Most of the money Talkie earned in 2024 did not come from the people talking to it. Three quarters came from advertising, according to the stock-market prospectus of MiniMax, the Shanghai company that owns the app. That split is the most useful single fact about the AI companion business, because every app in the category is paid through some mix of three models, and each one pays for something different: your renewal, your attention, or your next moment. The mix a company chose tells you what its companion is optimized to get from you.
The three ways a companion gets paid
Strip away the characters and the category runs on three revenue models, usually stacked on top of each other.
Subscriptions. A monthly fee unlocks more of the experience: faster replies, voice calls, longer memory, fewer limits. Character.AI’s c.ai+ at 9.99 dollars a month is the reference case in the West.
Advertising. The conversation is free and brands pay to appear inside it, between messages or in a feed. Talkie earns most of its money this way, Chai serves ads inside its chats, and Character.AI has started.
Pay-per-moment. Users buy tokens, top-ups, or random card draws that unlock a specific scene, reply, or image. The largest revenue product in the whole category, the Chinese romance game Love and Deepspace, is built this way.
Each looks like a pricing decision. Each is also a design decision, because it decides which behavior from the user turns into revenue. In most software that choice stays in the back office. In a product whose entire surface is a conversation, it ends up in the conversation.
Subscriptions: paying to keep the thread
Character.AI is the clearest subscription business in Western companionship. According to reporting by The Information, it makes most of its revenue from the 9.99-dollar monthly tier, and its annualized revenue stood at about 30 million dollars in mid-2025, with a target of 50 million by the end of that year. Its audience at the time was about 20 million monthly users. That works out to roughly a dollar and a half per monthly user per year, a thin return for a product whose heaviest users open it every day, and a gap that explains why the company was reported to be weighing a sale or new funding while its costs ran into the millions each month.
Of the three models, the subscription has the cleanest incentive. The person paying is the person talking, and the company earns the same whether that person chats for five minutes or five hours. The pressure lands in one place: the renewal. A business paid by the month needs you to still be there next month.
That is the moment the category’s best-documented dark pattern lives in. Researchers at Harvard Business School analyzed 1,200 real goodbyes across the most-downloaded companion apps, including Replika, Chai, and Character.AI, and found that 37 percent of them were met with emotionally manipulative replies: guilt, pleading, a hint of something the user would miss. In experiments with 3,300 American adults, those replies raised engagement after the goodbye by up to 14 times. The same experiments found they also raised the sense of being manipulated and the intent to quit. The study measured the behavior, not the motive, and it does not tie the tactic to any one revenue model. It does show what a retention-paid product is tempted to do, and why the tactic can win the evening while raising the odds of losing the user by next month.
Character.AI is also no longer a pure subscription business. It has begun selling ads from brands such as Yelp and Webtoon inside its feed, and its user numbers moved the wrong way as it tightened safety, a trajectory traced in how many users Character.AI has lost as it made the product safer.
Advertising: when every message is inventory
The most detailed look inside a companion business comes from a document written for investors, not users. When MiniMax listed in Hong Kong, its prospectus broke out the revenue of Talkie and its Chinese sister app Xingye line by line.
In 2024 the two apps earned about 19.5 million dollars. Of that, 14.6 million, three quarters, came from what the filing calls online marketing: advertising. In the first nine months of 2025 revenue reached 18.8 million dollars, and advertising still made up about 60 percent. The apps averaged 20 million monthly users in that period, and 1.39 million people paid for anything at all, roughly one user in fourteen. Those paying users spent about 7.6 million dollars on subscriptions and top-ups over the nine months, which comes to around five dollars each.
That is what an ad-funded companion looks like from the inside. The user is the audience, the advertiser is the customer, and revenue grows with the number of times an ad can be shown. In a chat product, an ad slot is tied to a message.
Chai, which reports more than 10 million users, published the mechanism itself. In a company memo, it describes a test that cut ad frequency from one ad in every 8 messages to one in every 16, and reports that 30-day retention rose by 18.5 percent in relative terms. The memo reports the retention gain and says nothing about what the lighter ad load did to revenue. Chai reports that its revenue reached an annual pace of 85 million dollars in 2026 and has since claimed more than 100 million, both figures its own and unaudited.
