AI Hallucinations: Why Chatbots Make Things Up and How to Catch Them
AI chatbots sometimes invent facts, quotes and sources with total confidence. Learn why these hallucinations happen, where they're most likely, and a simple routine to check AI answers.
Information current as of November 2026 · How we research · Report an error
Ask an AI chatbot for a list of books on a topic, and it might include one that doesn’t exist. Ask for a legal case, and it might invent one with a realistic name and date. These errors are called AI hallucinations, and they’re one of the most important things to understand before you rely on AI for anything that matters.
What is an AI hallucination?
A hallucination is when an AI system produces information that sounds plausible but is false or unsupported, and presents it as if it were true. Common examples include:
- Made-up statistics or dates.
- Quotes that a person never said.
- Citations to articles, studies or web pages that don’t exist.
- Wrong steps in a technical procedure.
- Incorrect summaries of a document, adding details that aren’t in it.
The dangerous part is the tone. A hallucinated answer usually reads exactly like a correct one: fluent, confident and well formatted.
Why do chatbots make things up?
They generate likely text, not verified facts
Large language models are trained to produce the most plausible continuation of a text. Most of the time, plausible and true overlap. But when the model lacks reliable information, it can still produce something that looks like a correct answer, because that’s what it was trained to do.
Gaps in training data
If a topic is rare, very recent, or very specific (a small local business, an obscure research paper, last week’s news), the model may have seen little or nothing about it. Rather than staying silent, it may fill the gap.
Pressure to give an answer
Models are trained to be helpful. Developers work hard to teach them to say “I don’t know,” and modern models are much better at it, but the tendency to provide some answer hasn’t disappeared.
Ambiguous or leading questions
If you ask “Why did the company announce its merger in 2019?” when no merger happened, the question itself suggests a false fact. A model may go along with it instead of correcting you.
Where hallucinations are most likely
Be especially careful with:
- Citations and sources: titles, authors, links, page numbers.
- Numbers: statistics, prices, dates, measurements.
- Niche topics: specialized technical, legal or medical details.
- Recent events after the model’s training cutoff, unless it uses web search.
- People: biographies, especially of people who aren’t widely known.
- Long summaries, where small invented details can slip in.
How to catch hallucinations: a simple routine
1. Ask for sources, then open them
If the chatbot can browse the web, ask it to cite sources. Then actually click them and check that they exist and say what the AI claims. A link that doesn’t work or a page that doesn’t mention the claim is a red flag.
2. Check the important facts independently
For anything you’ll act on, confirm key facts with an official or reputable source: a government site, the manufacturer’s documentation, a reputable news organization or a professional.
3. Ask the same question differently
Rephrase your question or ask in a new chat. If the answers disagree on a key fact, treat that fact as uncertain.
4. Ask the AI to check itself
Prompts like “Which parts of your answer are you least sure about?” or “List any claims here that I should verify” often surface weak spots. This isn’t proof of accuracy, but it’s a useful filter.
5. Give it the source material
When you need answers about a specific document, paste the text or upload the file and ask it to answer only from that document and to quote the relevant passage. This greatly reduces invented details, though it doesn’t eliminate them.
6. Watch for overly specific details
Suspiciously precise numbers, perfect-sounding quotes and neat lists of references deserve a second look, especially if you didn’t provide them.
How to write prompts that reduce hallucinations
- Say “If you’re not sure, say so” explicitly.
- Ask for reasoning step by step on calculations and logic.
- Specify the time period you’re asking about.
- Avoid assuming facts in your question; ask “Did X happen?” before “Why did X happen?”
When not to rely on AI alone
AI can be a great starting point, but don’t let it be the final word for medical decisions, legal matters, financial or tax decisions, safety-critical tasks, or anything you’ll publish under your name without checking.
The bottom line
Chatbots hallucinate because they’re built to generate plausible text, not to guarantee truth. They’re improving, and tools like web search and document upload help, but the responsibility to verify is still yours. Treat AI answers as a well-written first draft: useful, fast, and always worth checking where it matters.
New to AI? Our beginner’s guide to AI explains how it works, what it’s good at, and how to use it safely.