The Two Kinds of "No": Why 70% of People Wouldn't Use an AI Companion

The two refusals are not the same statement.

"No interest" is a description of the respondent: this is not something they want. "I prefer relationships with real people" is a comparison: an AI companion is being weighed against something and losing. The first is indifference. The second is a preference, which means the thing being rejected has at least been considered.

I expected that distinction to be cosmetic. It is not. Every other answer in the survey splits along it.

Asked whether they could imagine developing genuine romantic feelings for an AI, 94.2% said never. Asked whether they could see themselves choosing an extremely realistic AI over a human relationship, 94.1% said an AI could never replace one.

Only 0.8% currently use an AI companion app and 1.9% have used one in the past. 97.3% have never tried one.

Their view of cheating follows the same logic. Asked whether a partner's romantic or sexual relationship with an AI would be cheating, 37.6% of this group said no, because an AI is not a real person. That is the highest share of the four groups, well above the 28.3% for the sample as a whole. Only 5.6% chose the conditional answer, that it becomes cheating once a partner grows attached, against 17.1% overall.

For this group the whole subject seems to be settled by a single premise: nothing real is happening. If the AI is not a person, you cannot love it, it cannot replace anyone, and your partner cannot betray you with it.

Asked about romantic feelings for an AI, 82.7% said never, a large majority but about twelve points below the no-interest group.

The difference is in the "maybe, if the AI became realistic enough" answer: 16.1% of this group chose it, more than three times the 5.0% among those with no interest.

The replacement question shows the same pattern. 73.8% said an AI could never replace a human relationship, against 94.1% in the other group. But 19.2% said "maybe, but I would still prefer a human relationship," five times the no-interest share, and 7.0% gave one of the two affirmative answers.

They were also far more likely to have tried one. 9.4% of this group were former users, compared with 1.9% of the no-interest group. Almost nobody in either group was a current user, but one in ten of the people who "prefer real people" had used an AI companion, and nearly all of them had stopped.

That is what a preference looks like as opposed to indifference. These respondents are not saying an AI relationship is nothing. They are saying they would rather have the human version, and a meaningful minority of them concede that a realistic enough AI might change the calculation.

The no-interest group is split between the two categorical positions, either any interaction is cheating or nothing counts, with very little in between. The prefer-real-people group is spread across the middle.

Its single most common answer, at 30.5%, is that a partner's AI romance is not cheating but would make them uncomfortable. That was the top answer in none of the other three groups.

I read that as consistent with everything else about them. People who prefer real relationships have not decided that AI relationships are unreal. They have decided they would rather not have one, which leaves them uneasy rather than dismissive when a partner might. The survey did not ask why anyone chose the answer they did, so that is my interpretation of the pattern, not something respondents told us. But the pattern is hard to read any other way.

The first is why they think anyone would use an AI companion. 56.9% of the no-interest group and 56.2% of the prefer-real-people group chose having someone to talk to or feeling less lonely, almost exactly the 56.6% across the whole sample.

Even people who want nothing to do with these apps picture loneliness, not romance, as the reason others do. The stereotype of the category does not survive even among its skeptics.

The second is who they are. Both groups skew older and more female than the yes-sayers. The no-interest group is 57.3% women and 28.0% aged 62 to 80; the yes-sayers are 54.0% men and 62.0% aged 30 to 45. Among respondents aged 18 to 29, 53.8% said no interest and a further 33.7% said they prefer real people. That is the largest combined refusal of any age group under 62, and it is why the youngest adults were among the least open to AI companions in this survey.

That is the highest "not sure" share of the four groups, and it comes from the people with the least interest in the AI option.

I do not think it means they are torn. My reading is that it reflects disengagement: a respondent who has ruled AI out entirely has no reason to weigh it against anything, and "not sure" is the closest answer to "the question doesn't apply to me." Only 2.7% of them would include an AI in any form. The prefer-real-people group, by contrast, was more decisive, with 64.7% choosing dating apps outright.

If that reading is right, it is a reminder that "not sure" in a survey is not always ambivalence. Sometimes it is a shrug.

The caveat, kept visible. Age is tangled up in this one. "Not sure" was even more common among respondents aged 62 and over (48.4% of those 62 to 80), and the no-interest group is the oldest of the four. Some of that 36.7% is probably age rather than disengagement.

The practical point is that "70% wouldn't use an AI companion" hides two audiences that almost nothing would move in the same direction.

The no-interest group is not a market and does not appear to be a future one. 94% of them can't imagine feelings, can't imagine replacement, and mostly regard the whole thing as unreal.

The prefer-real-people group is different: one in six could imagine feelings if the technology improved, one in five could imagine at least considering an AI over a human, and one in ten has already tried. They are not customers. But they are the people whose answers would change first if "realistic enough" ever arrived, and they are also the people most likely to feel uneasy, rather than unbothered, if a partner got there before they did.

That last group also connects to a pattern we found elsewhere. The respondents most open to AI feelings were the ones most likely to call a partner's AI romance cheating, which I wrote about in the AI infidelity paradox. The prefer-real-people group sits one step down the same slope: not open, but not closed, and uncomfortable in exactly the way that pattern would predict.

The two "no" answers are how respondents chose to describe their refusal. The survey did not ask them to explain it, so the interpretations above, indifference versus preference, a closed door versus an open one, are readings of how the two groups answered everything else, not motives they reported. The groups also differ in age and gender, and some of the differences in their other answers may reflect that rather than the refusal itself. I have not controlled for it here.

Neither group should be described as representative of Americans who reject AI companions. The sample was balanced for age and gender but not region or income, and the 1,709 responses are the survivors of a quality screen described separately. All differences are associations in a two-day survey.

What the survey does show is that the majority position on AI companions is not one position. Seven in ten said no. Roughly two thirds of them meant it in a way that closes the question. The rest meant it in a way that keeps it open, and that is the third of the "no" vote worth watching.