AI language models generate text that sounds right, which isn't the same as being right. Occasionally they invent facts, quotes, statistics, sources or even entire products. This is often called “hallucination”. You can't eliminate it completely, but you can reduce it a lot.
Give it the source material
The most effective fix is to provide the facts yourself. Paste the document, data or article and say: “Answer only using the text below. If the answer isn't in the text, say you don't know.” This grounds the AI in real information.
Allow “I don't know”
Models are eager to help and may fill gaps with guesses. Explicitly permit uncertainty: “If you're not sure, say so and explain what you'd need to check.”
Ask for reasoning and sources
Ask the AI to explain how it reached an answer, and to list sources where the tool supports web search. Then actually open the sources: a link that doesn't exist or doesn't say what's claimed is a red flag.
Watch the high-risk areas
- Specific numbers, dates and statistics
- Quotes attributed to real people
- Legal, medical and financial details
- Citations, book titles and research papers
- Very recent events
Verify what matters
Use AI for drafting, structuring and explaining, and verify facts before publishing or acting on them. A quick check against an official or primary source takes minutes and can save real embarrassment.
A grounding prompt to reuse
Use only the information in the text below.
If something isn't covered, write "Not in source".
Quote the sentence you used for each claim.
Text: [paste source]