# How to spot AI hallucinations

> To spot an AI hallucination, look for over-specific details, numbers that do not add up, sources you cannot open, and answers that change when you ask again. Then check one claim yourself: ask the AI how sure it is, click its sources, and make it show its working for any numbers.

Source: https://shaaz.online/guides/how-to-spot-ai-hallucinations.html · Published 2026-10-04 · Updated 2026-10-05

An AI hallucination is an answer that sounds right and is not. The tricky part: the AI sounds exactly as sure when it is wrong as when it is right.

## Why chatbots make things up

Most chatbots work by predicting the next piece of text. That works beautifully for sentences, but it means the AI will happily produce something that *looks* like a fact, a quote or a source whether or not it exists. It never says "erm". It hands you a neat, confident answer, and your brain accepts it.

## Five red flags

- **A very specific detail you did not ask for**: an exact date, a page number, a named expert.
- **Numbers that do not add up**: totals, percentages, conversions and invoices are where AI quietly drops a digit.
- **Official-sounding sources** with no way to check them.
- **A different answer when you ask again.** If the story changes, the first version was a guess.
- **Total agreement.** If it never pushes back on your idea, it is flattering you, not checking it.

## Three checks that take under a minute

1. **Ask how sure it is.** "How sure are you about that, out of ten? And what would make you wrong?" Often the answer changes.
2. **Click the source.** Open the link and find the actual line. Some links are perfect, some are real pages that do not say what the AI claimed, and some do not exist.
3. **Make it show the maths.** "Show every step, and use a calculator or code for the numbers."

These three habits are the first week of [The 30-Day AI Habit](https://shaaz.online/guides/30-day-ai-challenge.html), practised with short daily tasks and quizzes where you spot the slip hidden inside a confident answer.

## Sources

- [European Broadcasting Union and BBC: News Integrity in AI Assistants (October 2025), as reported by VRT](https://www.vrtinternational.com/news/largest-study-of-its-kind-shows-ai-assistants-misrepresent-news-content-45)

## Frequently asked questions

**What is an AI hallucination?**
A confident answer from an AI that is wrong or made up, such as an invented fact, number, quote or source. The AI does not know it is wrong.

**How can I tell if ChatGPT is making something up?**
Look for over-specific details, numbers that do not add up, sources you cannot open, answers that change when you ask again, and total agreement with whatever you say. Then check one claim yourself.

**Can AI hallucinations be prevented completely?**
No. You can reduce them with better prompts, but the reliable fix is a checking habit: ask how sure it is, click the sources and verify the numbers.