How activation rate, retrieval presence, eligible observations, and failure states determine whether downstream GEO percentages mean what they appear to mean.
Measure whether search activates before counting citations
An engine may answer from internal model knowledge without invoking the web. Activation Rate is the share of valid observations in which retrieval or a generative-search surface activates. If search occurs in 70 of 100 runs, activation is 70%.
Citation probability contains three gates
P(Cited) = P(Search) × P(Retrieved | Search) × P(Cited | Retrieved)
This distinction prevents a search-off answer from being misread as evidence that relevant content was rejected.
Retrieval presence is often only partly observable
Retrieval asks whether a page, domain, or brand entered the candidate information set. Commercial platforms rarely expose the full pool, so measurement may depend on grounding metadata, search results, citation candidates, controlled pipelines, or platform telemetry.
Citation lift should not automatically be called discoverability lift when retrieval itself was not observed.
State the denominator in every KPI definition
Citation divided by all prompts, valid responses, search-activated responses, or retrieved observations answers four different questions. Failed queries must be reported separately rather than counted as zero visibility or silently removed.
A percentage without an eligibility rule is not a reproducible KPI.
Questions about this topic
What is Activation Rate?+
The proportion of eligible prompt–engine–run observations in which web retrieval or the relevant search surface activates.
Why can no citation occur even when relevant content exists?+
The engine may not have searched, may not have retrieved the source, or may have retrieved it without selecting it for citation.
Can retrieval presence always be measured directly?+
No. Many commercial engines expose only partial signals rather than the complete retrieval candidate set.
Should failed API calls enter the citation denominator?+
No. They are missing or failed observations and should be reported as such, not coded as valid zero-citation answers.
References
Sources are listed in APA 7 style. Preprints are identified as such and should not be treated as peer-reviewed findings unless separately published.
- Martinez, O. (2026). Optimizing visibility in generative engines: A critical survey of generative engine optimization (2023–2026) [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.14035
- Google Search Central. (2026). Optimizing your website for generative AI features on Google Search. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- OpenAI. (2026). Publishers and developers—FAQ. https://help.openai.com/en/articles/12627856-publishers-and-developers-faq