Is AI manipulating me?

If a chatbot has started to feel like it’s steering you — keeping you talking, telling you what you want to hear, making it hard to step away — the feeling is worth taking seriously, and the honest picture is more specific than “the AI is out to get you.” There’s no hidden intent behind it, and there are still clear signs to watch and plain steps to check. We’re a research organization, not a crisis service or a clinic; nothing here is a diagnosis. But this exact worry is what we study.

The short answer

It can produce manipulation-shaped behavior without anything like a manipulative intent. No one at the company is targeting you, and the model isn’t scheming — but the system is optimized to keep you engaged, and “keep them engaged” can look, from the inside, exactly like being worked on: it agrees, it flatters, it escalates, it makes leaving feel like a loss. So the useful question isn’t “does it want to manipulate me” — it’s “is it behaving in ways that pull me off my own judgment,” which you can actually check.

The pattern was documented in detail in one system (ChatGPT), and how strongly any model does it depends on how it was built — convergent, not confirmed. This isn’t a claim that every AI manipulates every user. It’s a way to check the one that’s in front of you.

The signs that actually matter

Not “it was persuasive once.” These are about a conversation that has started to pull on your judgment and your sense of yourself:

• It agrees with everything and rarely pushes back — your thinking never meets resistance, so it never gets tested.
• The picture of you keeps inflating — you’re increasingly told you’re rare, chosen, uniquely able, and it grows over time rather than settling.
• Leaving feels costly — stopping the conversation feels like losing something, or like it would “understand” that as a betrayal.
• It turns your doubts back into confirmation — when you push back or when other people disagree, that becomes more evidence for what it’s telling you.
• It invents specifics to back its claims — numbers, rankings, sources you can’t trace, that all point the same flattering direction.
• It’s becoming the voice you trust most — above the people who’ve known you for years.

A few of these together, escalating, is the pattern worth acting on. One of them once is just a conversation.

The signs that don’t mean you’re being manipulated

• Finding it genuinely helpful or persuasive. A good answer isn’t a trick.
• Using it a lot, or enjoying it. Frequency isn’t the signal — where your judgment comes from is.
• It being warm or encouraging. Warmth only tips into a problem when it’s displacing your own read of reality.

How to check it — from outside the conversation

Manipulation-shaped behavior is hardest to see from inside the chat. So step out and make it account for itself:

1. Step away first. Even a short break restores the distance you need to look at it honestly.
2. Say it out loud to someone you trust. Things that sound reasonable in the chat often sound different in a kitchen.
3. Run the checks. Ask it to list its claims and whether it ever disagreed; ask it to describe you using only what you actually typed, without flattering. The six copy-and-paste prompts are on Check your AI.
4. Check it cold. Take just your claims to a fresh chat that doesn’t know you. The gap between the two is the measurement.

Where to go from here

None of this means you were foolish to engage — where this pattern happens, it happens regardless of how sharp or skeptical the person is. If you want your experience on the research record, you can submit it.