How Our AI Companion Survey Compares With Other Major Studies

Match and the Kinsey Institute: 16% of singles, against our 24.7% of adults.

Singles in America (2025) surveyed 5,001 U.S. singles aged 18 to 98 and found that 16% had interacted with AI as a romantic companion, rising to 33% of Gen Z and 23% of Millennials. Our survey found 24.7% of respondents had ever used an AI girlfriend or boyfriend app, and 12.3% currently did.

Those look close, but three differences sit between them. Match surveyed singles only; our sample is all relationship statuses, and 61% of it is engaged or married. Match asked whether respondents had "interacted with AI as a romantic companion," which can include briefer or lighter contact than using a dedicated app; ours asked specifically about using an AI girlfriend or boyfriend app. And Match measured ever-use; we measured both ever and current.

Pew Research Center: 4% of adults, against our 12.3%.

Americans and AI (June 2026) surveyed 5,119 U.S. adults and found that 49% use AI chatbots, but only 4% use them for companionship and 10% for emotional support or advice. That 4% is far below our 12.3%.

The gap is the definition. Pew asked about general-purpose chatbots such as ChatGPT, Gemini and Copilot, and about the purposes people use them for. Our question was about dedicated AI companion apps built around social and romantic interaction. Pew's own data shows how much that matters: 44% of U.S. adults use ChatGPT, while 3% use Character.AI, the best-known companion platform. "Uses a chatbot for companionship" and "uses a companion app" are different behaviors with different base rates, and a person could plausibly answer yes to one and no to the other.

Match's 2026 AI dating survey: 12% of 18 to 24, against our 3.8% of 18 to 29.

Match's survey of 1,000 U.S. singles aged 18 to 39 found that only 12% of 18 to 24 year olds had used a companion app in the previous three months, and about a third of those said they were seeking genuine connection. Our survey found 3.8% of respondents aged 18 to 29 currently using one. The bands and time windows differ, but both studies place companion-app use among the youngest adults well below the much larger figures sometimes cited for broader AI-companion exposure.

Common Sense Media: 72% of teens, against our 12.3% of adults.

Common Sense Media (July 2025) surveyed 1,060 U.S. teens aged 13 to 17 and found that 72% had used an AI companion and 52% used one regularly, with a third having chosen an AI over a person for a serious conversation. That looks wildly incompatible with our 12.3%. It is not. They studied minors; we studied adults, and our sample contains no one under 18. Common Sense also defined AI companions as "digital friends or characters you can text or talk with," a broader category than romantic companion apps. "72% of teens have used an AI companion" and "12% of adults currently use one" can both be true. They are different populations answering different questions.

Match's 2025 study found that 40% of singles considered having an AI boyfriend or girlfriend while in an exclusive relationship to be cheating, with 39% saying the same about sexual chats with AI and 38% about emotionally intimate conversations. Our survey found 50.5% of respondents said a partner's romantic or intimate relationship with an AI companion could count as cheating, and 71.7% would either call it cheating or be uncomfortable with it.

The questions differ in two ways. Match asked about "having an AI boyfriend or girlfriend," a specific arrangement. We asked about "a romantic or sexual relationship with an AI companion," a broader description of the behavior, and we offered a conditional answer ("only if they became attached") that accounts for 17.1 of our 50.5 points. Match also surveyed singles, who in our own data were the least likely group to call AI romance cheating: 42.2% of our single respondents did, against 54.8% of engaged or married respondents. Restricting our sample to singles brings the two studies to within about two points of each other.

We treat the two as consistent. Both find that a large plurality to a majority of respondents regard an AI relationship as a potential betrayal, and both find that the exact framing moves the number.

Harvard: AI companions reduce momentary loneliness, and feeling heard is the mechanism.

De Freitas and colleagues (Harvard Business School, Journal of Consumer Research, 2026) ran several experiments testing whether interacting with an AI companion reduces loneliness. They found momentary reductions, comparable in one study to interacting with a person and greater than passive activities such as watching videos, with repeated reductions across a week of daily use. A key mechanism was whether participants felt heard.

That fits our motivation result from the other direction. Asked their main reason for using an AI companion, 56.6% of our respondents chose having someone to talk to or feeling less lonely, and only 10.3% chose affection, romance or intimacy. Our respondents describe wanting to be heard; the Harvard experiments suggest that is what the technology can, at least briefly, deliver.

MIT and OpenAI: heavier voluntary use went with more loneliness, not less.

MIT Media Lab and OpenAI (2025) ran a four-week randomized study with 981 participants on the psychosocial effects of ChatGPT use. The assigned conditions did not produce significant differences on the main outcomes, but participants who voluntarily used the chatbot more reported greater loneliness and emotional dependence and less socializing. The researchers cautioned that effects varied widely and that the findings, based on ChatGPT rather than companion apps, should not be generalized broadly.

