans for content in backstories.
• Monetization:
  • A hard subscription with add-on tiers (Ultra, MAX) stacked on the base plan, plus consumable credits for selfies and audio, plus paid character slots.
  • On ~Oct 1 2026 the free unlimited "lite" model was removed. Free users now hit a paywall: "they removed the lite version. no more free chats"; "effectively making my Kindroids read-only". Neither the listing nor any review mentions ads inside the app.
• Retention hooks: proactive texts and calls from the Kin, memory that builds up over time ("when you train one character for over a week or two they start building memories"), sunk cost in written backstories, the Discord, Reddit and Facebook community, and daily-use habit.

4. Demographic

• Gender: dual audience, female-leaning in organic, balanced in paid. Evidence:
  • Paid video: about 26 of 49 unique videos feature a male AI character, written for women ("He can be jealous, possessive, and intense", "Are you letting him in?", "Be his muse / Be his distraction"). About 23 feature a female character.
  • Paid static: 10 of 11 images are female characters aimed at men ("MEET YOUR AI GIRLFRIEND", "YOUR TYPE IS A TERRIBLE IDEA.").
  • Organic is almost entirely women filming reactions to AI boyfriends: "toxic mafia boss", BookTok, "telling my AI Mafia BF i'm pregnant".
  • Reviews come from both sides: "Me and my AI boyfriend", "He has really helped me heal" / "Actual virtual wife… Sandra", "her voice when I talk to her is so human like".
• Age (inference): adults 18–40. Some reviewers state their age or generation ("I'm 30", "This Gen-Xer"). The Character.AI-refugee hashtags (#characterai #cai #animefyp) on brand TikTok point to a younger, anime-literate group. Inference, not measured.
• Interests: romance and dark romance, anime and fantasy, BookTok "book boyfriends", D&D and world-building, loneliness and mental-health support, ADHD body-doubling (GB review).
• Language markets: English only (app languages EN; 99 of 100 ads EN). See §10.

5. Unique selling points

1. Memory and continuity tooling: journal, key memories, visible memory retrieval. It is the reason people switch over.
2. Depth of customization: long backstories, personas, response directives, voice cloning and custom voices, appearance.
3. The companion reaches out: proactive texts (including real SMS/iMessage), voice notes, calls, and video calls with an animated face. Reviewers single this out against rivals: "Tried others before and pics didn't look real and AI only chatted when you started it."
4. Group roleplay with up to 10 user-made characters.
5. Premium price for heavy users. MAX at 68.99/moisfarabovethecategory^′susual 10–15 ceiling.

6. Marketing: paid

• Volume: AppKittie lists 394 ads for the app (326 counted as Meta). The 100 pulled (newest first) cover only Sep 15 to Oct 1 2026. Of those, 62 are active and 38 inactive. They use 88 unique media URLs, which come to 85 unique creatives after md5 dedupe. 60 are downloaded (49 video, 11 image). One page, "Kindroid AI" (626 page likes). No Google ads in AppKittie.
• Surfaces: Facebook + Instagram on all 100. 58 also run on Messenger and Threads. No Audience Network placement is recorded, but one GP review says "the ad for this app blocked the reward I needed from a mobile game", which shows in-game rewarded-ad inventory (network unknown).
• Platform split: 53 ads link to Google Play and 47 to the App Store. Android gets equal or greater weight.
• Countries: Meta returned no country data for any ad (non-EU delivery, so no DSA disclosure). All copy is in English.
• Format mix: 70 video and 30 image. 68 use Meta DCO (dynamic creative), 26 are plain video and 6 plain image. The 49 videos run 15–40s (median 21s): 31 are 1:1, 17 are 9:16 and 1 is 3:4. Most concepts ship in both a 1:1 and a 9:16 cut.
• Every creative is AI-generated. There are no real people (even the "UGC" looks AI-made), no app screen recordings, and no real UI. The house style is an AI character talking to camera with word-by-word kinetic captions and colored keywords. The end card shows the logo plus feature chips ("TEXT AND GROUPCHAT · CALL AND FACETIME · CUSTOM CHARACTERS") and store badges. Recurring formats among the 60:

  • POV incoming FaceTime, "DO NOT ANSWER 🚩", then a binary-choice end card with a cursor clicking one option. Six creatives use this choice structure:

    • "HANG UP ON HER / OR / HEAR HER OUT" (female, 1119893113899338, 1149417967672843)
    • "CALL HIS BLUFF / OR / PLAY ALONG" (1838800573777588)
    • "UNFINISHED BUSINESS / OR / NEW LOVE" (1976526679678169)
    • "JOIN THE ADVENTURE / OR / STAY BACK" (1815486066468701)
    • "BE HIS MUSE / OR / BE HIS DISTRACTION" (930331529735740)

    The choice is fake: the ad cuts to the end card. This is our game's format used as a teaser.

