How to spot AI hallucinations: 6 quick checks
What a ChatGPT or Gemini hallucination is, which details it gets wrong most according to OpenAI, and how to check an answer in two minutes with sources and search.
A hallucination is when AI tells you something false as if it were obvious. It’s not rare: OpenAI explains in its help center that, like any language model, ChatGPT can produce incorrect or misleading outputs, sometimes sounding very confident.
Where it fails most (according to OpenAI)
- Incorrect definitions, dates or facts.
- Fabricated quotes, studies or citations, even non-existent sources.
- Overconfident answers to ambiguous or complex questions.
- Recent information: the model learns up to a cutoff; without search it doesn’t know what happened after.
- Oversimplifying or presenting one view as absolute truth.
6 quick checks
- Mark the specifics. Underline every figure, date, name, law or link. That’s what needs checking.
- Ask for sources. In ChatGPT with search on, answers cite websites; in Gemini, Google says a Sources button appears when available (Gemini help).
- Open every link. Check it exists and says the same. A made-up link or one that doesn’t mention the fact is a red flag.
- Rephrase. Ask the same thing differently or in another chatbot. If the figure changes, don’t use it.
- Make it doubt. Use this page’s prompt: it forces a split between what’s solid and what’s uncertain.
- Go to the official source for health, money, tax or law. Google says not to use Gemini for medical, legal or financial advice.
Golden rule
Use AI as a first draft, not a final source. That’s what OpenAI itself recommends.
Key concept: a model’s bias can tilt its answers; always review them with your own judgement.
Frequently asked questions
What exactly is a hallucination?
OpenAI defines it as a response that isn’t factually accurate, such as wrong definitions, dates or facts, fabricated quotes or studies, or references to non-existent sources.
If the AI sounds very sure, is it more reliable?
No. OpenAI warns the model can express high confidence even when it’s wrong.
Does web search fix it?
It helps, because answers come with cited sources, but OpenAI recommends visiting the links and checking them when accuracy matters.
Glossary · Safety and good use
Keep reading
Sources
Change log
- : First published with OpenAI and Google official help as checked that day.
Educational content, prepared with the help of AI tools, with sources cited. It is not a substitute for professional advice. Tools change fast: if you spot something out of date, write to us at contacto@cognenza.com. How we work.