Hallucinations and fact-checking
Why AI sometimes states false things with confidence, and how to check its answers.
What is it?
A hallucination is when an AI model gives an answer that sounds confident and fluent but is wrong or made up. It can invent facts, quotes, statistics, book titles, legal cases or web links that do not exist.
This happens because language models predict likely text rather than look facts up in a verified database. When the model lacks reliable information, it may still produce a plausible-sounding answer. Fact-checking means treating AI output as a draft and confirming important claims against trustworthy sources.
Which models use it?
- All large language models can hallucinate, including the GPT, Claude, Gemini and Llama families.
- Chat assistants with web search, such as Perplexity, ChatGPT search and Gemini, can show sources you can open and check.
- Retrieval-augmented generation (RAG) systems reduce hallucinations by giving the model relevant documents to answer from.
- Image and video generators can also produce made-up details, such as wrong text, hands or logos.
How to use it
- Decide how much accuracy matters: brainstorming needs little checking, while health, legal, money or news topics need careful checking.
- Ask the AI to say when it is unsure and to separate facts from guesses.
- Ask for sources, then open each link yourself; a cited source may not exist or may not say what the AI claims.
- Check names, numbers, dates and quotes against an official or primary source.
- Give the AI the source text yourself (paste a document) and ask it to answer only from that text.
- For important decisions, confirm with a qualified human expert.
Benefits
- Knowing about hallucinations helps you use AI safely and avoid spreading mistakes.
- Simple checking habits catch most errors before they cause harm.
- Asking the AI to show uncertainty often makes its answers more careful and easier to verify.
Limits and risks
- No current model is free of hallucinations, even with search turned on.
- Hallucinations can be hard to spot because the wording sounds confident and professional.
- Citations can be fabricated or point to pages that do not support the claim.
- Fact-checking takes time, so AI does not remove the need for human review.
Prompt template
Copy it, replace the parts in [brackets] and paste it into any AI chat.
Answer the question below. Rules: 1) Only state facts you are confident about. 2) Mark anything uncertain with [UNSURE]. 3) Do not invent sources, quotes, numbers or links. 4) At the end, list the 3 claims I should verify first and where I could check them. Question: [your question] Context I already know: [optional background]
More in Safety and society
- Bias and fairnessHow AI can repeat unfair patterns from its data, and what you can do about it.
- Privacy and your dataWhat happens to what you type into AI tools, and how to protect personal information.
- Prompt injection and AI securityHow hidden instructions in content can trick AI tools, and how to stay safe.
- Deepfakes, watermarking and content credentialsHow AI-made images, voices and videos are spotted and labelled.
- Copyright and AI contentWho owns AI-made work, and why the rules are still being decided.
- AI safety and responsible useSimple habits for using AI honestly, safely and with good judgement.