What is a hallucination?
A hallucination is AI output that sounds plausible and confident but is false or unsupported by any source. OpenAI defines hallucinations as plausible but false statements generated by language models, and NIST's generative AI profile uses the term confabulation for the same risk. Common forms are invented figures, made-up quotes, wrong dates and citations to documents that do not exist.
Why it matters to a revenue leader
A hallucination can end up in a forecast note, a customer email or a board slide, and its confident tone makes it easy to miss. OpenAI's research argues that standard training and evaluation reward guessing over admitting uncertainty, so the risk will not simply disappear with the next model. The defense is design. Anthropic's guidance includes letting the model say it does not know, requiring a citation for each claim and limiting answers to the documents provided, and a person should still review anything customer-facing.
Example
Fictitious numbersA VP of Sales asks an AI tool for last quarter's renewal rate by branch. The tool answers 94% for the northern branch, but the CRM shows 87%, and the tool had no access to renewal data at all. The fix is to connect the tool to the renewal records and require it to show where every number came from.
"When our AI tools give us a number, can we trace it to a record, and what do they do when they cannot find one?"
Sources
- Why language models hallucinate | OpenAI checked 9 Oct 2026
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) checked 9 Oct 2026
- Reduce hallucinations | Claude Platform Docs checked 9 Oct 2026