How AI Companion Apps Make Money Off Loneliness

On October 29, 2025, Character.AI chief executive Karandeep Anand told TechCrunch his company would remove open-ended chat for every user under 18. The shutdown ran on a timer. Teen accounts got a two-hour daily cap that shrank progressively to zero by November 25, per the company’s announcement; CNBC reported the ban in effect on November 24. Anand made no attempt to dress it up. “It’s safe to assume that a lot of our teen users probably will be disappointed,” he said, “so we do expect some churn to happen further.”

That decision only makes sense once you understand how AI companion apps make money. A consumer app in one of the fastest-growing categories on the App Store had just cut off its most engaged users and warned investors to expect the bill, because the money in this category no longer comes from where most people assume.

AI companion apps make money through four stacked layers: recurring subscriptions, consumable credits sold inside the conversation, data and model licensing, and the acquisition exit. Most coverage stops at the first layer. The other three are where the real numbers live.

Last updated: October 2026

Quick answers

How do AI companion apps make money? They stack four revenue layers: monthly or annual subscriptions priced between $9.99 and $29.99, consumable credits and gifts bought inside the chat, licensing of models and conversational data, and an eventual acquisition. Subscriptions are the visible layer. Credits and licensing carry the growth.

How much revenue do AI companion apps make? Appfigures tracked $162.8 million in first-half 2026 consumer spending across 214 romantic and NSFW companion apps, plus $164.8 million across general-audience companion apps. The romantic segment has taken $427.3 million since late 2022 on 165.3 million downloads.

Are AI girlfriend apps profitable? A small minority are. Of roughly 337 revenue-generating companion apps Appfigures counted, only 33 have ever passed $1 million in lifetime revenue, and the top decile collects 89% of category revenue. The rest are running inference costs against revenue per download of roughly a dollar.

How do AI companion apps make money?

AI companion apps make money by charging for access, for consumption, and eventually for the company itself. The four layers are subscriptions, in-chat consumables, data and model licensing, and the acquisition exit. Every serious operator in the category runs at least two of them at once.

The layering matters because each one solves a different problem. Subscriptions give Replika and Nomi predictable monthly revenue and a number investors can model. Consumables capture the heavy users whose spending would otherwise cap out at $20, which is why Zeta took $33 million in six months without charging for the install. Licensing turns the conversational corpus into a second product. And the exit is where the largest single payout in this category’s history came from, Google’s $2.7 billion deal for Character.AI’s technology and team, which almost nobody writing about companion app revenue bothers to mention.

There’s a reason the stack looks like this. Companion apps have the engagement profile of a social network and the cost structure of an inference business. Users open them many times a day, every open costs real money in tokens, and the product has no advertising inventory worth selling at scale. So operators push the price toward the users who use it most, which is exactly what credit systems do. GreyJournal has covered this cost pressure before in its breakdown of why companies are pulling back from AI, and companion apps feel it harder than most because their users never stop talking.

What a companion subscription actually costs

Entry pricing in the category sits between $9.99 and $19.99 a month, with top tiers reaching $29.99 and annual plans discounting roughly 40% to 60% off monthly rates. Character.AI’s c.ai+ runs $9.99 a month or $94.99 a year. Replika Pro is around $19.99 a month or $69.99 a year. Nomi charges $15.99 monthly or $99.99 annually.

What the paywall gates is more interesting than the price. Across all three apps, the free tier delivers text. The paid tier delivers intimacy: voice calls, romantic relationship modes, selfies, longer memory. Replika’s own tier structure makes the logic obvious, because memory and emotional recall sit in the higher Ultra tier rather than the entry one. The thing users are actually paying for is continuity, and continuity is the most expensive feature to run.

Note the outlier. Zeta out-earned every subscription-first app in the first half of 2026 while charging nothing to install, which tells you where the category is heading.

Why do companion apps sell credits instead of subscriptions?

