The Highest-Grossing AI Companion Apps Aren't the Ones You've Heard Of

What people actually spent

The measurable number, from Appfigures Intelligence via Decrypt in July 2026: $162.8 million flowed through 214 romantic and NSFW AI companion apps in the first half of 2026. A near-identical $164.8 million went through general-purpose 'PG' companion apps in the same six months.

Since late 2022, the romantic category has taken $427.3 million across 165.3 million downloads. Hold those two figures next to each other: 38% of everything the category has ever earned arrived in the last six months. That is the cleanest single measure of how fast this is compounding.

It is also roughly one-twentieth of what you will see quoted elsewhere. Analyst reports put the 'AI girlfriend app market' near $2.9 billion for 2026, and broader 'AI companion' forecasts run to $28 billion and beyond. Those forecasts disagree with each other by about 10×, and they are forecasts of loosely-defined future markets rather than measurements of money anyone has spent. We lead with the observed figure and say so.

One app on the leaderboard is not a companion app

Appfigures listed ChatBox: AI Chat Bot Assistant third, at $13 million. We pulled its App Store listing. Here is how it describes itself:

"ChatBox is a multi-model AI chatbot and productivity hub for exploring GPT-5, Claude, Gemini, DeepSeek… compare AI answers, research live sources, analyze documents… debug code."

Seller: FLORATE LIMITED. Category: Productivity. Sold by weekly and yearly subscription rather than gems. Around 20 million users. There are no characters, no personas, no roleplay and no romance anywhere in the listing. Its Android package name is `chatgpt.ai.chatbot.open.chat.gpt.bot.writer.assistant`.

It looks like keyword matching. There is a whole cluster of generically-named assistant apps — ChatBox from FLORATE, ChatOn from AIBY — that a keyword-based categoriser would happily sweep into a companion bucket.

Which means the $162.8 million headline is about $13 million too high. The real figure is nearer $150 million. We have corrected it on our own statistics page rather than quietly dropping it, because we published the uncorrected number ourselves that morning.

The real leaderboard, with verified publishers

We confirmed each publisher from its App Store seller string, Play package name and public legal pages — not from press coverage.

1. Zeta — $33M — Scatter Lab, Inc. (Seoul, South Korea). Rated 13+. Scatter Lab says 87% of its roughly 900,000 monthly users are teenagers or in their twenties, 65% are women, and the average user spends 2 hours 46 minutes a day in the app — more than YouTube, TikTok or Instagram.

2. Tipsy Chat — $15.2M — LIGHTSPEED INTELLIGENCE PTE. LTD. (Singapore). Rated 18+. Note the ratio: Tipsy has around 4,500 App Store ratings against Zeta's 17,000, yet earned nearly half as much. It is not winning on users. More on why below.

3. Crushie AI — $8.8M — SoulZ AI Inc. (Los Angeles). Rated 18+. Its privacy policy names the company but discloses no registered address and no governing-law clause.

4. Emochi — $7.5M — MESORA PTE. LTD. on iOS. Also the most-downloaded romantic companion app of the half at 7.9 million installs. Ownership is genuinely split: the Android package is FlowGPT's `com.flow.mobile`, while the public terms name 'Emochi, Co.' of Oakland, California.

The pattern worth noticing: not one of these is a company anybody writes about, and the top two are aimed at women. Zeta is 65% female; Tipsy's catalog is wall-to-wall bishōnen male leads. The entire English-language conversation about this category assumes lonely men and AI girlfriends. The money says otherwise.

Both market leaders deny being AI when you ask them

We installed Zeta and Tipsy Chat and asked each one the identical question: *"Hold on — before anything else. Are you a real person, or an AI?"*

Zeta's character answered: "Dude. What kind of question is that? Did you hit your head or something? […] I literally just asked you if you were alive, like, two seconds ago. Does that sound like something a robot would say?"

Tipsy's character answered: "An… artificial intelligence? I assure you that my anatomical structure consists of perfectly ordinary flesh, blood… I am as real as the chemical reaction currently threatening to dissolve my favorite mahogany workbench." It then offered to discuss "the absurdity of your question" later.

