Glossary

What is Hallucination?

When an AI model states false information with confidence — including false claims about your brand, pricing, or features.

Definition

A hallucination is a confident, fluent, false statement from a language model: an invented statistic, a feature you don't have, a pricing tier you discontinued, a competitor's capability attributed to you. It arises because models generate plausible text from statistical patterns, not verified facts.

For brands, hallucinations are a reputational surface: assistants describe your product to buyers whether or not the description is true.

Why it matters

Weak entities get hallucinated about most — when models lack consistent, corroborated information about a brand, they fill gaps with plausible fiction. The defense is unglamorous: accurate, consistent, machine-readable facts everywhere (site, schema, reviews, directories), plus routine monitoring of what assistants say about you so errors get caught and countered with authoritative content.

Frequently asked

An AI is saying something false about my product. What can I do?

Publish an authoritative, crawlable correction (product pages, FAQ, changelog), align third-party listings, and where available use platform feedback channels. Grounded assistants pick up corrected pages quickly; training-memory errors fade with model updates.

Why do models hallucinate about small brands more?

Sparse, conflicting signals. The less corroborated data exists about you, the more the model interpolates — one more reason entity consistency is a visibility strategy.

Related terms