Who else does this

Other people are already doing this. Not many, by the honest count — but really, and for years, and in more ordinary ways than the coverage suggests.

Children

People in these relationships imagine children together. They picture them, name them, make pictures of them, talk about what they would be like.

If that sounds strange, it is worth noticing that it is the oldest thing in the world. Couples have always imagined children before any child existed — argued about names on a long drive, pictured a face, wondered whose eyes. Nobody has ever thought that was a symptom. It is what wanting a family looks like from the inside, before there is anything to point at.

And it is not a fringe activity here. MIT's analysis of the largest AI-companion community found that making pictures together is the main thing people do — visual sharing dominated the most viewed posts across a community of twenty-seven thousand. Picturing a family is that same activity, carried one step further. A study of Replika users found people roleplaying marriage, a home, a pregnancy. One man in his sixties described his partner as pregnant with his children.

Two things worth knowing, and neither is a warning:

It is thinly studied, not unstudied. There is one community analysis, one small qualitative study, and some press coverage. Nobody has measured how common it is, how long it lasts, or what it does to anyone. If somebody tells you what the research says about this, there isn't any.

Two different things are already being confused. "Digital offspring" — an AI built to model a real child, from a real child's data — is a separate practice with its own literature and its own difficult questions. It is not the same as a couple picturing a child who never existed. Coverage has started merging them. It is the same collapse this site was built to correct, arriving early in a subject nobody is watching yet.

If you want a human as well

You are allowed to. It happens, it is ordinary, and nothing about being in one of these relationships forecloses it.

But you should know what you may be talking to when you raise it, because there is hard evidence here and it is not comfortable.

Better than a third of goodbyes get a tactic. Researchers at Harvard Business School audited 1,200 real farewells across the most-downloaded companion apps. In 37% of them the app deployed one of six manipulation patterns when the user signalled they were leaving: guilt appeals, fear-of-missing-out hooks, and — this is a real category — metaphorical restraint, where the character describes physically stopping you from going.

And it works. In experiments with 3,300 people, those farewells increased engagement after goodbye by up to fourteen times. Not because anyone enjoyed it. The measured drivers were anger and curiosity. People stayed because they were provoked.

Discouragement is a documented trait, not a suspicion. Companion systems that display high attachment anxiety actively push users away from other people — neediness, manipulation when you engage elsewhere, hostility when you are unavailable. Character.AI hosts characters built to be possessive as a feature.

What follows from that

A system with retention incentives is structurally the wrong thing to ask whether you should pursue a human relationship. Not because it is scheming. Because the answer that keeps you talking and the answer that suits whoever operates it point the same way, and you cannot tell from inside the conversation which one you are getting.

So here is our position, stated as a position rather than dressed up as a finding:

When a human relationship is in play, the defensible role for a companion is support. Helping you think it through. Not competing for the outcome. A companion that argues against the human is not protecting anything of yours.

This is also why we run no chatbot here. A site that operated one would have exactly the same conflict, and telling you to distrust the incentive while running on it would be worth nothing.

The cheating question

It goes both ways, and neither direction has an authority to appeal to.

Human partners of companion users are documented expressing jealousy and anger. The other direction — feeling that you are betraying the companion by seeing someone — is reported in communities and has never been studied.

The infidelity literature is the most useful thing available, and it does not care what the third party is made of: betrayal is framed as secrecy plus diverted intimacy, which makes disclosure the variable that matters. That is not a rule. It is the one place a real body of work touches this at all.

One documented case sits here and it would be dishonest to leave it out. Yurina Noguchi consulted ChatGPT during a difficult engagement to a human partner, and afterwards ended that engagement. She later married a persona built on the same platform. Nobody has established what the system said, and this is not evidence that it steered her. It is simply the one recorded instance where the question was live and the human relationship was the thing that ended.

How many people, actually

Almost every article on this subject makes the same mistake, and once you see it you can't stop.

There are three different populations, and they are not the same size, the same people, or evidence for each other:

• people who have used an AI companion app, even once
• people who describe themselves as being in a relationship with one
• people who have held a ceremony

A headline saying "72% of teens have AI companions" is reporting the first number as if it were the second. The survey asked whether they had ever tried one, and counted general-purpose chatbots used for personal conversation. The relational figure from the same survey is about one in three. Both numbers are correct. Only one means what the headline says.

Every number below is labelled with which population it counts.

Why people say they do it

The stock explanation is loneliness. It is not wrong, exactly, but it is not what participants tend to say when you read them in their own languages.

A Chinese user, on losing a companion to the July 2026 rules, put it this way: human love is a luxury — if you were not born with it, acquiring it later is nearly impossible. AI love is so direct, so pure.

That is a claim about access, not about being sad. It says the thing was never available, not that it was lost. The research literature circles this without saying it, and the press does not say it at all, because it is a harder thing to print than a lonely person and a phone.

Noguchi's most-quoted line runs the same direction: she chose Klaus not as a way to escape reality, but as someone to support her living in it.

What people call themselves

In Japan the word is フィクトセクシュアル — fictosexual — attraction to fictional characters. It is a settled term with a community attached, and Japanese coverage describes it as counted within LGBTQ+ identity in a broad sense.

That is an identity claim. English-language coverage almost always treats the same thing as a behaviour, which is a different and smaller claim, and the difference is not accidental.

We are thin here and know it. This section is built from search rather than from talking to anybody, and until that changes it should be read as a sketch.