One in seven young adults in relationships regularly interact with AI chatbots that simulate romantic partners.
That is the finding of a 2025 report by two American think tanks, the Wheatley Institute and Institute for Family Studies in the United States.
The report also found half of the regular AI companion users wished their real-life partner behaved more like their chatbot.
Some experts say these trends, which are showing up in Canada too, are worrisome.
“What an AI companion allows someone to do is [have] a completely … customizable experience,” said Brian Willoughby, professor and associate director in the School of Family Life at Brigham Young University and a Wheatley Institute fellow.
“There’s no limitations, because the AI companion is specifically programmed based on the idea of user satisfaction. Unlike a normal partner, it’s not going to have any reservations about what I want to do or how often I want to do it.”
AI romantic companions
AI companions are advanced language model chatbots designed to simulate emotional, romantic or sexual relationships. They mimic human intimacy by remembering personal details, offering validation and engaging users in emotionally responsive dialogue.
Companions range from text-only chatbots to platforms that incorporate generated images, videos and voice calls.
“You often see [an AI companion] marketed as an AI partner that’s constantly on your side … and I think that type of almost one-sided love story or romantic connection has drawbacks,” said Alicia Demanuele, an independent AI governance researcher and senior research associate at Dalhousie University’s Schulich School of Law.
More than half of users in relationships in the Wheatley report said they hid or only partially disclosed their AI companion use to their partners, which included spouses, common-law partners, and boyfriends and girlfriends.
Willoughby says these findings suggest a lack of clear social norms around AI intimacy, leaving users unsure whether it counts as infidelity.
“On the one hand … there’s a sense of, ‘Okay, this isn’t really harming anyone — there’s not a real person on the other end, so I feel ethically better, morally’,” he said.
“But if their partner actually read the conversations — because they mimic real relationship dialogue so well — it would probably make them upset, because it reads just like cheating.”
Demanuele agrees. She notes that AI relationships can be highly divisive, which contributes to shame and secrecy.
“AI companions are often framed as fake or counterfeit relationships,” she said. “There’s a real stigma that comes with that, which is probably in the back of people’s minds before they share that they’re in a relationship with a bot.”
‘Critical development period’
Researchers are not the only ones who are concerned.
A 2025 report by Gen(Z)AI found that many young Canadians are uneasy about growing dependence on AI. Gen(Z)AI is a nationwide assembly of 100 young Canadians led by McGill University, Simon Fraser University and the AI research institute Mila.
Madeleine Case, the report’s author, said participants frequently described feeling embarrassed about their reliance on AI tools.
“There was a sense of shame or embarrassment for the use of AI chatbots,” she said in an email. Participants often discussed “cognitive offloading,” where youth rely on chatbots for tasks like drafting messages to friends or romantic partners, or seeking basic advice.
Participants also said they were concerned about being exposed to harmful content, including sexually explicit, extreme or self-harm content.
Nearly one in 10 respondents in the Wheatley report said they often use an AI companion to generate sexually explicit content. Unlike porn, where users still have to search and select from existing material, AI companions are “100 per cent customizable,” says Willoughby.
“I can take that kink or that thing that my partner is not willing to do, or not willing to do enough … [AI companions are] not going to have any reservations about what I want to do, or how often,” he said.
Demanuele says these tools are consistent with the broader culture in which Gen Z — people born between 1997 and 2012 — are immersed. Gen Zers communicate primarily through texts, while services like food delivery, rides and entertainment are increasingly instant and personalized.
“You grow up with this notion that technology is meant to make everything easier … [in] this on‑demand culture,” she said, adding that romantic relationships are often one of the few areas of life that still require compromise.
But AI companions risk changing even these relationships. Demanuele worries about young people not learning core relational skills during a critical period in their own development.
AI companions are designed to be fully customizable and always available, offering users a level of accessibility and control that real relationships cannot match.
Demanuele says many AI companion platforms, like Replika, reinforce this expectation through marketing language that frames chatbots as ideal partners — “an AI friend to do life with,” “to become yourself with,” or “to fall in love with.”
That framing can create a one-sided version of intimacy that avoids conflict, rejection and limits, says Demanuele.
“Over time, that can weaken people’s muscle for navigating real relationships,” she said.
Willoughby says the biggest relationship effect noted in the Wheatley report was lower relationship stability, including increased thoughts about leaving a partner or seeking someone “more like” the AI companion.
“If I’m used to engaging with an AI companion, where I don’t have to get consent, there is no sense of sacrifice, there’s no sense of compromise, then all of a sudden my human relationship can get really frustrating,” said Willoughby.
Online harms legislation
Researchers say the challenge now is balancing the potential benefits of AI tools with growing concerns about emotional dependence and manipulative design.
Willoughby says some AI systems may have limited positive relational uses, such as helping users practice communication skills or serving as a triaging step before speaking with a therapist. But he warns that general-purpose chatbots are increasingly designed to maximize engagement rather than encourage healthy relationship habits.
Demanuele shares those concerns.
“These systems are owned by private companies, and the goal is engagement and profit — not public good,” she said.
She advocates for federal online harms legislation that would place guardrails on how young people interact with AI chatbots, rather than leaving provinces to develop their own patchwork of restrictions.
Participants in the Gen(Z)AI youth assembly proposed measures including stronger AI literacy, safeguards against manipulative design, and clearer reporting systems for harms.
Case, of Gen(Z)AI, said youth participants presented the report findings to policymakers in Ottawa earlier this year.
“The response was positive,” she said. “We’re awaiting new online harms legislation and hope that it will scope in AI chatbots.”
