You ask ChatGPT a question, it gives you a confident, well-written answer — and the answer is simply wrong. A statistic that does not exist, a quote nobody said, a book that was never written. This is common enough to have a name: a hallucination. Understanding why it happens is the key to using AI safely, so here it is in plain language.
What 'making things up' means
A hallucination is when an AI states something false as if it were fact, with no signal that it is unsure. That last part is what makes it dangerous: there is no nervous tone, no “I think.” Wrong answers arrive with exactly the same confidence as right ones.
Why it happens at all
Here is the core idea. An AI chatbot is not looking facts up in a database. It works by predicting the most likely next words based on patterns it learned from huge amounts of text. Most of the time, the most likely words happen to be true — which is why it is useful. But when it does not actually “know” something, it does not stop; it generates the most plausible-sounding continuation, and plausible is not the same as correct. It is less like a librarian and more like an extremely well-read improviser who never wants to leave you hanging.
AI predicts likely words, it does not retrieve verified facts. That single sentence explains almost every hallucination you will ever see.
When it happens most
- Obscure or niche topics — the less it saw in training, the more it fills gaps with plausible guesses.
- Specific facts — exact dates, numbers, names, quotes, citations. These are the most likely to be invented.
- Very recent events — anything after its training cutoff, unless it can search the web.
- When pushed for an answer — if you insist, it would often rather invent something than say “I don’t know.”
How to spot it
Be most alert with suspiciously precise numbers, direct quotes, named sources, and anything you cannot easily verify. A good habit: separate the ideas in an answer (usually solid) from the specifics (most likely to be wrong), and check the specifics. Our full fact-checking guide walks through a simple routine for this.
Build AI you can trust more
The free AI Prompt Builder helps you write prompts and assistants with clear limits — no signup.
Try the AI Prompt Builder →How to reduce it
- Ask it to show its reasoning or say how confident it is — it will often flag its own shaky ground.
- Use a tool that can search the web for current or factual questions, so answers are grounded in sources.
- Give it the source material and ask it to answer only from that, rather than from memory.
- Cross-check anything important against a second source before you rely on it.
The bottom line
Hallucinations are not a bug that will be fully “fixed” soon — they come from how these models fundamentally work. That does not make AI unreliable for everyday use; it means you keep one habit: trust the help, verify the specifics. Do that, and confident-but-wrong answers stop being a problem.
Frequently asked questions
They are getting less frequent as models improve and as tools add web search, but because hallucinations stem from how these systems predict text, it is wise to assume any AI can still make things up. The habit of verifying specifics remains important.
No. There is no intent. It cannot always tell the difference between a fact it learned and a pattern that merely looks like one, so it generates the most plausible-sounding text — which is sometimes false.
All major assistants can hallucinate, and the rate shifts as they update. Tools that can search the web tend to do better on factual questions. The verifying habit matters more than the choice of tool.
Yes, as a starting point — for finding angles, structure, and explanations. Just verify names, numbers, quotes, and sources independently before you rely on them, exactly as you would with any assistant.