THE DIRECT ANSWER

An AI hallucination is an output presented as factual or grounded that is unsupported or incorrect. It can include invented citations, names or details. Fluency and confidence are poor substitutes for checking evidence, especially when an answer will guide an important action.

Why fluency misleads

Language models generate text based on learned patterns and current context. They can produce a convincing name, citation or explanation without a reliable basis for that particular claim. The problem is especially easy to miss when the output uses familiar professional language.

Make the evidence visible

For source-based work, ask the assistant to identify the passage behind each important conclusion. Open the sources yourself when the claim matters. A citation marker alone is insufficient: the linked page may exist without supporting the sentence. Separate direct evidence from interpretation and mark missing information explicitly.

Test the uncomfortable questions

Include questions with no answer in the supplied material, ambiguous dates and conflicting documents. Reward a clear statement of uncertainty when it is appropriate. For example, a policy assistant should explain that it lacks the relevant regional policy rather than adapting an unrelated one. Improvements may involve better retrieval, clearer instructions or a different workflow. No single wording trick removes the need to verify consequential claims against dependable evidence.

A worked example

Ask a source-bound assistant about a policy detail absent from its documents. A useful response states that the material does not answer the question. A plausible invented rule is a failure even if it sounds consistent with the rest of the policy. Add such unanswerable cases to evaluations instead of testing only questions with obvious answers.

COMMON QUESTIONS

Questions readers ask

Can a confident answer still be a hallucination?

Yes. Presentation does not establish support. Verify important claims against appropriate sources and check whether the system had access to the required information.

SOURCES & NOTES

Follow the evidence.

AI-assisted explanation with linked technical references. Examples are illustrative learning exercises, not reported customer deployments or experimental results.

  1. OpenAI: Optimizing model accuracy

Public sources support the factual background. Analysis and illustrative exercises are identified in the text. How we assess AI · Editorial standards

Spot a factual error or an outdated detail?