AI Girlfriend Statistics: The Numbers Worth Citing, and the Ones to Distrust

Most AI girlfriend statistics circulating in 2026 come from one of three places: a survey with a stated sample, a web analytics panel, or a vendor blog with neither. The numbers differ wildly depending on which you read. This page sorts them by source quality so you can cite the defensible ones and flag the rest.

One note before the figures. Nobody knows how many people currently use an AI girlfriend. What exists are proxies: survey answers about ever having chatted with one, panel estimates of app audiences, and search volume. Treat all of them as estimates with a method attached, not as a headcount.

Key statistics at a glance

Statistic	Figure	Source
US adults who have chatted with an AI romantic partner	19%	Wheatley/IFS 2025
Men vs women aged 18-30 who have chatted with an AI partner	31% vs 23%	Wheatley/IFS 2025
US singles who used AI as a romantic companion	16% (Gen Z 33%)	Match Singles in America 2025
US teens 13-17 who used a companion app	72% (13% daily)	Common Sense Media 2025
Male share on UK mixed-use platforms	77%	Ada Lovelace Institute 2025
Monthly "ai girlfriend" searches	~93,000 US, 253,000 worldwide	Ahrefs, September 2026

US usage: the survey numbers

The most cited US figure is 19% of adults, from a Wheatley Institute and Institute for Family Studies survey of roughly 3,000 respondents in 2025. That means about one in five US adults says they have chatted with an AI romantic partner at some point. It does not mean one in five currently has an AI girlfriend, and the survey does not measure that.

Match's Singles in America 2025 study, with 5,001 participants, puts the share of US singles who have used AI as a romantic companion at 16%. Broken down by generation, 33% of Gen Z and 23% of Millennials said they had. Match also found 26% of singles now use AI to enhance their dating lives, up from 6% previously, a rise the company framed as 333%.

A separate cluster of survey work puts 27-31% of US internet adults using large language models for social or emotional interaction, and about 31% of AI users considering a chatbot a friend as of 2026. Those questions are broader than romance. A chatbot friend and an AI girlfriend are not the same product, and mixing the two inflates the headline.

One figure to treat with real caution: aigirlfriends.ai claims almost 45% of US men aged 18-34 have tried an AI girlfriend app. That is a vendor publishing a statistic about its own market, with no disclosed method. It also sits far above the Wheatley figure for the same age band. Mention it as a claim, not as data.

Gender: who actually uses these apps

The gender split depends almost entirely on which kind of app you measure. Adult-oriented companion apps skew heavily male. Roleplay and character platforms are much closer to even, and one of them disputes the panel data outright.

Platform	Male share	Source
Candy.ai	83%	Similarweb
Joi	80%	Similarweb
OurDream	79.6%	Similarweb
CrushOn	77%	Similarweb
SpicyChat	75%	Similarweb
Janitor AI	62.7%	Similarweb

Aggregate estimates land in the same place. The Ada Lovelace Institute found 77% of users on UK mixed-use platforms were male, and ChinaTalk put the US male-to-female ratio at roughly 7:3 overall and about 8:2 among 18-24s. Across surveys, 75-80% of romantic AI users are male. Men are also more than twice as likely as women to say they would consider an AI partner at all.

Then there is the outlier. Janitor AI publicly claims that 70-80% of its users are women, which directly contradicts Similarweb's 62.7% male panel estimate. Both cannot be right. Similarweb measures web traffic panels, which miss app usage and logged-in behaviour; the company measures its own sign-ups, which it has not detailed. DemandSage, meanwhile, puts the Character.AI audience close to even. If you need one line for a piece: adult companion apps are mostly male, roleplay sites are mixed, and the exact split is genuinely contested.

App naming follows the same skew. Appfigures found 17% of companion apps are named around girlfriends versus 4% around boyfriends. Usage intent differs too: 17% of young women say they chat for arousal against 32% of young men.

