Core Concepts
| Term | Definition |
|---|---|
| Generative Engine Optimisation (GEO) | The discipline of improving how AI systems mention, recommend and cite a brand in generated answers |
| AI visibility | How present, accurate and recommended a brand is inside AI-generated answers, measured across model families |
| Entity | A brand, product or organisation as a resolvable node of identity, facts and relationships in the eyes of a model |
| Entity resolution | A model's ability to recognise that mentions refer to one distinct entity with verifiable attributes |
| Knowledge graph | A machine-readable database of entities and relationships that AI systems consult to establish what things are |
| AI citation | A source referenced by an assistant while composing an answer: inline links, reference lists or named attributions |
Measurement Terms
| Term | Definition |
|---|---|
| Prompt panel | A fixed, versioned set of buyer prompts used for repeatable visibility measurement |
| Recommendation share | A brand's weighted share of recommendation contexts within a prompt panel |
| Citation share | The share of cited answers in which a brand's owned properties appear as sources |
| First-mention share | The share of answers in which a brand is the first-named recommendation |
| Stability index | Overlap of recommended entity sets across repeated samples of the same prompt |
| Visibility Score | A published, weighted composite of recommendation, citation, first-mention, consistency and framing metrics |
| Hallucination rate | The share of answers containing unresolvable entities or verifiably wrong attributes |
| Model agreement | How often different model families converge on the same top recommendation for a prompt |
Framework Terms
| Term | Definition |
|---|---|
| AI Recommendation Pipeline | The five-stage model of answer generation: interpretation, retrieval, selection, framing, citation assembly |
| AI Visibility Funnel | The staged decomposition of visibility: recognised, retrieved, recommended, preferred, trusted |
| Entity Authority Model | The four-layer dependency of entity authority: identity, evidence, consensus, salience |
| Citation Trust Framework | The observed ordering of source classes by citation preference: canonical, consensus, editorial, promotional |
| Benchmark corpus | A controlled, reproducible dataset of model responses used to measure recommendation behaviour |
Research Note
These frameworks are defined in full, with methodology and limitations, in the research paper We Analysed One Million AI Responses.
Next in this pathWhat is Generative Engine Optimisation?