About 19% of American adults reported having conversed with an artificial intelligence built to act like a romantic partner, according to a survey of roughly 3,000 people released in February 2025 by the Wheatley Institute at Brigham Young University. Use is concentrated among younger adults: in the 18- to 30-year-old group the survey found 31% of men and 23% of women had tried such systems.
What the numbers show — and what they do not
The Wheatley Institute figures have been cited widely and interpreted in alarming ways, but experts and the researchers who conducted the work urge caution. The data come from self-reported responses collected via Qualtrics, and the investigators acknowledge the difficulty of producing a single, stable prevalence rate for a rapidly changing technology.
“We readily acknowledge the challenge of identifying an exact prevalence rate with rapidly evolving behavior such as the use of AI technologies,” said Brian Willoughby, the lead author of the Wheatley study.
Other surveys produce lower estimates. A Gallup poll commissioned by the Walton Family Foundation found monthly use of AI companionship tools among 18- to 28-year-olds closer to 10%. A separate 2026 study, titled “Secret Soulmates,” surveyed 2,431 partnered young adults and reported that about one in seven dating, engaged or married young adults regularly interacted with chatbots designed to simulate a committed partner.
Patterns and possible drivers
Beyond headline prevalence, the research highlights behavioral subtleties. In the Wheatley survey, about 21% of people who had used an AI companion said they preferred talking with that AI to speaking with a real person. The work suggests the appeal is not merely novelty; users may be drawn by predictability, emotional safety, or the absence of reciprocal demands.
- Demographic concentration: Young adults are the primary users, with men reporting higher use than women in the Wheatley sample.
- Variation across studies: Estimates range from roughly 10% to more than 30% depending on methods and population surveyed.
- Self-report limitations: These are descriptive surveys, not controlled experiments, and cannot establish causation.
Consequences for mental health, research and policy
Researchers and clinicians say the current evidence is useful for identifying trends but insufficient for conclusions about long-term effects. The available work does not track outcomes such as relationship stability, loneliness trajectories or diagnostic harms over time. That leaves unanswered questions about whether AI companions function as coping mechanisms, harmful substitutes for human connection, or something more complex.
The phenomenon also raises practical questions for regulators and platform makers. As AI tools that simulate intimacy become more sophisticated and more widely distributed, policymakers may be pressed to consider consumer protections, transparency around capabilities and limits, age restrictions, and how to integrate such technologies into mental-health frameworks.
Data at a glance
| Study | Sample | Key finding |
|---|---|---|
| Wheatley Institute (Feb 2025) | ~3,000 adults (Qualtrics) | 19% of adults used AI romantic partners; 31% of men 18–30; 23% of women 18–30 |
| Secret Soulmates (2026) | 2,431 partnered young adults | ~1 in 7 regularly interact with AI companions |
| Gallup (Walton Family Foundation) | 18–28-year-olds | ~10% monthly use |
For journalists and policymakers, the immediate imperative is twofold: to improve measurement and to fund longitudinal work that can separate short-term experimentation from sustained behavioral change. For clinicians, the guidance remains pragmatic: if an individual’s emotional well-being or relationships appear affected by AI use, evaluation by a qualified counselor or therapist is advisable.
The emerging trend underscores how quickly interactive technologies are entering intimate spheres of life. Whether these systems will reconfigure social habits, augment care, amplify isolation or do some combination of those outcomes is a question that current surveys can flag but not yet answer definitively.
