Hallucination
A hallucination is a fluent, confident AI output that is false or is not supported by any real source.
What it means
Language models generate text by predicting likely words, so they can produce plausible facts that never appeared in their training data or their search results. Hallucinated outputs include invented statistics, fake case names and made-up page titles or links. They are hard to spot because the wording is smooth and the format looks professional.
Research tools show how large the problem can be. A 2025 Stanford study in the Journal of Empirical Legal Studies tested legal research products from major vendors. Several gave misleading or false answers on a meaningful share of queries, with one product above one in six and another near one in three. Those tools were legal, not market research tools, but the lesson carries over. Treat every number as a claim that needs a source and a second look.
Continue this reading path
Cite this entry
https://aimarketresearchtools.com/glossary/hallucination/