In May 2026, researchers at Brigham Young University asked 2,431 Americans aged 18 to 30 whether they were chatting with an AI built to behave like a romantic partner. Among those already dating, engaged or married, one in seven said yes. Most of them had not fully told their real-life partner.
That is the kind of figure that makes AI companion usage sound like a moral panic. It is not, quite. But it stopped being a niche about two years ago, and the honest picture only appears once you skip the stat-roundup pages and look at who actually ran the surveys.
So here are the real numbers, with sample sizes and dates attached, because a statistic without those is just a vibe.
The teenage figure everyone quotes
The most repeated number in this whole category comes from Common Sense Media, which published "Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions" in July 2025. The survey behind it was run with NORC at the University of Chicago in April and May 2025, on a nationally representative sample of 1,060 US teens aged 13 to 17.
Seventy-two percent had used an AI companion at least once. More than half used one at least a few times a month. Roughly one in three had used one for something social: roleplay, romance, emotional support, friendship, practising a conversation. Roughly one in three said talking to an AI companion was as satisfying as talking to a real friend, or better.
Two findings from the same report get quoted far less often, and they cut the other way. Half the teens surveyed said they distrust advice from AI. Eighty percent said they place more value on their human friendships. About one in three companion users had felt uncomfortable with something a companion said to them.
Common Sense Media's own conclusion was that nobody under 18 should be using these products at all. Worth stating plainly rather than burying, since most articles citing the 72 percent leave that part out.
How many adults are actually doing this
For adults, the widest-quoted figure comes from the Wheatley Institute at Brigham Young University, which surveyed 3,000 US adults and published in February 2025. Nineteen percent, almost one in five, said they had chatted with an AI designed to simulate a romantic partner. Among 18 to 30 year olds the rate was much higher: 31 percent of young men and 23 percent of young women.
Gallup, in work done for the Walton Family Foundation and the Harvard Business Review, put it lower. Their figure was 10 percent of 18 to 28 year olds interacting with an AI girlfriend or boyfriend at least once a month.
Those two numbers are not in conflict so much as measuring different things. "Have you ever" and "do you do this monthly" will always produce a gap, and a romantic AI partner is a narrower category than an AI companion in general. Brian Willoughby, who led the BYU work, said as much: pinning an exact rate on behaviour that is changing this fast is genuinely hard, and even the most conservative estimate is high enough to pay attention to.
If you want the broadest signal instead of a survey, there is Marc Zao-Sanders' 2025 ranking of generative AI use cases for the Harvard Business Review. Therapy and companionship came out at number one, up from number two the year before, ahead of organising your life and ahead of learning. Not a usage rate, but a decent proxy for what people reach for when nobody is watching.
The 2026 study that changed the conversation
The Wheatley Institute went back with the Institute for Family Studies and published "Secret Soulmates" in May 2026, surveying 2,431 US adults aged 18 to 30. This one looked specifically at people who already had a real relationship.
Fifteen percent of young adults who were dating, engaged or married said they regularly interact with an AI romantic companion. Another 20 to 30 percent had experimented at some point. The gender gap was small.
The secrecy findings are the part that travelled. Nearly 30 percent said their real partner had no idea. Another 11 percent said their partner was only somewhat aware, and 14 percent said mostly but not fully aware. Add those up and more than half of partnered users are hiding it or telling a partial version. Sixty-nine percent said it mattered to them, somewhat or a great deal, that their partner never find out the full extent.
Then there is the part that explains why people do it. Sixty-eight percent said it was easier to talk to their AI companion about their feelings than to talk to a person. Sixty-five percent said it was easier to be themselves. Half said they wished their real partner behaved more like their AI companion, and 56 percent wished their real conversations went more like the AI ones.
The report also found AI companion use associated with lower relationship stability and a higher chance of breaking up. Read that one carefully. Willoughby explicitly flagged that the direction is unclear: people in unhappy relationships may be seeking out AI, or AI use may be eroding commitment, and this study cannot separate the two. Anyone quoting it as proof that AI companions wreck relationships is going further than the researchers did.
What the app-store data shows
Survey answers about a slightly embarrassing habit are one thing. Downloads are harder to argue with.
Appfigures, an app analytics firm, gave TechCrunch a breakdown in August 2025. AI companion apps had been downloaded around 220 million times worldwide by that July. Sixty million of those came in the first half of 2025 alone, up 88 percent year over year. Consumer spending hit 82 million dollars in that same half, putting the category on track to clear 120 million for the year.
