How to Catch AI When It's Making Things Up

Free guide by AI with Zana · Works in ChatGPT, Claude and Gemini · 8-minute read

AI can give you a wrong answer in the same confident voice as a right one. That's called a hallucination: when the model fills a gap with something that sounds true. A made-up statistic in your sales page or a fake "study" in your post can cost you trust fast.

Here's why it happens, where it's most likely, and five habits that catch it.

Why AI makes things up

AI doesn't look facts up in a database by default. It predicts the most likely next words. Most of the time that's accurate. But when it doesn't know, it doesn't always say so. It produces something that fits the pattern of a correct answer.

Where to be extra careful

• Numbers and statistics, especially very precise ones ("73.4% of buyers…")
• Quotes, sources and links
• Recent news, prices and product features (things change after the AI was trained)
• Legal, medical and tax details
• Niche topics with little information online

5 habits that catch it

1. Give it permission to say "I don't know"

If you're not sure about something, say so clearly instead of guessing.
Mark anything you're less than confident about with [CHECK].

2. Ask for sources, then open them

For every factual claim, give me the source. If you can't point to a
real source, say "no source" instead.

Then click the links. If a link doesn't exist or says something else, treat the claim as false. Use tools with web search turned on for anything recent.

3. Give it the facts yourself

AI is much more reliable when it works from your material instead of its memory.

Using ONLY the document below, answer my question.
If the answer isn't in the document, say "not in the document".
[paste document]

4. Run the fact-check pass

After you get a draft, open a new chat and paste it in:

Act as a strict fact-checker. List every factual claim in this text.
For each one: is it likely true, likely false, or unverifiable?
Explain which ones I must check before publishing.
[paste text]

A fresh chat doesn't "defend" what it wrote earlier, so it's more willing to spot problems.

5. Ask the same question twice, differently

If two differently worded questions give you different answers, you've found a weak spot. Verify it yourself.

The simple rule

Use AI for thinking, drafting and structure. Verify anything that's a fact, number, name, date or price before it goes public. Thirty seconds of checking protects years of trust.

For more practical AI guides, follow @aiwithzana.