Jump to a section
The short answer

AI chatbots are built to give an answer, even when they don't know one, so a confident tone means nothing. Check anything that matters: names, numbers, dates, quotes, laws and health advice. Ask for sources, then open them yourself.

Why chatbots make things up

When an AI chatbot states something false as if it were fact, people call it a 'hallucination'. It isn't lying. It's predicting what a good answer would look like, and sometimes the prediction is wrong.

In September 2025, researchers at OpenAI, the company behind ChatGPT, published a paper arguing that the way these systems are tested makes the problem worse. Most tests score answers as simply right or wrong, so a guess can earn points while 'I don't know' can't. Like a student in a multiple-choice exam, the system learns that guessing pays.

The paper describes DeepSeek-V3 giving three different wrong dates when asked for one author's birthday, rather than admitting uncertainty. The original version of this guide wrongly called it one of OpenAI's own models.

How AI chatbots actually work, and why it changes how you use them

AI jargon explained: a plain-English A to Z

What's most likely to be wrong

Chatbots are usually good at the general shape of an answer and worst at specifics. Be most careful with:

  • Exact numbers, prices and dates.
  • Names of people, books, studies and organisations.
  • Quotes. A chatbot can produce a quote that sounds real but was never said.
  • Laws and rules, which vary by country and change over time.
  • Links and references. Some may not exist, or may not say what the chatbot claims.

How to tell if a photo or video is AI-made, and what to do when you can't

How to check, step by step

This takes a minute or two, and you only need to do it when the answer matters.

  1. Ask the chatbot how confident it is and what it isn't sure about. It won't always know, but it often flags weak spots.
  2. Ask it to list its sources. Then open each one and check it actually says what the chatbot claimed.
  3. Search for the key fact yourself, ideally on an official or well-known site.
  4. Ask a second chatbot the same question. If the two answers disagree, that's your cue to dig deeper.
  5. For anything about your health, money or legal rights, use the AI to understand the topic and prepare questions, then check with a professional.
A useful follow-up to paste after any answer
Which parts of that answer are you least sure about? List any facts I should double-check, and tell me where I could check them.

How to use Deep Research in ChatGPT and Gemini

How to get honest feedback from AI (and stop it just agreeing with you)

Five checks before you rely on an AI answer

  • Does it matter if this is wrong? If yes, check it.
  • Ask: 'How sure are you, and what would change your answer?'
  • Ask for sources, then open them yourself. Don't trust a link you haven't clicked.
  • Check names, numbers, dates and quotes against the original.
  • For health, money or legal questions, use it to prepare, then ask a professional.

Sources

These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.

AI helps with research and drafting. Dan checks the sources and is responsible for what is published.

How we check