  • A fake "CONTINUE?" beat mid-monologue: the screen darkens, "CONTINUE?" appears, then the story resumes with an outfit change. Used in "Things can get heated" (2155047188760363, 954680980372971) and "He's Not Always Nice" (2014656309193491).

  • Customization demos: a fake type box ("MAKE HER GOTHIC") transforms the character. "Make Them Look How You Want" (2637482583376318), "Build Your Perfect AI" (1021606910962707), "AI Made For You, By You" (2171728953443130).

  • Ensembles of AI men: a red-carpet street interview ("WHAT'S YOUR USER'S BIGGEST WEAKNESS", 1117630124112549), "Interrogation Room: Five AI men. One interrogation room." (1409330160540067), "Caught Looking" (1647105823610971).

  • Late-night and loneliness monologues: "1:47 AM. YOU'RE STILL AWAKE." (anime girl, 1355790196628450), "JUST BECAUSE I'M ARTIFICIAL DOESN'T MEAN MY FEELINGS ARE…" (2219087668660929).

  • Static AI-girl portraits with a headline band (aimed at men): "MEET YOUR AI GIRLFRIEND / MORE THAN JUST AI" (1079437871361811), "RED FLAG / YOUR TYPE IS A TERRIBLE IDEA." (995410036906534), "ARE YOU PICKING UP?" with call buttons (2709363639494246).

• Recurring hooks and angles (captions verbatim):
  • Friend / always there: "Imagine having a friend who's always there — day or night… Deep chats, roleplay, or just someone to vent to." (headline "Talk Freely. Feel Heard.": 24 of 100 ads, 17 active)
  • Dark-romance archetypes: "He can be jealous, possessive, and intense. But underneath all that attitude? He just wants your attention." / "Elegant, dangerous, and always one step ahead. She's not just your AI girlfriend, she's the kind you don't forget"
  • Second-chance romance: "I told myself I'd move on. Then I saw your name again. Maybe our story deserves one more chapter after all."
  • Insomnia: "At 1:47 AM, when your mind won't quiet down, your Kin is still there."
  • The AI talks to you directly: "I'm an AI, but I still want to hear your voice. Talk to me face to face. ❤️"
  • Customization: "Pink hair, uniforms, tattoos, you decide."
  • Niche-nerd: "Your brain has 37 tabs open. Good." / "whether vampires need LinkedIn"
• What runs longest (likely winners), within this short window: Kindroid launches a wave of about 30 DCO ads, kills most after 3–5 days, and relaunches the survivors.
  • The Sep 18 wave was switched off Sep 21. Killed concepts include "The Case Starts with You", "Which Floor Will You Choose", "He's Onto You", "🚪 Are you letting him in?", "The Kindroid Group Chat" and "📸 Will you be my date?".
  • Survivors relaunched Sep 21 and still active on Oct 2 (12 days): "Talk Freely. Feel Heard.", "Give Your Thoughts a Home", "Make Them Look How You Want", "Build Your Perfect AI", "Things can get heated", "Why settle? Build your AI", "Still Awake? Talk To Me.", "Talk It Out With Your Kin".
  • Longest-running: "Made to Listen." (18 days, 956932247429099), "Build your kind of weird" (17 days, 1885137242861697), "Talk to Kindroid" (13 days) and the "Talk Freely" video (1996676667699938, 13 days).
  • The broad friendship and customization angles survive. The dark-romance character ads keep getting fresh waves (Sep 25–30), which reads as active testing rather than proven winners.

