AI learning guide
Check AI answers before trusting them
Learn why a convincing answer can still be wrong and how to verify its claims.
By SuperAIGuy.com · AI-assisted educational content · Updated
An AI hallucination is a plausible statement that is false or unsupported. Confidence, polished wording, and a detailed citation do not establish truth. OpenAI's research describes how training and evaluation incentives can encourage guessing when information is missing. Anthropic's guidance also explains that uncertainty and grounding techniques can reduce hallucinations without eliminating them.
Use a claim-by-claim check
- Extract the statements that matter to your decision.
- Open the cited sources and check that they exist.
- Find the actual passage supporting each statement.
- Check the date, definitions, population, units, and limitations.
- Mark unsupported claims and resolve conflicting evidence.
Make uncertainty visible
Ask the assistant to separate evidence from inference and say when the supplied material cannot answer the question. These instructions can help, but the model can still make mistakes. Treat generated confidence percentages as uncalibrated unless their meaning has been evaluated.
Practice
Ask for a summary of a short document you know well. Compare every factual statement with the original. Count supported statements, unsupported statements, and appropriate statements of uncertainty separately.
For consequential decisions, use authoritative material and qualified human judgment. This site's chat is a learning tool, and its generated answers are not independently verified.
Put it into practice
Sources to check
- Reduce hallucinations · Anthropic · Link checked
- Why language models hallucinate · OpenAI Research · Link checked
Check dates record source-link reviews. These are source-based explanations prepared with AI assistance; no independent expert review is claimed. Read our method and limitations.