How do I stop the AI from making up facts?
Set the rules up front
Models generate plausible text, not verified facts, so they will sometimes invent names, dates, quotes, and citations. This is often called hallucination, and no prompt eliminates it.
Tell the model explicitly: 'Use only the information below. If the answer is not there, say you don't know.' Then paste the source text into the prompt. Grounding the answer in supplied material cuts down invention a lot.
Ask for confidence levels or for the model to flag uncertain claims. This does not guarantee accuracy, but it makes guessing easier to spot.
Verify before you use it
Treat every specific fact, number, quote, and citation as unverified until you check it against a reliable source. This matters most for medical, legal, financial, and academic work.
If the model gives a source, open it. Fake citations that look real are common. When accuracy is critical, use tools designed to search live sources rather than relying on the model's memory.
For high-stakes decisions, have a qualified human review the output. A prompt is a writing aid, not a fact-checker.
- Paste the source text and say 'answer only from this.'
- Ask the model to say 'I don't know' when unsure.
- Request a list of assumptions behind the answer.
- Check every number, name, date, and quote independently.
- Use search-enabled tools for current or niche facts.
Common mistakes
- Trusting a confident tone; fluency and accuracy are not the same thing.
- Asking for citations without checking that the cited sources actually exist.
- Using a chatbot as the only source for legal, medical, or financial decisions.