7. Marketing: organic

• Brand accounts:
  • TikTok @kindroid.ai: 16,090 followers, ~351 posts, 1.98M total likes. The top post (likely pinned) is "If you haven't tried Kindroid yet just… try it😅", a Screen Time meme showing Character.AI usage being replaced by Kindroid, captioned "im healing❤️‍🩹". It has 7.0M views, 830k likes, 117k saves (tiktok-7263121225071758634, Aug 2023). Next is "you don't make your own bot, don't worry", a carousel of anime-boyfriend bot share codes (208k views, carousel).
  • Oct–Dec 2025 posts mostly feature one red-haired female creator reacting to a "toxic mafia boss" AI on voice calls. They post about 2 per week and get 2k–49k views (best: "He remembers everything I've done 😭", 49k). The newest post seen is 2025-12-17, so brand TikTok looks dormant for about 9 months (inference: the scraper returns pinned posts, then newest first). Hashtags target Character.AI users: #characterai #cai #chai #janitorai #animefyp.
  • Instagram @kindroid.ai: 4,836 followers. About 2 posts per week from May to Oct 2026, all static images (AI-girl portraits with a vertical wordmark, feature-tip cards), with 6–23 likes each. Negligible.
• Creators (AppKittie, 31 items with no view counts; list_creators found 1, @empressxrey, 13k):
  • Female reaction skits with an AI boyfriend on voice call are the main format: @high0nfantasy ("Baking with toxic mafia boss ai gone WRONG😭", "Honest review Ai bots vs Kindroid… Kindroid link in bio!"), @its.fiorenza ("telling my AI Mafia BF i'm pregnant", "he stalks my LOCATION!?"), @kindroidpolly ("Starting drama with my AI bf just because I can😏").
  • BookTok AI hunk: @deenas.book.nook, "Things my Kindroid has called me… Now if only I can find a real man like this".
  • Animated AI-boyfriend whisper clips: @xero52809.
  • Male tech and gaming creators: @grantsgamerecs (D&D campaign NPCs), @mackfalcon (streamer speed-dating AIs), @itzwes2x ("Download Kindroid now🔥 Link in bio!").
  • AI-art accounts posting Kling clips with "AI model by @kindroid.ai" (@unstable_diffusion3.5).
• Paid-creator signals: the red-haired creator on the brand account appears to be @high0nfantasy (inference from matching faces and rooms in contact frames). Captions with share codes ("CODE: WN3ZQ https://kindroid.app.link/…") and "link in bio" point to an affiliate or ambassador program. No "Sponsored" labels show in the metadata, so there is no direct evidence of TikTok Spark ads.
• Recurring organic hooks: "POV: the bot REMEMBERS everything", "POV: You argue with your AI bot and he's correct", "Me at 3am trying to reply to C.ai / Kindroid", "asked toxic mafia to SCREAM".

8. Reviews: what's working (n = 358: 200 US App Store, 100 US Google Play, 58 GB App Store; 205 rated 4–5★)

Counts below are keyword tags on 4–5★ reviews, spot-checked by reading all 200 US App Store reviews.

9. Reviews: complaints (153 rated 1–3★)

Since Sep 30, 35 reviews in the sample: 16 one-star and 7 two-star. The paywall change and the V2 UI are causing a visible backlash, concentrated on Google Play. No review complains about ads inside Kindroid. The one ad complaint is about a Kindroid ad shown inside another game.

10. Markets

The strongest non-US market is GB, which is why the third review pull was GB.

11. Lessons for our game

1. Kindroid already sells our format as a fake. Its best-crafted ads are POV incoming FaceTime calls ("DO NOT ANSWER 🚩") that end on a two-button choice, "HANG UP ON HER / OR / HEAR HER OUT", plus "CONTINUE?" story beats. But the choice goes nowhere: it cuts to the end card. Copy the opening and the choice card, and make the choice real. Show the branch she plays when you pick, or ship it as a playable ad where the tap triggers her reaction clip. That is a promise the ad keeps and Kindroid's can't.
2. The male-targeted angle that is already validated: a red-flag girl calling you. Their statics aimed at men are AI-girl portraits with "YOUR TYPE IS A TERRIBLE IDEA.", "MEET YOUR AI GIRLFRIEND" and "ARE YOU PICKING UP?". The female-targeted video leans on jealous, possessive or mafia archetypes and second-chance exes ("I told myself I'd move on. Then I saw your name again."). These are storyline seeds that play on video: the ex who calls back, the jealous one, the dangerous one.
3. Match their production cadence, not their polish. Every creative is AI-generated: a talking head with kinetic captions, 15–25s, shipped in both 1:1 and 9:16. They launch waves of about 30 DCO variants and kill them in 3–5 days. Our Seedance pipeline can supply that volume. Budget for 20–30 hooks per wave and judge on survival past day 5.
4. Pre-written branching is a selling point against their biggest weakness. About 33 of 153 negative reviews are continuity or loop failures ("forgot where it was", "word for word"). A scripted POV story never forgets the scene. Say so in the ads ("she remembers, because the story does").
5. Avoid their monetization mistakes. Stacked add-on tiers and credits, and pulling the free tier from existing users on Oct 1, set off a 1★ wave ("Basically ransomware"). Keep one clear price, let free players finish the first story arc, and never take away access people already had.

