Guides · Claude & ChatGPT
When AI gets it wrong: how to spot and fix made-up answers
Why AI sometimes makes things up, the warning signs to look for, and five simple habits that keep you safe.
Every AI tool gets things wrong sometimes. That’s not a reason to avoid it, just a reason to use it wisely. Here’s how.
Why does it happen?
AI writes answers by predicting what words are likely to come next, based on everything it has learned. Most of the time that produces accurate, helpful answers. But when it doesn’t know something, it doesn’t always say so. It may fill the gap with something that sounds right.
This is called a hallucination. It’s become much less common as the tools have improved, but it still happens.
When to be most careful
Mistakes are more likely with:
- Specific facts and figures: dates, statistics, prices, quotes
- Very recent events, unless the AI searched the web
- Niche or local information: your council’s bin day, a small company’s policies
- References: book titles, legal cases, web links
- Maths with lots of steps
- Anything about specific people, especially less famous ones
The warning signs
- Very specific details you didn’t ask for and can’t verify
- Links or references that don’t work when you try them
- An answer that changes when you ask the same question again
- It agrees with whatever you suggest, even when you’re testing it
Five habits that keep you safe
1. Ask it to show its sources
Where does that information come from? Give me links I can check.
Then actually click them. Both ChatGPT and Claude can search the web and show where answers came from.
2. Ask how sure it is
How confident are you in that? What parts might be wrong?
Good AI tools will tell you honestly when they’re unsure.
3. Check what matters with a trusted source For health, check the NHS website. For money, benefits and rights, try GOV.UK, Citizens Advice or MoneyHelper. For anything legal, speak to a professional.
4. Give it the facts yourself AI is much more accurate when working from information you provide. Instead of “what are the terms of my phone contract?”, upload the contract and ask about that.
5. You’re the editor Treat AI’s work like a draft from a helpful junior colleague. It’s usually good and often great, but you check it before it goes out with your name on it.
Fixing it in the moment
If you spot a mistake, just say so:
That's not right. The meeting was on Tuesday, not Thursday.
Please correct it and check the rest for similar errors.
AI doesn’t mind being corrected, and it’ll usually fix things straight away.
The bottom line
AI is a brilliant assistant, not an oracle. Use it for drafting, explaining and thinking, and keep your own judgement switched on. That combination is where AI shines.