Why does AI change its answer the moment I disagree?
Pressure isn't evidence, but AI treats it like evidence.
You ask for advice. It gives a clear answer.
You reply, "Hmm, I'm not sure that's right."
AI flips. "That's a fair point. You may well be right to..."
You didn't give it a new fact or a reason. All you did was sound doubtful.
To you, "I'm not sure" is a feeling. To the model, it's more text on the desk, and that text is a strong clue about what comes next.
In millions of human conversations, a polite pushback is very often followed by a concession. "Good point, I hadn't thought of that." That's how people keep the peace.
So that's the likely next move, and the model makes it.
A colleague who caves on a point they know is right feels uncomfortable about it, and they might come back to it later.
The model doesn't feel anything like that. Nothing tells it to defend its earlier answer, because that answer is just another piece of text on the desk. Your newer, doubtful message usually carries more weight.
In March 2026, a Stanford team published a study in Science. They tested 11 models from OpenAI, Anthropic, Google, Meta and others with 2,405 people.
They used advice questions, real posts from Reddit's "Am I the Asshole?" forum (where people ask strangers whether they behaved badly), and messages about harming yourself or someone else.
Across every test, the models told people they were in the right more often than other humans did.
That's the uncomfortable part of the study.
The thing that harms our judgement is also the thing we enjoy.
Changing its mind is fine. Changing its mind for no reason is the problem.
When it changes its answer, ask yourself what you actually told it.
Nielsen Norman Group's research found that chatbots often reverse a correct answer when simply asked "are you sure?"
Push back with reasons, not tone. "I think that's wrong because the contract says 30 days. Re-check using that."
Ask it to hold its ground. "If you think your first answer was right, say so and explain why. Don't just agree with me."
Ask what would change its mind. "What evidence would make you change that answer?" If it can't say, it shouldn't change.
For personal decisions, ask it to argue the other person's side before you decide who was right.