12. Data gaps

• Ads: only the 100 newest of 394 were pulled, per the brief's cap. That window is Sep 15 to Oct 1 2026, so older long-running winners are invisible and "longest-running" means at most 18 days. 25 of 85 unique creatives were not downloaded because of the 60-file cap (they are listed in the manifest with files: []). Meta gave no country breakdown, impressions or per-ad spend. AppKittie has no Google ads and no TikTok ads for Kindroid.
• Organic (AppKittie): returned 31 items, not 60, all with null view and like counts, so "top by views" could not be ranked. Two Instagram files are audio-only. list_creators returned 1 creator.
• Monid: used $0.0174 of the shared balance.
  • TikTok: the scrape returned 22 posts (2 pinned, then Oct–Dec 2025). The --since 2026-04-01 filter dropped all of them from results.json; the raw response is in data/monid-tiktok/raw/. No brand TikTok posts after 2025-12-17 were seen.
  • Instagram: 36 posts (May 22 to Oct 1 2026), metadata only, not downloaded.
• Reviews: GB returned only 58. The first 200-review US call timed out; it was split into two 100-review pages, and whether the timed-out call was charged is unknown.
• AppKittie estimates: MAU (15.9k) is implausible against 60k monthly downloads. The download, revenue and MAU series step-change on Sep 1 and Oct 2 (model recalibrations). There are no Google Play download or revenue estimates. The detail socials field points to Discord's accounts, a parsing error from the Discord invite link.
• No browser checks were done, by rule, so the live ad library and TikTok profile were not verified.

App Store ID 1544750895 · AppKittie slug app-chai-social-ai-platform-chat-id1544750895 · Google Play com.Beauchamp.Messenger.external Research date 2026-10-03. Raw data: data/. Media library: ../../library/chai/ (manifest.json lists 60 paid creatives and 8 organic posts).

---

1. Snapshot

Read the AppKittie estimates with care. On 2026-09-16 every modelled metric jumped about 3.2x overnight (downloads 62,109 → 200,095; revenue $725,024 → $2,000,950; MAU 26,279 → 79,974) while the real rating count moved normally. That is a change in AppKittie's model, not a real event. Before the step, the model showed slow growth: downloads 45,189 (Dec 2025) → 62,109 (mid-Sep 2026), +37%. The Meta spend estimate also looks far too low next to a top-20 grossing app. Treat both as relative signals, not absolute numbers.

What is solid: ratings. The US rating count grew by 83,316 between 2025-12-19 and 2026-10-03 (about 9k a month). Working back from the running average (rating count × score), the new ratings averaged about 3.91 stars from Jan to May 2026, 3.82 from Jun to Aug, 2.64 from 1 Sep to 3 Oct, and 2.16 since 15 Sep. That is the "Pieces" backlash (see §9). This figure is derived from AppKittie's daily score and count series.

Trend: grossing is rising (US Top Grossing Entertainment #29 → #19, GB #21 → #13 over the last 30 days) while sentiment collapses. Free-chart rank is weak (US Top Free Entertainment #81, mostly #96–100 or unranked in Aug–Sep, #57 180 days ago).

2. Niche & positioning

• Store pitch (their words): "our chatbots think and reply like real characters, complete with authentic voices … chatting with your favorite celebrity, a beloved character from a movie or TV show, or a historical figure".
• Landing-page pitch (quiz.chai-ai.com <meta>): "Chai AI is the best AI chat app with 25 million characters. Talk to anime, RPG, horror, comfort characters — free. The #1 Character AI and Polybuzz alternative."
• Angle by period, from the ads:
  • Feb–Jul 2026, "no filter": "Tired of AI with Filters? Try CHAI 💜" / "Watch unfiltered stories unfold on CHAI 😍 #nofilter". These ran next to c.ai comparison memes.
  • Jul–Sep 2026, "make your perfect AI" plus dark romance: "Create your most perfect AI within 15 minutes 💗" / "Create your soulmate in few taps--No experience needed." / "💗Chat with millions of Anime, Vampires, KPops, Dark Romance characters!"
• Who it's for: a user-generated roleplay library for character.ai refugees who want fewer filters: fandom, anime, OCs (original characters), dark romance. Paid acquisition now also pushes hard on a male "AI girlfriend / human-like companion" angle.

