Methodology
GEO MethodologyHow AI search visibility is measured
Definition
AI search visibility is the presence and representation of an organization in answers generated by AI systems, observed over a defined query set, under stated conditions. It is an observation of a sample, not a property of the organization.
Query selection
Queries are chosen with the organization before measurement and grouped by intent: definitional, comparative, decision and navigational. They include questions real customers ask. The full query set is part of the result and is provided with every report.
Sampling and repeated runs
AI answers vary between runs. Each query is sampled several times, and results are reported as the share of runs, never as a single observation. Sample sizes are stated with every figure.
AI systems tested
Each report names every AI system sampled, including the product surface used (for example an assistant with web access, or an AI overview within search results). Systems change without notice, so results are tied to the dates sampled.
Date, time, locale and language
Every run records date and time, locale, language and whether the session was signed out or personalised. Results from different conditions are not combined.
Metric definitions
| Measure | Definition |
|---|---|
| Mention | The answer names the organization. |
| Citation | The answer references one of the organization's pages as a source. |
| Recommendation | The answer suggests the organization for the task asked. |
| Entity accuracy | The share of factual statements about the organization that are correct. |
| Source visibility | Which sources are cited when the organization is mentioned. |
| Competitor visibility | The same measures for named competitors on the same queries. |
Mention, citation and recommendation are reported separately because they can move independently.
Limitations
- AI output is not deterministic; repeated runs reduce but do not remove variance.
- Systems change without notice; a change in results may reflect the system, not the organization.
- Personalisation and conversation context are not fully observable.
- Samples cover the stated queries only, not all possible questions.
Methodology versions
Methodology v1.0, published 1 October 2026 alongside GEO Standard v1.0. Changes will be versioned and listed in the changelog.
Research references
- Aggarwal, P. et al. (2023). GEO: Generative Engine Optimization. arXiv:2311.09735
- Schema.org vocabulary: schema.org
- Google Search Central, Introduction to structured data
Measurement principles
Six rules every GEO measurement follows.