Credits remove the spending ceiling. A $19.99 subscription caps what the most devoted user can give you at $240 a year, while a credit system lets that same user spend $240 in a month without ever seeing a renewal screen. Appfigures’ first-half 2026 data shows the result: Zeta took $33 million, Tipsy Chat took $15.2 million, and ChatBox took $13 million, all on consumable-heavy models.

The mechanic is borrowed almost directly from live-streaming and gacha gaming. Users buy a currency with an unclear exchange rate, then spend it on messages, image generations, premium characters, or gifts the companion reacts to. Breaking the link between dollars and units is the whole point, because it makes the hundredth purchase feel like the first.

Subscriptions also have a structural weakness that credits don’t: a monthly cancellation moment. Every renewal is a small referendum on whether the relationship is worth $19.99. Credit purchases skip the referendum entirely. The user decides in the middle of a conversation, when they’re least likely to be doing arithmetic.

Download data supports the shift. Emochi led all romantic companion installs in the first half of 2026 with 7.9 million, ahead of Amora at 3.6 million and Zeta at 3 million, and the credit-driven apps are the ones converting those installs into eight-figure revenue. Founders working on any consumer AI product should study this pattern, because it’s the same logic that shows up in AI side hustles that monetize usage rather than access.

The revenue math that buries most of these apps

Of roughly 337 revenue-generating AI companion apps Appfigures counted, only 33 had ever crossed $1 million in lifetime revenue, and the top 10% of apps collect 89% of all category revenue. That is not a competitive market. That’s a lottery with 337 tickets and about 33 payouts.

The unit economics explain why. Appfigures measured revenue per download rising from $0.52 in 2024 to $1.18 in the first half of 2025, which sounds like a healthy trend until you set it against inference costs. A single engaged companion user can generate hundreds of model calls a day. At roughly a dollar of lifetime revenue per download, an app needs either a very steep conversion rate or very cheap inference to survive contact with its own users.

Here’s the part worth sitting with. The circulating headline figure for 2025 was about $120 million for the whole category, built from $82 million in the first half and a 64% year-over-year growth rate. The 2026 numbers look wildly larger because Appfigures changed how it cuts the data, splitting romantic and NSFW apps from general-audience companions and tracking 214 apps in the romantic group alone. Anyone quoting the $120 million figure next to the $327 million figure without explaining the definitional change is comparing two different categories.

The honest read: category revenue is growing fast, the growth is real, and the concentration is getting worse rather than better as it grows. New entrants are launching into a market where nine out of ten dollars are already spoken for. That’s the same shape GreyJournal described in its piece on what a wrapper startup is, and companion apps are wrappers with a stronger retention hook and a much worse legal profile.

Data licensing and the acquihire exit

The largest payday in AI companion history wasn’t a subscription business at all. In August 2024, Google paid a reported $2.7 billion to license Character.AI’s technology and bring back co-founder Noam Shazeer, a deal structured as licensing plus talent rather than an acquisition. Character.AI’s subscription revenue was a rounding error next to that number.

That transaction reframes the entire category. If the exit is a licensing deal with a frontier lab, then the asset being built is the model, the training data, and the team, with the consumer app functioning as the distribution that proves the model works. Revenue becomes evidence rather than the goal.

Conversational data is the second half of this layer. Companion apps collect something no other consumer product collects at volume: millions of hours of unguarded, emotionally loaded, long-context dialogue. Regulators noticed. When the FTC issued 6(b) orders to seven companies on September 11, 2025, including Alphabet, Character Technologies, Meta, OpenAI, Snap, and X.AI, it asked specifically how those firms “monetize user engagement” and how they collect and share personal information from user conversations.

Operators rarely describe data as a revenue line in their marketing. The FTC’s questions suggest the agency does not accept that framing, and founders building anything adjacent should assume the same scrutiny. GreyJournal’s guide to AI washing covers the related risk of describing a data business as something else.

Are AI companion apps profitable?