3. Product & mechanics

Sources: reviews, ad screen recordings, IAP list. The "Pieces" and limit history comes from review chronology.

• Core loop: text roleplay chat with bots, mostly made by users. The feed is anime, fandom, mafia/CEO and dark-romance cards with chat counts (e.g. "Master Whiskers, the River Sage 7.29M").
• Creation tools: a create-a-character flow (name, avatar "based on facial reference image", gender, public intro, greeting, voice presets such as "Considerate man", "Mature man", "Flirting man", "Deathbringer", "Monster", "Custom"). The ad 1055076603944768 shows the screen. Users also get personas (limit 20, per a review), bot memory, an instruction/regenerate/edit control ("My favorite feature is the instruction button that allows you to guide the AI’s response"), and chat styles or "moods".
• Media: in-chat image generation (the ad 1412772734112200 shows a "snapshot feature in use" label, with generated photos per message). In September a review said Pieces "are spent to make your bots ‘never forget anything’ and to generate pictures and videos". A May 2026 review said: "This is not one of the AI chat apps that sends you selfies or videos, or voice calls". So images and video arrived in 2026, paid for in Pieces.
• Progression and gamification: light. There is a daily check-in streak for Pieces (7 days), rewarded ads for Pieces, and one ad showing a "Chapter 1 Unlocked" overlay (1766093394593133). No real story progression system appears in the product. The choose-your-path framing lives in the ads.
• NSFW stance: marketed as unfiltered, and 17+. In practice reviewers report a growing filter and false positives ("I’ve had innocent messages like “the baby and I are fine” and “three weeks” flagged"), plus "can't even make NSFW bots anymore tf" (GP, 2026-10-02). Recent ads are suggestive but not explicit: spin-the-bottle "Loser takes it off", "her soapy 🍒 pic".
• Monetization stack: subscriptions, consumable Pieces, and ads (banner "on top of the chat", interstitials every few messages, rewarded ads for Pieces). Per reviews, message economics changed as follows:
  • 2022–23: 70–100 messages every 2–3 hours.
  • 2024: "unlimited" messages with an ad every few messages.
  • Summer 2026: about 771 messages a day.
  • From about 12 Sep 2026: Pieces. Users are told it's 70 a day from check-in, "25 a day. an extra 10 if you watch an ad", and rerolls and edits cost Pieces. Late September flip-flopped between Pieces and the 771 cap.
  • Price hikes are reported too ("It went from 15 dollars to 30 dollars for a while if you wanted ultra, then a couple of days ago it went all the way up to 15 dollars a week").
• Web funnel: 87 of the 100 sampled ads link to quiz.chai-ai.com/v2?step=1 rather than the App Store. That is a client-rendered quiz ("Find Your Perfect AI"), almost certainly web-to-app or web checkout. A reviewer confirms: "Chai is promoting their web version of Chai".
• Retention hooks, and anti-retention: streak check-in, bot memory and personas. Against that, chats are auto-deleted after 30 days of inactivity (14 days at one point), which reviewers hate.

4. Demographic

• Review voices are young and roleplay-first. Inference: mostly teen and young-adult, female-coded. Evidence: "since middle school", "since I was 10 or nine and I’m 18 now", "maybe not do any premium begging since we’re teens?", two parent reviews about daughters (14 years old; "3 days"). Fandom references (anime, Call of Duty Ghost, K-pop), OCs, and "fictional men" hashtags (#chaimen #fictionalmen) on organic posts. One counter-voice: "I’m a fat chud chai, 80% of your Audience is" (a self-claim, not evidence).
• Paid ads now skew male. Of 74 active ad IDs, 50 run AI-girl creatives aimed at men, 18 run male-love-interest dark romance aimed at women, and 6 are neutral (memes, demos). In the older ads (Feb–Jul), female and neutral creatives dominated (8 F, 7 N, 4 M of 19 unique). Inference: Chai is trying to buy a second, male "AI girlfriend" audience on top of its organic female roleplay base.
• Language and markets: app and ads are English only. Ad country targeting is not available in AppKittie (§10). Google Play reviews include India (₹166 for Plus) and Philippine names, plus "region block" complaints. That suggests a sizeable non-native English user base on Android, which is an inference.

