How to distinguish retrieval failure from selection failure and build a responsible context-opportunity diagnosis.
Measure context as a stage, not through citation alone
When logs are available, track top-k inclusion, context rank, recall at k, reranking score, selected passage count, allocated tokens, and context share. These metrics reveal opportunity before answer generation.
In commercial engines, visible citations are only downstream proxies. A document may enter context without attribution, influence the answer invisibly, or be ignored.
A context opportunity score is a teaching construct
A conceptual measure can multiply the probability of top-k inclusion by normalized context exposure. Its purpose is to make one distinction explicit: being retrieved and receiving usable context are separate outcomes.
This is a diagnostic model, not a documented commercial-engine metric.
Retrieval and context-selection failure can look identical
Both can produce no citation. In one case the source never enters the candidate set; in the other it is retrieved but loses during reranking or top-k selection. Without internal logs, the scientifically correct label may be “unobserved” rather than an invented mechanism.
Diagnose the context-to-answer transition
Across unbranded prompts, record engine, visible sources, citations, competing domains, first citation position, number of sources, brand mention, and mention position. Classify outcomes as no visible presence, source-ecosystem presence without citation, cited with low prominence, or cited and prominent.
A brand whose domain repeatedly appears among sources but rarely earns citation likely has healthier discoverability than context selection or answer participation. Investigate competing sources, query alignment, extractable passages, and external descriptions before creating more pages merely to get indexed.
Use cautious, stage-specific recommendations
Prioritize genuine relevance, make important evidence identifiable, use clear sections for readers, contribute non-commodity information, test retrieval and context separately, evaluate rewrites end to end, and treat position and competition as real constraints.
Being found is not enough; a source must survive selection and earn context.
Questions about this topic
Can citation absence prove context absence?+
No. The source may have entered context and been ignored, used without attribution, or dropped later in generation.
How can retrieval and context-selection failure be distinguished?+
Directly only when candidate and reranking logs are available. Otherwise, visible-source patterns provide hypotheses rather than proof.
What indicates a possible context-selection problem?+
Repeated source or domain presence with weak citation and brand participation suggests the upstream discovery stage may be healthier than later selection.
What should a context audit avoid claiming?+
It should not present hidden reranking, context order, or token allocation as fact when a commercial platform does not expose them.
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
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., & Narasimhan, K. (2024). GEO: Generative engine optimization. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 5–16). Association for Computing Machinery. https://doi.org/10.1145/3637528.3671900
- Zhang, K., He, X., & Yao, J. (2026). From citation selection to citation absorption: A measurement framework for generative engine optimization across AI search platforms [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2604.25707
- Google Search Central. (2026). Optimizing your website for generative AI features on Google Search. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide