Grounding
Constraining a language model's output to information retrieved from a specified source — a search index, a document set or a database —…
GEO AIConfabulation
Hallucination
A confident, fluent but factually wrong or fabricated statement produced by a language model, including invented citations, prices, features or quotes.
In full
Hallucination arises because models generate plausible continuations rather than looking facts up; grounding reduces it by supplying retrieved evidence but does not remove it, since models can still misread or over-extrapolate from context. For brands the practical exposure is misinformation at scale: wrong prices, invented policies, or attributions to the wrong company. Mitigations are stating key facts explicitly and unambiguously on your own pages, keeping third-party sources current, and monitoring answers for recurring errors.
Example
An assistant tells users a SaaS product has a 30-day free trial that was discontinued two years earlier.
Related terms
Brand mention monitoring
Tracking where a brand is named across the web and inside AI answers, including whether the description is accurate and the sentiment…
Retrieval-augmented generation
An architecture where a system retrieves relevant documents or passages first and then has a language model generate an answer conditioned…
AI visibility
The extent to which a brand appears, is cited, or is recommended across AI-generated answers, measured by sampling prompts rather than by…
AI-generated content policy
Google's position that content is judged on quality and purpose, not method of production — but mass-produced AI content made to game…