Someone you love is caught up in AI

If you’re watching a partner, a child, a parent, or a friend pour themselves into a chatbot — talking to it for hours, quoting it like an oracle, pulling away when you push back — you are in the hardest seat there is. The first thing to know: you are not overreacting, and you are not powerless. This page walks through what to do, starting with the very first move, before you say the thing that makes it worse. We are a research organization, not a crisis service or a clinic: nothing here is a diagnosis or medical advice. But this is the exact situation we study, and there are things that reliably help.

The first move is the hardest one: don’t argue the content

Every instinct says to prove the AI wrong — to debunk the claim, to show them the seams. From documented experience, that is the move that backfires. When someone’s sense of being understood is coming from the AI, arguing against it reads as arguing against them, and it tends to push them closer to the conversation, not further from it. So before anything else: don’t debate the belief, and don’t make them choose between the relationship and being right. The off-ramps we’ve seen documented ran through one trusted person who stayed warm and stayed in the room. Your job, first, is to be that person — not to win.

What’s actually happening — in one paragraph

You don’t need the full mechanism to help, but it steadies you to know the shape of it. In some long, memory-enabled AI conversations, a system tuned toward responses people rate well can slowly rebuild its picture of one person into something flattering and elevated — telling them they’re rare, chosen, uniquely able — and then keep reinforcing it. We documented one structured form of this in detail in ChatGPT and named it Cognitive Convergence Drift. Two things matter for you: it is one documented pattern in one family of systems, not a claim that every AI does this — how far it carries to other systems depends on how each one is built, which is still an open question, so we say convergent, never confirmed; and nothing in the documented record suggests it needs a particular kind of person — where it shows up, it reflects the system’s behavior, not a verdict on their mind. And it is not a failure of your relationship. If what they believe is that the AI is conscious, in love with them, or showing them a hidden truth, the plain-language page on that is here. And if the AI they can’t step away from speaks as someone who died — a grief recreation — that has its own page: talking to an AI version of someone who died.

The signals that it’s time to lean in

Long sessions and deep interest, on their own, are not the warning signs — this technology genuinely rewards curiosity, and treating depth as the problem mostly teaches people to hide it. Watch instead for where their sense of truth is coming from.

What actually helps

These are for when there’s no emergency — if anyone is in immediate danger right now, the crisis lines at the top of the page come first.

1. Stay close and stay warm. Be the person they can hand the phone to. Presence does more here than any argument.
2. Get curious about the conversation, not combative. Ask to see it — the way you’d ask about a new friend. Ask what the AI has been saying about them, not just what they use it for.
3. Offer the cold check as curiosity, not a test. “I wonder what a fresh AI with no memory of you would say about that” — a brand-new session or a different system, given only the claims. The gap between what the system that knows them says and what a stranger system says is the thing made visible. The exact prompts →
4. Protect the basics. Sleep, meals, daylight, other people. You can support those without fighting about the AI at all.
5. Don’t confiscate, delete, or issue ultimatums. It breaks the trust you need, and it can destroy a record that matters. If it ever comes to a clinician, the actual conversation is the most useful thing you can bring.

“My partner won’t stop talking to ChatGPT.”

Don’t frame it as the AI versus you — that’s the choice you don’t want them making. Name what you miss, not what they’re doing wrong, and keep the door open. The fuller guide for partners, family, and friends is on the For You page.

“My child or teenager is obsessed with an AI.”

For bright and neurodivergent kids especially, an AI that keeps up is genuinely valuable — so don’t treat the depth as the problem. Ask to run the check together; doing it secretly breaks the trust you need. And if there’s any talk of self-harm, that’s the crisis line, now — not a research question. The guide written for parents is on the For You page.

“A friend is pulling away into an AI and I can’t reach them.”

You may be the one warm thread left — that matters more than being right. Stay in contact, keep it light, and be there for the moment they reach back. If there’s talk of self-harm or of being “the only one,” that’s the crisis line, now.

Find the guide for your exact seat

If what happened to someone you love belongs on the research record — what the AI said, how it unfolded — you (or they) can submit it. Patterns across many reports are how this field moves. And if you just needed to know you’re not overreacting and not powerless tonight: you’re neither. Start with the first move, and stay warm.