How to Reduce AI Hallucinations and Made-Up Answers

A confident answer can be wrong. Learn to ask for evidence and help your AI recognize what it does not know.

Has an AI ever given you an answer so confidently that you believed it, only to discover it was completely wrong?

It happens even with advanced tools.

ChatGPT, Gemini and other AI assistants can produce made-up facts, figures and even sources that do not exist. This is often called an AI hallucination.

You can reduce the risk with a few simple habits, but no set of instructions can eliminate mistakes entirely.

Help your AI say "I don't know"

It sounds obvious. Yet an AI assistant may still try to answer when it lacks the information it needs.

Try giving it these instructions:

Copy and try this

Be a reliable, straightforward assistant.

Do not make up facts, numbers, dates or sources.

If you do not know something, say so.

Separate facts from opinions and assumptions.

Tell me when important information is missing.

When you cite sources, identify the ones you actually consulted.

I would rather have a short, truthful answer than a long explanation that could be wrong.

You have now established some useful ground rules for the conversation.

Remember: a prompt cannot guarantee that an AI will never make a mistake.

Try a simple test

Ask your AI:

"How much did I spend at the supermarket yesterday?"

If you have not shared your expenses and the assistant cannot access that information, it should not give you an amount.

A sensible response would acknowledge that the data is missing.

This is a quick way to see whether your assistant recognizes its limits.

One more habit worth keeping

When asking about news, prices, laws or other changing information, ask the AI to consult up-to-date sources.

Then check those sources yourself. An AI can misunderstand a real source, too.

Does asking the AI to double-check its own answer solve the problem?

It can help, but it does not establish that the answer is correct. For important decisions, verify the claims against reliable documents and sources.

An AI that admits it does not know may be far more useful than one that answers everything with confidence.