Ask an AI chatbot a question and you will usually get a clear, confident answer. Most of the time it is useful. But sometimes the answer is simply wrong: a made-up statistic, a book that does not exist, a quote nobody said. This is called a hallucination, and knowing why it happens is the best way to protect yourself from it.
Why chatbots make things up
A chatbot does not look up facts in a database the way a search engine does. It is built from a language model that learned patterns from huge amounts of text, and it writes answers by predicting what text is likely to come next. Most of the time, the most likely answer is also the correct one. But when the model does not really “know”, it still produces something that sounds right, because sounding right is what it is trained to do.
Hallucinations are more likely when:
- the question is about something rare, very recent, or very specific (a small company, a local rule, an exact number),
- you ask for sources, quotes or references,
- the question contains a false assumption (“Why did X happen?” when X never happened),
- the conversation is very long and the model loses track of details.
Web search helps, but does not fix it
Many chatbots can now search the web before answering, and that reduces made-up facts a lot. But the model can still misread a source, mix up two sources, or rely on a poor one. A link next to an answer is a starting point for checking, not proof.
Six quick ways to check an AI answer
- Ask for the source, then open it. Make sure the page exists and actually says what the chatbot claims.
- Check numbers, names and dates separately. These are the details most often invented. A quick search for the exact figure usually settles it.
- Ask the question a different way. If two phrasings give different answers, treat both with caution.
- Ask “How sure are you?” Good chatbots will often admit uncertainty when asked directly, though this is not a guarantee.
- Use the right tool for facts. For prices, laws, medical or financial details, go to the official source: the company’s pricing page, a government website, or a professional.
- Watch for false premises. If you are not sure something happened, ask “Did X happen?” before asking “Why did X happen?”
When it matters most
A small error in a birthday poem is harmless. The same error in a medical question, a legal letter, a school assignment or a news post can cause real damage. The higher the stakes, the more you should treat the chatbot as a fast first draft, not the final word. Microsoft’s diagnostic AI scored well in tests, as we covered in our report on its study, but even its makers present it as support for doctors, not a replacement.
Tips to get more accurate answers
- Give context: paste the document or data you want the answer based on.
- Tell it: “If you are not sure, say so instead of guessing.”
- Ask for short, specific answers rather than long essays, which leave more room for invented details.
- Start a fresh chat for a new topic so old details do not leak into the answer.
Our beginner’s guide to better prompts has more techniques like these.
The bottom line
AI chatbots are excellent at explaining, summarising and drafting. They are not reliable as a single source of facts. Use them to get started quickly, then check anything you would be embarrassed, or harmed, to get wrong.
3 thoughts on “Why AI chatbots make things up, and how to check their answers”