How candidate sources are found, compared, and filtered before a generative model ever sees them.
Retrieval builds the candidate set
After a system has interpreted the request, retrieval selects potentially useful documents or passages from its available information sources. This is an upstream constraint: a page that never enters the candidate set cannot shape a retrieval-grounded answer.
The foundational GEO experiment used a fixed set of top search results, which created strong evidence about post-retrieval content effects. It did not test whether a rewritten page would be organically crawled, indexed, and retrieved in a live product.
Reranking decides which candidates remain competitive
Retrieval may return more material than a model can practically use. A later ranking stage can reprioritize candidates using relevance, freshness, source quality, language, user context, and proprietary signals.
Commercial reranking functions are largely hidden, making GEO a partially observable optimization problem. A document may be successfully retrieved yet rank too low to reach the final context.
Relevance is relational, not absolute
A page is not simply “high quality” in isolation. It competes against other sources for a particular reformulated query, user context, and moment in time. Clear topical focus, current evidence, distinctive information, and credible support can help a candidate remain useful during selection.
Retrieval means the system found the source. Reranking determines whether it survives.
Questions about this topic
What is a candidate set?+
It is the collection of documents or passages retrieved as potentially relevant before later filtering and context selection.
Is retrieval the same as appearing in the final answer?+
No. A retrieved source can be reranked below the cutoff, receive little context, influence the answer without a citation, or be omitted during generation.
Can we see a commercial engine’s reranking score?+
Usually not. Practitioners infer behavior from controlled tests, visible sources, repeated outputs, and other observational proxies.
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.
- 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
- 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