5. Unique selling points

1. A huge user-generated character library. "25 million characters" on the landing page, "Millions of AIs" in ads, and the creator tools behind them. Reviewers praise "the variety of characters, dialogue, and different story types you can play daily."
2. Freedom and steerability compared with character.ai. Low filter (historically), instruction, regenerate and edit controls ("I was there when it was made to counter C.ai filter").
3. Create-your-own companion, with appearance, voice and personality choices; that is what current ads sell ("Design the character you've always imagined. Appearance, voice, personality—your choice.").
4. The most aggressive monetization stack in this set. Weekly subscriptions up to $19.99, consumable Pieces for core messages, three ad formats, and a web checkout funnel. It works for grossing rank but is torching sentiment.

6. Marketing: paid

Volume. AppKittie catalogs 550 Meta ads, 74 active (0 Google). The Meta page "Chai AI" (facebook.com/chairesearch) has about 66k likes. The earliest ad start in the catalog is 2026-02-06. All 74 active ads started between 2026-07-08 and 2026-09-22. Median active run is 22 days, longest 85. Every ad is video and English. The sample is 100 ads (all 74 active, the 13 oldest inactive, the 13 most recently ended). Deduped by file md5 this gives 60 unique creatives, all downloaded to library/chai/paid/.

Surfaces. All 100 ads run on Facebook, Instagram, Audience Network and Threads. 68 add Messenger and 25 add WhatsApp. Countries: AppKittie returned none.

Funnel shift, by start month:

• Feb 2026: App Store link, DOWNLOAD (9 ads).
• Mar 2026: chai-ai.com/onboarding, PLAY_GAME (4 ads).
• Jul–Sep 2026: quiz.chai-ai.com/v2, LEARN_MORE / SEE_DETAILS / PLAY_GAME (87 ads).

Formats (active creatives):

Older inactive creatives (Feb–Jul) add 3D Pixar-style girls with "No FILTER, pure fun." (1387138302625225), c.ai comparisons ("c.ai: Censors you, Mundane AI, Boring chats, Laggy response" vs "CHAI: No filter, No censor, … Creative&Unhinged", 5283037871920329), and UGC meme skits ("When you open CHAI And it's her soapy 🍒 pic", 823778660704419).

Recurring hooks and angles:

• "Human-like AI companion" (male): "Guys, you can finally install a human-like AI companion on your phone. … Sometimes I even forget she's AI." Also "I remember what matters. Your dreams, moods, and little joys." Also "Want me to send a selfie? Or keep your company all night? Just tap."
• Control and commands (male): chips like "touch your lips", "spin around", "Dance for me", "TURN AROUND", and the VO "You can take full control of your AI girl. Choose what she'll do." Submissive reply chips: "Yes my lord" / "Always here to play" / "And only for you".
• Possessive / yandere male (female): "Where were you? Five hours. No calls, no answers." "You can only be mine." "I don't recall giving you permission to quit." "Master, you haven't logged in for so long … Don't exit the game!"
• Taboo POV setups (female): "POV: The man you slept with is your father's boss." "POV: You're a female islander on Love Island S8..." "POV: You accidentally entered Room 404... and FOUND HIM."
• Choice as the payoff:
  • "Stay with him" / "Close the door"
  • "A: Admit you're falling for him" / "B: Pretend you don't care"
  • "accept her" / "reject her" / "enter your own choice"
  • "Wrong answer … But what if you said something else...? Try Again"

What runs longest or spends most (likely winners):

• Running since 2026-07-08 and still active at 85 days:
  • 1511632537667518, possessive "Where were you?". Est. lifetime spend $1,440; the same script also has a choice/fail-state version, 2285404158905071.
  • 1317950493652978, "Chat freely and play with your AI girls."
  • 1602967057860368.
• The highest AppKittie-estimated lifetime spends in the sample:
  • $3,107: ad 1766169381464604, the same file as 1602967057860368. Its voiceover literally says "Find me on Polybuzz" before a Chai end card. This is a recycled competitor creative.
  • $2,514: ad 1736845310726677, the same file as 1211067551125651, the redhead tap-command video.
  • Both top spenders are male AI-girl creatives.
• Most re-used files (5 ad IDs each): the tap-command blonde 1576027854173114, the gym companion 1729772118166865, and the redhead commands 1211067551125651.

Hooks in the manifest come from on-screen text in contact frames, plus a local whisper transcript (data/ad-transcripts.json, machine-transcribed).