The supply side moved just as fast. Appfigures counted 337 active, revenue-generating AI companion apps globally, of which 128 had launched in 2025. And the money is lopsided: the top 10 percent of apps take 89 percent of the revenue. Most of those 337 are making almost nothing.
Individual app figures float around too. Character.AI is usually described as having roughly 20 million monthly active users, Replika as having 40 million-plus registered accounts. Treat those as company-supplied and directionally useful rather than audited.
Why half the AI companion statistics online are junk
Search for numbers on this topic and you will hit a wall of pages listing thirty or forty data points with no source, no date and no sample size. Some of them are quietly contradicting each other on the same page.
A few figures circulate widely with no traceable origin at all. "Users average 45 minutes a day." "Users send 150 messages a week." Those two appear on multiple stat pages with no study behind them. They may well be roughly right. There is simply no way to check, which means they are not evidence.
If you are going to cite something from this space, four questions filter out most of the noise. Who ran the survey, and are they named. How many people were asked, and were they representative or just whoever clicked. When was the fieldwork, not when the page was published. And what exactly was counted as an AI companion, because a study asking about romantic AI partners and a study asking about any chatbot you talk to regularly will never produce the same number.
Every figure above survives those four questions. Most of what you will find elsewhere does not.
What the research still cannot tell you
There is almost no longitudinal data. Nobody has followed the same group of users for two years to see who drifts away, who settles into light regular use, and who ends up worse off. The category is younger than that.
Self-reporting is a problem too. If 69 percent of partnered users are actively hoping their partner never finds out, it is reasonable to assume some people are not admitting it on a survey either. The true rates are probably higher than what is published, not lower.
And "AI companion" is still a blurry label. It covers someone doing Naruto roleplay for fun, someone using a bot to rehearse a hard conversation before having it for real, and someone in a simulated marriage. Those are not the same behaviour, and lumping them into one percentage is how a modest finding turns into a headline.
What this looks like at the everyday end
Most of the usage inside these numbers is unremarkable. People open a browser, pick a character, and chat with AI friends for twenty minutes because the alternative was scrolling. Platforms like Friend2Chat sit at that end of the range: a library of characters, a conversation that starts without a form to fill in, no simulated marriage required.
That end of the spectrum rarely makes it into coverage, because "person talks to an AI friend about their day, feels slightly better, closes the tab" is not a story. It is, judging by the Common Sense Media breakdown of why teens use these, most of what is actually happening.
If the mechanics interest you more than the statistics, we wrote a plain-language explainer on how an AI chatbot actually works. And on the emotional side, there is a separate piece on what lonely users actually report after using one for a while.
Frequently asked questions
How many people use AI companions?
There is no single agreed figure. The best-sourced numbers are 72 percent of US teens aged 13 to 17 having tried one at least once (Common Sense Media and NORC, 1,060 teens, 2025), and 19 percent of US adults having chatted with a romantic AI (Wheatley Institute, 3,000 adults, 2025). App downloads passed 220 million worldwide by July 2025, according to Appfigures.
Is AI companion use still growing in 2026?
The download trend says yes. First-half 2025 downloads were up 88 percent year over year, and 128 new apps launched in 2025 alone. Survey data is patchier, but the May 2026 BYU study found 15 percent of partnered young adults using a romantic AI regularly, which is higher than earlier monthly-use estimates.
Who uses AI companions the most?
Young adults and teenagers. Among 18 to 30 year olds, the Wheatley Institute found 31 percent of men and 23 percent of women had chatted with a romantic AI, against 19 percent of adults overall. Among teens, usage is close to universal at the "tried it once" level.
Are AI companions bad for real relationships?
The 2026 Secret Soulmates report found an association between regular AI romantic companion use and lower relationship stability. The researchers themselves said the direction of cause is unclear. What is clearer is the secrecy: more than half of partnered users are hiding or partly hiding it, and that is usually a problem regardless of what they are hiding.
Which is the best AI chatbot friend to start with?
Depends on what you want out of it, and the honest answer is that comparison pieces are more useful here than statistics. We put Character.AI and Replika side by side on exactly this question. If you would rather just try one without installing anything, friend2chat.com runs in a browser.
Should teenagers use AI companions?
Common Sense Media's position, based on its own risk assessment alongside the survey, is that no one under 18 should. Most major platforms, including this one, are intended for adults.