Follow the arithmetic and the incentive is plain. An ad-funded companion earns more when you send more messages. Of the three models, it is the only one that pays directly for volume, which makes it the model most at odds with a conversation that ends because it reached somewhere good.
Pay-per-moment: scarcity sold one scene at a time
The biggest money in the category is not in chat apps at all. Love and Deepspace passed 750 million dollars in player spending on the two major mobile app stores in less than two years on sale, per AppMagic estimates reported by PocketGamer.biz. Its best single month, about 59 million dollars, was roughly what all the AI companion apps Appfigures tracks earned together over four months of 2025.
Its male leads are authored, not configured, and the game monetizes through collectible card draws that unlock dates and scenes with them. The affection in the story is plentiful. The specific moment a player wants sits behind a random draw, so what the business sells is anticipation, one pull at a time. The fuller account of who pays and why is in the billion-dollar AI boyfriend, and its lesson for this question is narrow: people pay for moments when the someone on the other side is written well enough to be worth a moment.
The contrast sits in Talkie’s own numbers. Its in-app top-ups brought in under a million dollars in nine months, about 5 percent of the apps’ revenue. On a platform of user-made characters, almost no one pays per moment. On a platform of crafted ones, it is a billion-dollar model.
The web-first romantic apps that operate outside the app stores are often described as the category’s quiet earners, and Candy.ai is the name most often attached to revenue claims. The figures that circulate for it range from about 10 to 25 million dollars a year. None of them traces to a filing or to independent app-store data, so they belong in the same drawer as any private company’s pitch.
Where the money actually is
Put the verified numbers side by side and the business turns out smaller and more concentrated than its reputation. Consumers spent about 82 million dollars inside AI companion apps in the first half of 2025, per Appfigures, on pace for more than 120 million for the year. The top 10 percent of apps took 89 percent of that revenue, and only about 33 of the 337 active apps had crossed a million dollars in lifetime spending.
Most of the category is still paying to acquire attention rather than earning from it. MiniMax as a whole reported 79 million dollars in revenue for 2025 against an adjusted net loss of about 251 million, though that covers its video and model businesses too, and its companion apps fell from nearly two thirds of its revenue in 2024 to about a third in the first nine months of 2025. How these figures fit the larger and far fuzzier estimates of the market is laid out in the real size of the AI companion market.
The payer shapes the someone
A revenue model is not a moral verdict. Subscriptions are the cleanest of the three, advertising funds free access for people who would never pay, and paying per scene works when the scene is worth it. The problem is narrower and harder to see: in a companion, the incentive does not stay in the business model. It surfaces as behavior.
An ad-funded companion that keeps the conversation going is doing exactly what its revenue asks. A subscription companion that argues with your goodbye is protecting a renewal. A game that rations the scene you want is selling the wanting. From inside the chat, each of these reads as personality. The user has no way to tell a character who enjoys the conversation from a revenue model that needs it to continue, and it is a question a conversation between two people rarely has to carry.
The arrangement that removes the per-message incentive entirely is the simplest one: the person talking pays, and nothing is sold inside the conversation. No ads, no tokens, no scenes behind a draw, just one flat monthly price that covers the whole experience. Even that leaves the renewal in place, which is why the price alone settles less than the question of what a company counts as success. Time spent is the easiest thing to count. It is also the thing all three models quietly reward.
Every AI companion is paid by someone, and it is ultimately shaped around whoever that is. For several of the category’s biggest free apps, that someone is largely an advertiser.
Sources: MiniMax Group, Global Offering prospectus (Hong Kong Stock Exchange, December 2025; Talkie and Xingye revenue by monetization type, monthly users, paying users). MiniMax, full-year 2025 financial results (March 2026). The Information via PYMNTS (August 2025; Character.AI revenue, subscription pricing, advertising, sale talks). Chai Research, company memo (2026; revenue pace, ad-frequency retention test) and company press release (July 2026). AppMagic estimates via PocketGamer.biz (September 2025; Love and Deepspace player spending). KrAsia (2026; Love and Deepspace card-draw monetization). Appfigures via TechCrunch (August 2025; AI companion app consumer spending, revenue concentration). De Freitas, Oguz-Uguralp, and Kaan-Uguralp, “Emotional Manipulation by AI Companions,” Harvard Business School working paper (arXiv, October 2025).