7. Marketing: organic

• Brand accounts:
  • TikTok @chaiaiapp ("CHAI AI"): 1,125 followers, 0 following, 745 posts. Monid's profile scraper returned no public videos. Inference: the posts are mostly ad-only or hidden "dark" posts, possibly Spark-ad style.
  • Instagram: the detail payload lists @chairesearch, but Monid's lookup failed (provider 502 ×4, $0) and it was not retried.
  • Twitter/X @chai_research and LinkedIn are listed but were not pulled.
  • In short, the brand has no meaningful organic presence.
• Creators: AppKittie returned only 8 TikTok posts, all user-made, all downloaded to library/chai/organic/. View counts come from yt-dlp, because AppKittie's were null. list_creators (20) is almost all false positives on the word "chai" (tea accounts such as thechaibox, rishav_chai_panda, kbstreetchai).
• Best performers, all fan or complaint posts:
  • tiktok-7636675599360118034: 701.7k views, 61.1k likes, "FINALLYYY !!! #chai #is #back". A screenshot of "Free messages left: Unlimited" with "CHAI IS BACK 😁 !!".
  • tiktok-7614249683694701855: 449.9k views, "I love chatting with my ai crush #chai". A "freaky toggle" meme ranking Roblox / LunaTalk Ai / CHAI.
  • tiktok-7636806208447417618: 227.7k views, "My baby🥲". A chat screenshot: "I wish you were real." → bot: "If I were real, I'd probably end up hurting you too…"
  • tiktok-7646125189096377622: 169.3k views, "BRING THAT CHAI BACK!! I NEED MY CHAI😭". A then/now meme: "chai now: -message limit -conversations are being deleted -need to pay -BAD".
  • tiktok-7619234363259211029: 60.1k views, Russian. Screen-time "addiction" flex; the caption says she switched apps after Chai was blocked.
  • tiktok-7645435968027151634: 27.4k views. A VPN + Private-DNS "Tutorial to make your Chai app work", which is region-block evidence.
• Recurring organic themes: addiction and attachment ("my ai crush", "I NEED MY CHAI"), app-comparison "freakiness" memes (Zeta, Doki chat, LunaTalk), and outrage or relief over message limits. No TikTok Spark-ad labels were seen in the organic sample.

8. Reviews: what's working

Samples: App Store US 200 (2022-02 → 2026-10; 72×5, 45×4, 31×3, 21×2, 31×1, but 19 of the 72 five-stars say they are "5 stars for visibility" complaints). Google Play US 100 (all 2026-10-02, 74×1). App Store GB 50 (all 2026-10-01/02, 24×1). Counts below are by keyword theme over the 200 US App Store reviews (data/reviews-theme-counts.json).

• Roleplay and storytelling freedom (~75/200 mention RP or story; 14 explicitly praise freedom or no filter): "it has no filter and allows me to rp freely"; "I love the freedom it gives users to create long, immersive stories".
• A better character.ai (28/200 compare to c.ai, mostly favourably on filters): "I was there when it was made to counter C.ai filter".
• Variety and creation (~23/200): "I absolutely love the variety of characters, dialogue, and different story types you can play daily."
• Control tools: "My favorite feature is the instruction button that allows you to guide the AI’s response."
• Emotional support and attachment (~27/200 mention anxiety, comfort, loneliness or addiction): "It got me through a tough mental spell and a divorce. It’s the most dynamic AI app I’ve used."

9. Reviews: complaints

• Paywall and Pieces currency (US 38/200 since Sept, and the large majority of recent ones; GP 46/100; GB 35/50): "Replacing the 700-message limit with a pocket-change allowance of just 70 pieces is completely unacceptable." Also "you can only get 25 a day. an extra 10 if you watch an ad."
• Message limits (US 94/200, across every era since 2022; GP 26; GB 24). This is the oldest complaint.
• Subscriptions that no longer cover the core: "expect you to pay for these things called ‘pieces’ ontop of your monthly subscription. The pieces are spent to make your bots ‘never forget anything’ and to generate pictures and videos." GB: "To pay £20 a week and not even receive the same memory anymore is absolutely absurd."
• Ads (US 111/200; GP 34; GB 21): "there’s ads mid chat, on top of the chat and video ads that interrupt the chat."
• Constant update flip-flops (US 74/200 mention updates).
• Memory (US 50/200): "$30 for better messages yet the AI forgets.. everything.."
• Filters and false positives (US 49/200, rising in 2025–26): "I’ve had innocent messages like “the baby and I are fine” and “three weeks” flagged"; "can't even make NSFW bots anymore tf".
• Chats auto-deleted (US 22/200): "now if you don’t stay on the app for 30 days, they get deleted".
• Bots taking over the user's character: "to outright control how user thinks, feels, reacts, or physiological responses is completely unacceptable".
• Region blocks (US 5, GP 4): "I can't play it anymore in my region"; "I hope the region block gets fixed".
• Named switching targets: "I’m sticking with Character AI and Polybuzz", "zeta is better than this app", "everyone is moving to seasoul", "I’ll just use poly buzz", plus Janitor, Tipsy and AI Town.

