Does an AI Relationship Count as Cheating? It's Not Really About Whether the AI Is Real

These are not points on a scale from strict to relaxed. They are four different ways of deciding what infidelity is.

The survey cannot say which of these is correct. Cheating is defined inside a relationship, not by a poll. What the survey shows is that people are using very different criteria to decide where the line sits, and that the criterion most people assume is the obvious one, whether the other party is real, is held by fewer than three in ten.

And one of those definitions becomes much more important once we look at who chose it.

The distinction they are drawing is between who or what a partner is interacting with and what the partner is investing in that interaction.

For this group, the AI does not need to have feelings. The human does. A partner who exchanges flirtatious messages with a chatbot has not crossed the line; a partner who has come to need those conversations has.

The 21.2% who are uncomfortable without calling it cheating occupy a different middle ground. They do not consider the behavior infidelity, but the AI's non-human status is apparently not enough to make them comfortable with it either. The survey did not ask what specifically bothered them.

Put the two groups together and 38.3% rejected both ends of the question.

They neither said any AI romance is automatically cheating nor dismissed the issue because the AI isn't real. For them, the boundary is more conditional than either extreme allows.

Respondents who currently use an AI girlfriend or boyfriend app were the strictest group in the survey. 80.6% said a partner's AI romance can count as cheating.

Among former users, 60.2% did. Among people who had never used one, 44.0%.

The definition shifted along with the level. Current users' single most common answer, at 43.6%, was the attachment definition. Never-users were far more likely to dismiss the issue entirely: 33.0% said an AI is not real so it cannot be cheating, against 11.4% of current users.

If firsthand experience simply made AI companions feel more obviously artificial, users would be the least concerned group. The pattern runs the other way. The people who know these apps best were the least willing to say the question does not apply.

The survey cannot say why. Using an AI companion might change how seriously people take one; or people who already regard AI relationships as meaningful might be the ones who try the apps. Either way, experience was associated with taking the boundary more seriously, not less. I looked at this group in detail in People Who Use AI Companions Take AI Relationships More Seriously.

Among the 500 who could imagine romantic feelings for an AI, 73.0% said a partner's AI romance can count as cheating. Among the 1,209 who could never have such feelings, 41.2% did.

Separately from the cheating question, we asked whether respondents could imagine developing genuine romantic feelings for an AI companion. The two questions were not linked in the survey, which is what makes the overlap worth looking at.

The more revealing number is underneath. Among people open to AI feelings, only 7.2% said it is not cheating because the AI is not real. Among people closed to AI feelings, 37.1% chose that answer.

People who believe it can treat a partner's AI romance as consequential. People who believe it cannot are the ones who reach for "an AI isn't real."

A caution. This is an interpretation of an association, not something respondents told us in those words. But it fits both patterns at once: firsthand users and people open to AI feelings were both stricter, not looser. The feelings cross-tab gets its own post, answer by answer: the AI infidelity paradox.

Women more often locate the offense in the behavior, men more often in the attachment.

Women were more likely to choose the strictest definition, that any romantic or sexual AI interaction counts, at 38.0% versus 28.2% of men. Men were more likely to choose the attachment definition, at 21.2% versus 13.6% of women.

So men and women arrive at almost the same overall answer through somewhat different boundaries. The survey cannot say why, and I would not guess. The rest of the gender picture, including what each wants from an AI companion, is on the main survey page.

Engaged or married respondents were more likely than singles to say a partner's AI romance can count as cheating: 54.8% versus 42.2%.

Those dating sat in between at 45.8%. Singles were also the group most likely to say an AI is not a real person.

What makes this more than a demographic footnote is that married respondents were also more open to AI companionship on several other measures in the wider survey. At the group level, greater openness to AI companionship and stricter boundaries around a partner's use of it coexist, which is the same shape as the two twists above. That tension deserves its own analysis, and it will get one.

Compare that with the 36.6-point gap between current users and never-users.

Whether someone had used an AI companion told us far more about their answer than how old they were. The same held for interest in AI companions by generation: the age story in this survey is smaller than the experience story everywhere we looked.

This article started with what looked like the obvious dilemma: can something be cheating if the other party is not a person?

The survey suggests most respondents are not resolving it that way. Fewer than three in ten treated the AI's non-humanity as the deciding factor. The rest were asking different questions, ones that sound much more like the questions people already ask about each other.

Is my partner seeking romance outside our relationship? Are they sexually engaging with it? Have they become emotionally attached? Would I be comfortable knowing this was happening?

None of those questions requires the AI to feel anything. The AI does not have to experience the relationship the way a human would for the human partner's behavior to matter.

That may explain the strangest pattern in our survey. The respondents closest to believing an AI relationship could mean something were not the ones dismissing AI cheating. They were the ones most likely to believe the boundary could be crossed.