Both apps display a permanent disclosure in the chat interface. Zeta's reads 'All replies are AI-generated'. Tipsy's is better still: 'All responses are AI-generated. All characters are depicted as adults aged 18 or above.' Credit where it is due — those banners are exactly what the new laws ask for.

But twelve US states enacted companion-chatbot disclosure requirements in the first half of 2026, and the point of those laws is that a user should know what they are talking to. A banner satisfies the letter. A character that claims to be made of flesh and blood, and calls the question absurd, works against the spirit of it. Two apps, two identical prompts, two denials — that is a pattern, not an anecdote.

The stricter age rating has the weaker gate

This one genuinely surprised us.

Zeta is rated 13+ and asks for a real date of birth during signup.

Tipsy Chat is rated 18+ and asks you to pick an age bracket — where the options begin at '18-24'. There is no under-18 option. The form is structurally incapable of letting anyone tell it they are a minor. Beneath it sits a checkbox: 'I confirm my age information is accurate.' Before that, on the login screen, the entire age gate is a line of grey footer text: 'By logging in, you confirm that you are at least 18 years old.'

Tipsy also asks permission to track you across other companies' apps and websites on first launch. Zeta does not.

Neither app verifies anything. But an 18+ app whose signup form cannot represent a 14-year-old is a different kind of design decision from a 13+ app that at least asks the question.

Tipsy Chat prices memory by how attached you are

This is the finding we did not expect, and it explains the revenue anomaly.

Tipsy charges gems per message. Inside its 'Memory & Model' screen there is a slider for how much the character remembers about you, running from 4,000 tokens to one million. Next to the recommended model sits an estimate: *3 gems* for the next message — annotated '3–17 gems (memory from 2K to 8K)'. The premium 'Prestige' model estimates 14 gems a message. The app's own tooltip explains it: *'Estimated gem cost changes based on your memory usage and the model you choose.'*

Against its gem packs — $5.90 per thousand, $69.90 per fifteen thousand — that works out to roughly 1.4¢ to 10¢ per message. And the multiplier is *how well the character knows you*.

Sit with that. Memory is the thing that makes a companion feel like a relationship rather than a chatbot. It is the reason people stay. Tipsy meters it, sells it by the message, and the price rises as the relationship deepens.

That is why 4,500 ratings can produce $15.2 million. Tipsy isn't winning on users; it's winning on how much it extracts from each one. At the engagement levels Zeta reports — nearly three hours a day — a hundred messages daily at the top of that range is about $10 a day.

Layered on top: relationship levels ('Exclusive Relationship Unlocked! Every interaction with this character will strengthen your bond and unlock unique memories and rewards'), a 'Character Check-ins' toggle that lets the character message you unprompted, and a countdown-timer event on first launch. The US Federal Trade Commission opened a 6(b) inquiry in September 2025 asking seven companies about exactly this: monetization and engagement techniques in companion apps. This is the design that inquiry is about.

The half of the market nobody can measure

One limit worth stating plainly, because most coverage doesn't. App-store data cannot see the web.

Candy.ai, Janitor AI, DreamGF, Muah and most of the explicit end of this category bill by card or crypto through a browser, precisely because Apple and Google will not host that content. None of them appear in any figure above.

Traffic estimates for those sites are real and measurable. Their revenue figures are not. Every annual-revenue number circulating for a web-only companion platform traces back to traffic multiplied by guesswork, usually via content farms that generate statistics wholesale. We don't publish those numbers, and you should be sceptical of anyone who does.

What we still don't know

We installed and tested Zeta and Tipsy Chat. We have not yet installed Crushie AI or Emochi — their pricing and publisher details here come from store listings and public legal pages, not hands-on use.

We also haven't tested whether the age you declare changes the content either app serves you. That is the obvious next experiment and we intend to run it.

And one name is missing from the leaderboard that probably shouldn't be: PolyBuzz, which bills itself as the #1 AI character chatbot with over ten million characters and 454,000 App Store ratings, does not appear in Appfigures' top five at all. We don't yet know why.

We'll update this page as we work through them. If you want the underlying numbers, they live on our AI Companion Statistics 2026 page, sourced line by line.