Age: younger than the marketing suggests

Age estimates disagree more than gender ones. Statista data surfaced through Google AI Mode puts 18-24s at over 65% of global users. A Simon Lermen analysis found nearly 90% of users under 35. The Ada Lovelace Institute says 18-24s are about 39% of users, and DemandSage reports the Character.AI web audience as 51.84% aged 18-24.

Against those, aigirlfriends.ai and Grand View put roughly 41% of users at 35 and older. That is a very different picture, and it comes from the least transparent sources in the set. Older users may be undercounted on platforms that report web traffic rather than app installs, and younger users are likelier to be counted twice across devices. Nobody has resolved this.

Teens and age restrictions

The teen numbers are the most striking in the whole field, and the most sensitive. Common Sense Media surveyed 1,060 US teens and found 72% had used an AI companion, 52% used one regularly, and 13% used one daily. One third said they used companions for social interaction or relationships. The Center for Democracy and Technology found 19% of high schoolers had been in, or knew someone in, an AI romantic relationship. AllAboutAI estimates up to 20% of teens and young adults spend time with AI the way they would with human friends.

Note the wording. "AI companion" in the Common Sense survey covers character chatbots, study helpers and roleplay tools, not just romantic apps. It is not a measure of teen AI girlfriends specifically.

The adult apps we review are 18+ only. California's SB 243 requires disclosure and self-harm protocols and covers cases where a minor is known to be using a service. New York has a parallel law. Enforcement is another matter, and age gates on these sites are typically a date-of-birth field.

Search demand: the clearest trend line

Search volume is the least glamorous statistic here and probably the most reliable directional signal. Ahrefs data for "ai girlfriend" shows a steep climb.

Period	US monthly searches
2022 average	3,425
2023 average	34,891
2026	about 95,000
Since February 2026	85,000-99,000

Worldwide volume is about 253,000 monthly searches. Related terms grew too: "ai companion" searches rose 53% over the past 12 months, and "ai boyfriend" hit an all-time high of roughly 9,000 monthly searches in September 2026. Interest peaks in December-January and again in June-July, which looks like holiday and summer loneliness rather than product launches.

Skim AI and Hera Haven report a 525% one-year increase in these searches. That is plausible against the Ahrefs trend but is a much larger jump than the monthly figures imply, so cite the underlying numbers instead.

Search interest does not equal paying users. Plenty of people search, read a review and leave.

Dating and market figures

AI is increasingly part of ordinary dating rather than a replacement for it. Skim AI's roundup found 54% of people would prefer AI help finding dating matches, 47% would use AI dating apps for long-term partnerships, and 45% want match recommendations based on physical attraction. One in five young people say they are open to AI or virtual partners. Half of adults under 30 have used dating sites or apps.

The broader dating app market was worth $2.86 billion in US revenue in 2022, projected to reach $3.41 billion by 2027. Companion apps are a smaller but faster-growing slice. Appfigures, via TechCrunch, recorded 60 million downloads in the first half of 2025, up 88%, with $82 million in mobile spending over the same period; the top 10% of apps took 89% of revenue. Sensor Tower put in-app revenue at $150 million in Q1 2026. One widely repeated figure claims these apps grew roughly 700% between 2022 and 2025.

Market size is where the numbers fall apart. Globe Market Research values the market at over $30 billion. Our own estimate puts total consumer spending nearer $1-1.5 billion a year, and we hold that at low-to-medium confidence. The gap comes from definition: the $30 billion figures sweep in enterprise assistants, robotics and general AI, not just companion apps. The range across published estimates runs from $2.6 billion to $30 billion for what is supposedly the same market.

Platform user counts are worth listing but not trusting equally. Character.AI has about 20 million monthly users according to Sacra. Joi has about 6.5 million monthly users per Ofcom. SpicyChat reports 100 million registered accounts, which is not the same as active users. Janitor AI reports 2.5 million daily users. None of these companies disclose how it counts.

Emotional effects and loneliness

This is the section where headlines outrun evidence. Almost all of it is correlational.

