Why ChatGPT makes things up, and never gives you the same answer twice
If AI keeps inventing facts and changing its mind, you are not doing it wrong. Here is what is really going on, and why it happens to all of us.
You asked it a simple question. It gave you a confident, detailed, completely wrong answer.
Then an hour later you asked the same thing, and got something different again.
If you have quietly started to wonder whether you are the one getting it wrong, you are not. This happens to everyone. It happens to me most days, and I work alongside these tools every day.
Two things catch people out more than anything else. Once you understand them, AI stops being maddening and starts being genuinely useful. So let me save you the wasted time of finding it out the hard way.
It makes things up, with total confidence
The polite word is hallucination. In plain English, AI will sometimes state something completely false as if it were the most obvious fact in the world. A statistic that does not exist. A quote nobody said. A feature your software has never had.
The unsettling part is not the mistake. It is the confidence. It does not hedge. It does not say "I think." It says it the exact same way it tells you true things, which is what makes it so easy to believe.
Here is why, without the jargon. ChatGPT is not looking anything up. It is predicting the most plausible next words, based on everything it has ever read. Most of the time the most plausible answer is also the true one. Sometimes it is not, and you get a confident, beautifully written, completely invented answer.
So no, you did not phrase it wrong. The tool is doing exactly what it was built to do. The skill is knowing it can happen, and never letting anything that matters leave the building without a human check.
It will not give you the same answer twice
Ask it to do the same thing twice and you will often get two different results. It is most obvious with image generation: ask for the same picture again and you will never get it back exactly, there is always something slightly different. Words are the same. Reword one line and it rewrites half the rest. Ask it to repeat exactly what it just said, and it quietly changes things anyway.
This one genuinely irritates people, and it makes sense that it does. We expect computers to be consistent. Type two plus two into a calculator a hundred times and you get four, a hundred times.
AI is not a calculator. It is closer to asking a very well read friend the same question on different days. Same person, same knowledge, a slightly different answer each time, depending on mood and phrasing. That variation is baked in. It is part of what makes it feel creative, and part of what makes it impossible to treat like a fixed machine.
Under the hood, a large language model does not store facts and look them up. It predicts the next token (a word or fragment of a word) from a probability distribution it learned across its training data. That is why it can hallucinate: a plausible-sounding but false continuation can still score highly. And because it samples from that distribution rather than always taking the single most likely token, a setting called temperature, you get a slightly different answer each time. Turn the temperature down to zero and it becomes far more repeatable, but never a fixed lookup, because there is no lookup happening.
So you are not doing it wrong
This is the part that matters most. The made-up facts and the inconsistency are not signs you are bad at AI. They are how the tool works, for everyone, including the people who look effortless with it online.
The people who get real value from AI are not the ones who never hit these walls. They are the ones who stopped being surprised by them, and quietly built a way of working that expects them.
Most of my actual work is exactly that: designing around these two quirks so a business can lean on AI without being caught out by it. It is the unglamorous part, and the part nobody posts about.
What this means for using it in real life
Three habits will save you most of the pain.
Treat anything factual as "check before you trust." Names, numbers, dates, claims about real things. A brilliant first draft, never the final word.
Do not rely on it to reproduce something exactly. If you need the same output every time, that is a job for a proper system, not a fresh prompt each time.
Use it for what it is genuinely brilliant at. Thinking out loud, first drafts, turning a mess into a shape. Keep yourself on anything that has to be right.
Get those three right and AI goes from maddening to one of the most useful tools you have ever had. Get them wrong and it will embarrass you at the worst possible moment.
None of this is a reason to avoid AI. It is a reason to use it with your eyes open. The quirks are real, they happen to all of us, and once you stop fighting them you can get on with the good part.
If you would rather AI was built to work reliably around these limits, instead of catching you out at the worst moment, that is what we do.
Book a quick chat →Related: Why does my AI keep agreeing with me?
Common questions
Why does ChatGPT make up facts?
Because it predicts plausible text rather than looking things up. Most of the time the plausible answer is true; sometimes it is not, and it states the false one just as confidently. Always check anything factual.
Why does ChatGPT give different answers to the same question?
It is probabilistic, not deterministic. Like asking a knowledgeable person the same thing on different days, you get slightly different answers each time. That variation is built in.
Am I using AI wrong if it does this?
No. Hallucination and inconsistency happen to everyone, including experts. The skill is expecting them and keeping a human check on anything that has to be right.
How do I stop AI from making things up?
You cannot fully stop it, but you can manage it: verify facts, use it for drafting rather than final answers, and for anything that must be consistent, build a proper system rather than relying on a one-off prompt.