LLM hallucination · Fabrication
A hallucination is a false statement produced with confidence by an LLM. It is frequent when the model answers from its parametric memory without grounding, or when retrieved sources are poor or contradictory.
Standalone definition · 33 words
Why does an LLM hallucinate?
A model generates the most probable text, not the most true. With no retrieved source, it completes from what it learned, with gaps and a cutoff date. A literature survey (Ji et al., 2022) distinguishes intrinsic hallucinations, which contradict the source, from extrinsic ones, which cannot be verified from it.
What is the risk for a brand?
An LLM can attribute an inaccurate product, price or policy to a brand. The risk is higher for poorly documented brands. Regularly audit what engines say about the brand, and fix it at the source when an ambiguous or outdated page is to blame.
How do you reduce hallucinations about your brand?
Publish factual, dated, unambiguous pages, identify the entity (Organization, sameAs, Wikidata), harmonize information across third-party sources and remove outdated or contradictory content. You do not control the model, you reduce sources of error.
Frequently asked questions
Does grounding eliminate hallucinations?
It reduces them, but the model can misread a source.
Can I ask an AI engine for a correction?
Some providers offer reporting channels. Nothing guarantees a correction.
How do I detect a hallucination about my brand?
Test a set of prompts about the brand and compare the answers with verified facts.
Related terms
Further reading
- Checklist GEO and LLM / AI 2026
- 2026 GEO Tools Comparison: The Decision Guide
- ← Back to the GEO glossary (114 terms)