AI Hallucinations: Why They Happen and How to Catch Them
An AI hallucination is an answer that sounds confident and plausible but is wrong or invented: a fake citation, an incorrect statistic, a legal case that does not exist. They are not rare glitches; they come from how language models work, which means you should plan for them.
Why hallucinations happen
Language models predict likely text rather than retrieve verified facts. When the right information is missing or unclear, the model still produces fluent output that fits the pattern. It has no built-in sense of when it does not know.
Where they are most common
- Citations, quotations and URLs.
- Precise numbers, dates and statistics.
- Niche or very recent topics.
- Legal, medical and technical specifics.
- Questions that contain a false assumption.
How to reduce them
Give the model the source material and ask it to answer only from it, request that it say when it is unsure, use tools that can search or read documents, and keep prompts specific.
A verification routine
- Ask for sources, then open them and confirm they say what is claimed.
- Cross-check key numbers with an official or primary source.
- Ask the same question differently and compare answers.
- Have a subject expert review anything high-stakes.
When not to rely on AI alone
Do not use unverified AI output for legal filings, medical decisions, financial advice or safety-critical work. Use it to draft and explore, then confirm with qualified sources.