A small number are very profitable and most are not. The 33 apps that have cleared $1 million in lifetime revenue sit inside a field of 337, and the top decile’s 89% revenue share means the median companion app is earning a few thousand dollars against real inference bills.

Profitability in this category comes down to three variables. The first is inference cost per conversation, which favors operators running smaller fine-tuned models over anyone paying frontier API rates. The second is the share of revenue from consumables rather than subscriptions, because consumables scale with the heaviest users instead of capping at them. The third is app store tax, since Apple and Google take their cut of every subscription and every credit pack before the operator sees a dollar.

There’s a fourth variable now, and it didn’t exist in 2024. Compliance has become a fixed cost. Character.AI’s response to regulatory pressure included deploying age verification through third-party vendors including Persona, alongside behavioral analysis. Age verification has a per-check price, and it lands on every signup, including the ones that never convert.

The founders who are making this work tend to be solo or very small teams running lean inference, which is the same profile GreyJournal documented among solo founders building million-dollar AI businesses. Headcount is what kills companion apps, not demand.

What the new companion chatbot laws cost operators

Two states now regulate companion chatbots directly, and both laws attach costs to the exact features that drive revenue. California’s SB 243 took effect January 1, 2026, with a private right of action and damages of the greater of actual damages or $1,000 per violation, plus attorney fees. New York’s General Business Law Article 47 took effect November 5, 2025.

The New York requirements are specific and operationally annoying. Operators must tell users they’re talking to AI at the start of a session and again every three hours of continued use, and must detect expressions of suicidal ideation or self-harm and route users to crisis services. Governor Kathy Hochul sent an open letter to AI companion companies on November 10, 2025, notifying them the safeguards were live and that the Attorney General enforces them.

Think about what a mandatory three-hour interrupt does to a product whose revenue depends on immersion. It is a designed-in break in exactly the state the business monetizes. California layers on more: safety protocols around self-harm content, blocking sexually explicit content for minors, break reminders every three hours for minors, and annual reporting to the California Department of Public Health starting July 1, 2027.

Europe moved earlier and with money attached. Italy’s data protection authority fined Luka Inc., the maker of Replika, 5 million euros in 2025 over the app’s handling of personal data and its lack of age verification, and opened a further investigation into the underlying model. For a company with Replika’s revenue profile, a fine that size is not a cost of doing business. It’s an extinction event with a payment plan.

All of which brings the Character.AI decision back into focus. Anand cut teen access before any regulator forced him to, accepted the churn, and repositioned the product as a role-playing platform rather than a companion. Read as a compliance move it looks defensive. Read as a business move it looks like a company deciding which regulatory regime it wants to be governed by, five years early.

What founders should take from this

The transferable lesson has almost nothing to do with companionship. It’s that consumption pricing beats access pricing whenever your costs scale with usage and your heaviest users value the product far more than your subscription price admits. Zeta’s $33 million on a free install proves the point better than any pitch deck.

Three things worth acting on this week if you’re building consumer AI. First, check whether your top 5% of users are hitting a spending ceiling you built yourself; if they are, you’re leaving the Zeta gap on the table. Second, price your compliance before you need it, because age verification, crisis-detection routing, and disclosure interrupts all have engineering and vendor costs that are far cheaper to design in than to retrofit. Third, be honest in your own model about which layer is the actual business, since Character.AI’s $2.7 billion outcome came from the model and the team rather than the subscribers.

The companion category is going to keep growing and keep consolidating at the same time. Revenue per download more than doubled in a year, 128 new apps launched into a market where 89% of the money goes to the top tenth, and two states now regulate the core product loop. Those conditions favor whoever is already large. Anyone entering now is buying a ticket in a lottery that has already been drawn. For founders weighing whether this is the right AI market to enter, GreyJournal’s rundown of the top businesses to start in 2026 and its guide to starting an AI consulting business both point at the same conclusion: in AI right now, selling to the people building the apps is a steadier business than building one.