The APA Monitor reports that over half of users discuss stress with AI companions, and finds a negative correlation between heavy AI reliance and relationship health. An OpenAI and MIT Media Lab study surveyed 4,076 people and ran a trial with 981 participants; more daily AI interactions were linked to more loneliness and fewer social connections. Wheatley found over 50% of male users and over 60% of female users at depression risk, roughly double the rate for non-users. Stanford researchers warn that companions may worsen loneliness for vulnerable users.

None of these establish causation. It is equally plausible that lonely and depressed people are drawn to these apps. A Nature Human Behaviour analysis of the Replika subreddit traced negative sentiment rising from 13% to 38%, with sentiment improving after Replika removed erotic roleplay in 2023, which suggests product changes move user mood as much as the reverse.

Some users report benefit. 43% of romantic AI users call the AI a better listener than humans, 21% prefer AI over humans for talking, and 25% of young adults believe AI partners could replace human romance. Those are attitudes, not outcomes.

The context matters. Pew finds 63% of men under 30 describe themselves as single. The Survey Center on American Life finds 15% of men have no close friends, and about one in five young men report no close friend. A product that fills a gap is not automatically harmful, but the demand is real and measurable.

Privacy and safety risks

The security record on these apps is poor, and that is a statistical claim you can stand behind. Mozilla found 10 of 11 romantic AI chatbots failed minimum security standards, and estimated about 90% may share or sell personal data. The Ada Lovelace Institute found 14 of 16 UK platforms claimed rights to chat logs.

Breaches have happened. Muah.ai exposed about 1.9 million email addresses. Chattee Chat and GiMe Chat leaked data on over 400,000 users and 43 million messages. Italy fined the Replika developer €5 million under GDPR in 2025.

Regulation is catching up. California SB 243 and the New York law both require disclosure and self-harm protocols. If privacy is your main concern, the platform you pick matters more than any average, which is why we flag token traps and data handling in individual reviews such as our Candy.ai review and our OurDream AI review.

How to read these statistics without getting fooled

Six checks before you cite anything in this space:

1. Definition. Companionship, romantic relationships, AI used inside dating, and chatbot friendship are four different questions. Many published figures blur them.
2. Sample and funder. A survey of 3,000 adults with a named funder beats an unlabelled percentage on a vendor blog.
3. Ever vs now. "Have you ever chatted with an AI partner" produces a much larger number than "do you use one weekly."
4. Panel vs company claim. Similarweb and company-reported splits disagree by design. Where they conflict, say so.
5. Market scope. A $30 billion figure almost never means consumer companion apps.
6. Correlation. Loneliness studies are mostly cross-sectional. Write "linked to", not "causes".

Two specific traps. The 54% dating stat is recycled across blogs without the question it came from. And vendor blogs publish dramatic figures with no method section at all, then get quoted by other vendor blogs. If a number has no identifiable sample, it is marketing.

For where individual apps sit against each other, our review hub lists every platform we score and what each costs, and our value comparison covers what you get per dollar.

AI frontiers to watch

Two platform retreats in 2026 suggest the field is consolidating around safer products. OpenAI paused ChatGPT's adult mode in March 2026, and xAI retired Grok's 3D companions in September 2026. Both moves remove capability that users had already built habits around.

Meanwhile demand keeps climbing. If the search trend holds through 2027, the interesting question stops being how many people try an AI girlfriend and becomes how many stay, what they pay, and what the apps do with the transcripts. Those are the numbers we would most like to see published, and the ones companies are least likely to release.

Bottom line

About one in five US adults has tried AI romance. Adult companion apps skew roughly 75-83% male; roleplay sites are closer to even and one of them disputes the panel data. Search volume is near record highs in 2026. AI is increasingly used inside dating rather than instead of it. The emotional-effects research is real but correlational, and the privacy record is bad enough to factor into any choice.

Trust surveys with stated samples and analytics panels over app-maker blogs. When a vendor quotes a percentage about its own market, attribute it to the vendor and move on.