The $2.9 Billion AI Girlfriend Market Doesn't Exist — Here's What People Actually Spend

What we can actually measure

There is exactly one class of number in this category that is a measurement rather than a projection: app-store consumer spending, tracked by intelligence firms that read the App Store and Google Play revenue signals directly.

Appfigures reported that in the first half of 2026, $162.8 million flowed through 214 romantic and NSFW AI companion apps, with another $164.8 million through general-purpose companion apps. Since late 2022, the romantic category has taken $427.3 million in total, across 165.3 million downloads.

We think the romantic figure is about $13 million too high, because one app on their leaderboard is a general-purpose productivity assistant that appears to have been keyword-matched into the category. We set out that correction in detail. Call it ~$150 million.

That is the honest number. It is growing extremely fast — 38% of everything the category has ever earned arrived in the last six months — but it is not billions.

Where the billions come from

Three things inflate the forecasts, and they compound.

They define the category differently, and don't tell you how. 'AI girlfriend app' is a narrow slice. 'AI companion market' can be stretched to include voice assistants, multimodal agents, eldercare robotics, customer-service bots and enterprise deployments. A report that quietly counts enterprise conversational AI will land an order of magnitude above one counting romance apps — and both will be sold to you under similar-sounding titles.

They forecast rather than measure. Nearly every quoted figure is a projection with a CAGR attached, anchored to a base year the report rarely justifies. Once the base is inflated, a 25% compound rate carries the error forward for a decade.

They are marketing collateral. These reports exist to sell reports, typically for several thousand dollars. The headline number is the advertisement. There is no methodology section you can check, no sample, no disclosure of what was counted — and the press release is free precisely because the number travels.

The citation laundry

Watch what happens next, because this is the part that makes bad numbers durable.

A research firm publishes a press release with a headline figure. A tech blog reports the figure and links the firm. A second blog cites the tech blog. An SEO content farm — the gitnux, worldmetrics, zipdo tier — generates a page titled '110+ Verified AI Companion Statistics' that lists the number alongside dozens of others, many simply invented. A chatbot summarising the web then repeats it with total confidence.

At no point in that chain did anyone check the original methodology, because there isn't one to check. The number gets more citations and more authority at every hop while its evidentiary basis stays at zero. By the time it reaches you it has the texture of consensus.

We spent a while trying to verify revenue claims for individual web platforms — figures like '$25M ARR' or '$50M ARR' for the big AI girlfriend sites. Every trail ended at one of those content farms. Not one traces to a company filing, an audited statement, or a named source. We don't publish those numbers, and you should discount anyone who does.

We published one of these numbers ourselves

It would be comfortable to write this as an outsider. We can't.

An earlier article on this site told readers that analysts projected the market to grow "from about $49 billion in 2026 to more than $550 billion by 2035." We took it from secondary coverage, it sounded authoritative, and it sat on the page for weeks. It is not defensible, it contradicted our own statistics page, and we have now retracted it in place rather than quietly editing the sentence.

We are pointing at it because the mechanism that put it there is the same one that puts it everywhere: a big number, plausibly sourced, that nobody has any incentive to interrogate. Being sceptical of other people's figures is easy. The discipline is auditing your own.

The genuinely unmeasurable half

Here is the honest complication, and it cuts against our own headline.

App-store data cannot see the web. Candy.ai, Janitor AI, DreamGF, Muah and most of the explicit end of this market bill by card or crypto through a browser — specifically because Apple and Google won't distribute that content. None of them appear in the $150 million.

So the real total is higher. How much higher, nobody honestly knows. Traffic to these sites is measurable and substantial; converting traffic into revenue requires assuming a conversion rate and an average order value that no one outside those companies has.

That uncertainty is not a licence to pick a large number. It is a reason to say what is measured, say what isn't, and refuse to blend the two into a single confident figure. Our own best estimate for worldwide 2026 spending across mobile and web sits somewhere in the $1–1.5 billion range — and we label that an estimate, not a finding, because half of it rests on assumptions we can't verify.

How to read any market-size claim

Four questions that will disqualify most of what you'll read:

Is it a measurement or a forecast? If there's a CAGR attached, it's a forecast. Forecasts inherit every flaw in their base year.

What exactly was counted? If the report won't tell you whether enterprise chatbots are in the number, the number is unusable for consumer questions.

Who benefits from the figure being large? Research firms sell reports. Companies raise funding. Both are served by a big market and neither is penalised for being wrong in 2035.

Can you reach a primary source in under three clicks? If every link leads to another blog citing another blog, you have found citation laundry, not evidence.

Applied honestly, those questions leave you with a much smaller and much more interesting story: a category that turned over roughly $150 million on mobile in six months, is compounding faster than almost anything else in consumer apps, and is dominated by companies nobody has heard of. That's worth more than a made-up billion.

The sourced numbers, with every figure attributed and dated, live on our AI Companion Statistics 2026 page.