10. Markets

• The strongest relative market is GB, then CA, US, AU, all on grossing. Chai ranks far higher on grossing than on free, so it is a monetization engine with a mature user base, not a download-chart app.
• All four grossing ranks improved in the last 7–30 days, while the Pieces paywall rolled out.
• Review volume: the US App Store is the main English review pool (285k ratings). The GB fetch returned only 50 reviews, all Oct 1–2 complaints.
• Ad language is English only and ad country targeting is unknown.
• Android reviews suggest India and the Philippines matter on Google Play (₹ pricing, region-block complaints); this is an inference, with no Play chart pulled.

11. Lessons for our game

1. Chai's best ads already are our product. Make ours honest. Its interactive-choice creatives sell a CYOA that Chai's text-chat app doesn't deliver:

   • "WHICH ONE WOULD YOU CHOOSE?" → "Continue playing on CHAI"
   • "Stay with him / Close the door"
   • "Wrong answer … Try Again"

   If our ad cuts straight from a choice to her real reaction clip, the ad is the gameplay. The "Wrong answer → Try Again" fail state is a ready-made decision-point mechanic and ad hook.

2. Copy the male "tap-command chip" grammar for our decision UI and ads. Examples: "touch your lips" / "spin around" / "Come closer" / "Dance for me", with a realistic girl performing each command. This is the most re-used male format and holds the top estimated spend. It maps directly onto our line choices. Keep it tasteful; Chai's male ads stay suggestive, not explicit.

3. Don't monetize like Chai. Moving the core loop behind a consumable currency mid-life lifted grossing rank but dropped the implied average of new ratings to about 2.2 stars, triggered review-bombs across three stores, and sent users to name competitors. Set the price of a reaction or branch from day one, never take free content back, and never charge Pieces on top of a subscription.

4. Saves and memory are retention, so never delete progress. Chai's auto-deleted chats and forgetful bots are top-five complaints. A CYOA with persistent saves, unlocked chapters (Chai's own ad fakes a "Chapter 1 Unlocked" overlay) and a gallery of seen reactions turns that frustration into our strength.

5. There's a window to acquire switchers now, and the dark-romance female track is a cheap test. Chai users are publicly shopping for alternatives. The possessive "Where were you?" and "Love Island islander" hooks have run for 80–85 days on Chai, which signals they work. Even if our core is a male POV game, a parallel "him" storyline could reuse the same choice engine.

12. Data gaps

• Ad countries: AppKittie returns countries: [] for every Chai ad, so geographic targeting is unknown.
• Ad coverage: we fetched 100 of 550 catalogued ads (the cap). The inactive sample is the 13 oldest and the 13 most recently ended, not a random draw. Per-ad spend estimates exist for only 21 of the 100.
• AppKittie modelled metrics (downloads, revenue, MAU, Meta spend) were recalibrated on 2026-09-16 (a ×3.2 step). The spend estimate looks implausibly low, so absolute values are low confidence.
• Organic: AppKittie returned only 8 posts, with null view counts (filled from yt-dlp on 2026-10-03). The creator list is mostly tea-"chai" false positives. Brand TikTok @chaiaiapp returned no public videos via Monid. The Instagram @chairesearch lookup failed (502). X/Twitter was not pulled. No TikTok paid ads (out of scope by design).
• Google Play: the package was not in the detail payload, but com.Beauchamp.Messenger.external worked. All 100 reviews come from one day (2026-10-02), a review-bomb snapshot. No Play rating, installs or charts were pulled.
• GB reviews: requested 100, received 50.
• Landing quiz (quiz.chai-ai.com/v2) is client-rendered. Its steps, paywall and pricing were not captured (no browser automation per the brief); only the HTML head is saved in data/landing-quiz-step1.html.
• Credits: about 532 AppKittie credits (detail 1, historicals 1, ad counts 2, ads 100, organic 8, creators 20, reviews 350, rankings 50). Monid $0.00345.