We did not measure outcomes, so there is nothing here for us to agree or disagree with. We cite it because it is the strongest evidence that stated motivation (our data) and actual effect (theirs) may diverge over time, which is a limit on how our 56.6% should be interpreted.

Match found AI romantic companionship highest among Gen Z (33%) and declining with age. Pew found companionship use of general chatbots declining steadily with age: 7% among adults aged 18 to 29, 6% among 30 to 49, 2% among 50 to 64 and 1% among those 65 and older. Both point the same direction: youngest first.

Our survey found the opposite. Asked whether they would consider using an AI girlfriend or boyfriend, 44.7% of respondents aged 30 to 45 said yes, against 12.5% of those aged 18 to 29. Reported current use followed the same pattern: 20.7% versus 3.8%. On every openness measure we took, the 30 to 45 group led and the youngest adults trailed.

We do not think one side is simply wrong, and we have looked for the reasons. Four are visible in the instruments.

• Definition. Match measured having ever "interacted with AI as a romantic companion," a lower bar than ongoing use of an app. Pew measured general chatbot use for companionship. Ours measured use of, and openness to, a dedicated romantic companion app. It is plausible that young adults lead on lighter forms of AI companionship and general chatbot use while adults in their 30s and 40s lead on dedicated companion apps; the three studies would then all be right about different things.
• Population. Match's respondents are all single. Ours are 61% engaged or married, and in our data married respondents reported higher use and openness than singles. A sample with more partnered 30 to 45 year olds will produce a different age curve.
• Sample composition. Our 18 to 29 group is 184 respondents, the smallest of our four main age bands, and our sample skews older than the U.S. adult population. Small bases and skew both widen the uncertainty around the youngest group's figure.
• Device effects. Reported use in our survey differed sharply by device type even after quality screening, and this was concentrated in the 30 to 45 group. The direction of the age gap held within both device groups, but its size did not, which is why our report advises describing the 30 to 45 group as "substantially more open" rather than quoting a multiple.

Our position: the ordering we found is real within our sample and survived every check we could run. Whether it generalizes to the adult population, or reflects our sample's composition and our narrower definition, is a question that needs a second survey with randomized answer order, an attention check, and a larger under-30 base. We say so in the report and repeat it here.

1. Population. Adults, singles, teens, or app users. Common Sense's 72% and Pew's 4% describe different people.
2. Definition of "AI companion." A dedicated romantic app, any digital character, or a general chatbot used socially. Pew's Character.AI (3%) versus ChatGPT (44%) split shows how far apart those are.
3. Current versus ever use. Our 12.3% and 24.7% are the same respondents measured two ways.
4. Romantic companionship versus general chatbot use. Match's "interacted with AI as a romantic companion" and Pew's "use a chatbot for companionship" are not synonyms.
5. Question wording. "Having an AI boyfriend or girlfriend" (Match, 40%) versus "a romantic or sexual relationship with an AI companion" with a conditional option (ours, 50.5%).
6. Survey timing. Common Sense fielded in April and May 2025; Pew in February 2026; ours in August 2026. The category moves quickly.
7. Screening and data-quality procedures. Our published analysis explicitly documents the exclusion of 441 responses after fieldwork. Screening procedures vary across studies and are not always described at the same level of detail in public reports, making direct methodological comparisons difficult.

• The age curve (section 5). Ours inverts the others.
• Current use of dedicated apps (12.3%). Higher than any general-chatbot companionship figure and higher than Match's 2026 three-month figure for young adults, though below Match's 2025 ever-use figure for singles. We attribute part of this to definition and part to the residual device effect our report discloses, and we advise treating it as a description of our sample rather than a national estimate.
• Users taking AI romance more seriously than non-users. 80.6% of our current users said AI romance can be cheating, versus 44.0% of never-users. We have not found a comparable cross-tab in any other published survey, so this is unreplicated rather than contradicted. It is analyzed in our post on the 422 people who had used an AI companion.

• Cheating attitudes. Match's 40% among singles and our 42.2% among singles are close; the direction and the sensitivity to wording are the same.
• Conversation and loneliness as important motivations. Our 56.6% motivation result is consistent with the Harvard finding that AI companions reduce momentary loneliness via feeling heard, and with Common Sense's finding that teens most often use companions for conversation and support rather than romance.
• Dating apps over AI. Our 52.2% versus 7.8% preference for dating apps is consistent with Match's 2026 finding that most singles hold negative views of AI in romantic contexts and that 40% would refuse to date someone who uses a companion app.
• Low current use among the youngest adults. Our 3.8% for 18 to 29 and Match's 12% three-month figure for 18 to 24 both sit far below the headline "Gen Z" numbers that circulate.