LIVING DOCUMENT — UPDATED PERIODICALLY
last updated: 15/08/2026
Concepts A–Z
- Answer shaping: (Groundbase concept) The technique whereby content not only provides an answer, but also defines the criteria and decision framework on which the answer is based. Whoever sets the criteria indirectly steers the conclusion.
- Citation rate: The percentage of AI answers, measured across a defined prompt set, with an explicit source reference to your domain. Not to be confused with the mention rate — your brand named in the answer text; the two do not necessarily move together.
- Defensive GEO: The set of strategies aimed at monitoring and correcting incorrect or unwanted AI representations of a brand. See Cluster 9.
- E-E-A-T: Experience, Expertise, Authoritativeness, Trustworthiness. Google’s editorial quality framework from the Search guidelines. Useful in GEO as a quality bar; that other AI systems use it as a signal has not been demonstrated.
- Entity: A named concept — person, organisation, product, location or notion — that is uniquely identifiable. AI systems work strongly on an entity basis.
- Entity salience: How prominently and consistently an entity appears in content. High salience goes hand in hand with better recognisability in practice; a guaranteed effect it is not.
- Fragility Index: (Groundbase concept) The degree to which your position in AI answers is stable or vulnerable to small prompt variations. High Fragility means your visibility depends on chance phrasings.
- GEO (Generative Engine Optimisation): The practice of measuring and improving how generative AI systems find, mention, describe and recommend a brand — through on-site and off-site strategies. See 1.1.
- Grounding: Anchoring AI output to specific retrieved documents or facts, reducing hallucination.
- Hallucination: An AI system generating factually incorrect information with seemingly high confidence. The reason factual, well-structured content performs better.
- Inclusion threshold: (Groundbase concept) The observable point at which a brand appears consistently enough in open category questions to be in the game. A measurement concept — not a demonstrated internal model component. See Cluster 8.
- LLM (Large Language Model): A large language model trained on extensive text corpora. The engine behind generative AI systems.
- Mental availability in AI: The degree to which a brand is present in the ‘mental model’ of an AI system — independent of active retrieval. A brand-strategy analogy with brand awareness — useful as a compass, not a cognitive equivalent.
- Narrative Share: (Groundbase concept) The amount of answer space dedicated to your brand — not just whether you are mentioned, but how extensively and how centrally. See sub-page 4.5.
- Off-site GEO: GEO optimisation through external platforms: Wikipedia, reviews, sector publications, comparison sites. See sub-page 3.5.
- Prompt intent: The underlying goal behind a user query: informational, navigational, transactional or comparative.
- RAG (Retrieval Augmented Generation): A technique in which an AI retrieves documents — from an index, cache or the live web — before generating an answer.
- Schema.org: Standardised vocabulary for structured data on web pages. Makes the nature of content explicit for search engines and parsers; not a ticket into AI answers.
- Share of Voice (SoV) in AI: The share of your brand mentions relative to all measured brands, across a set of benchmark prompts (share of mentions). Always report the absolute mention rate alongside it. See 4.2.
- Training data: The text corpora on which an LLM is trained, with a fixed cut-off date.
- Volatility Score: (Groundbase concept) The degree to which the position in AI answers varies on repeated measurements of identical prompts. High Volatility indicates unstable visibility.
- YMYL (Your Money Your Life): Category of content with direct impact on financial wellbeing or safety. Google applies stricter quality criteria here; for other AI systems this is not documented, though high factual reliability is wise there regardless.
- Zero-click: An interaction in which the user gets their answer without clicking through. AI answers increase the share of zero-click interactions — but often do contain sources and links.