AI Hallucinations: How to Check AI Answers for Mistakes

AI is fast and fluent, and that is exactly what makes its mistakes dangerous: a wrong answer looks just as confident as a right one. This lesson teaches the habit that separates people who use AI well from people who get embarrassed by it.

Why AI makes things up

Remember lesson 1: the AI writes what a good answer looks like. Usually that is also what is true. But when it does not know something, it does not always say "I don't know". It may fill the gap with something that sounds right.

This is called a hallucination. Common examples:

A statistic or percentage that nobody ever published
A book, article, court case or website link that does not exist
A wrong formula or a calculation with a small error in the middle
An old rule, rate or deadline presented as current
A real person or company described with details that are not true

Warning signs

Be extra careful when the answer…	Why
Contains exact numbers, percentages or amounts	Numbers are easy to get slightly wrong
Mentions laws, tax rates, deadlines or rules	These change, and the AI may know an old version
Gives sources, links or quotes	They may be invented or misquoted
Is about something recent	It may be after the AI's knowledge cut-off
Is about a specific person or small company	There is little reliable information to learn from
Is exactly what you hoped to hear	That is when we all stop checking

The 5-step checklist

Use this before you rely on any important AI answer:

Check the numbers yourself. Recalculate totals in Excel or on a calculator. Never copy an AI-calculated figure straight into a report.
Open every source. If the AI gives a link or a reference, open it and find the exact line. If you cannot find it, treat the claim as unverified.
Go to the official source for rules. For GST, income tax, labour law or company policy, check the official government website, the official circular or your company's own document.
Ask the AI where it might be wrong. It is often good at spotting its own weak points when you ask directly:

Review your answer above. Which parts are you least sure about, and what should
I verify from an official source before relying on it?

Ask an expert for anything high-stakes. Money, legal, medical and tax decisions need a qualified person, even if the AI's answer looks perfect.

Tip: Some assistants can search the web and show their sources. That helps, but it does not remove the need to check: the AI can still misread a page.

Make mistakes less likely in the first place

Give it the facts. Paste the actual circular, invoice or data and ask questions about that, instead of asking from memory.
Allow "I don't know". Add to your prompt: "If you are not sure, say so instead of guessing."
Ask for reasoning. "Show your calculation step by step" makes errors easier to spot.

Practice task

Ask your assistant a factual question about your own field where you already know the correct answer, for example a rule, a formula or a process you use every day.

Then ask the review prompt from step 4. Compare its answer with what you know. Did it get anything wrong? Did it admit where it was unsure? Write down one thing you will always double-check from now on.

Key takeaways

AI can be confidently wrong. This is called a hallucination.
Be most careful with numbers, rules, sources and recent events.
Recalculate numbers, open every source and use official websites for rules.
Give the AI real documents to work from and allow it to say "I don't know".