THE SHORT ANSWER
An assistant may retrieve candidates from its own index, a search provider or user-supplied material, then select and present supporting sources according to a system that is usually not public. Relevance, clarity, specificity, evidence, freshness, access and source authority can make information more useful, but they are editorial and retrieval qualities—not a guaranteed list of citation factors.
State what the observation cannot reveal
Seeing one page linked for one prompt shows that it appeared in that response. It does not show the full candidate set, weighting, retrieval provider or reason for selection. Responses can change across repeated questions and interfaces.
Google publishes some information about its AI Search features, including query fan-out and supporting links, but not a complete selection algorithm. OpenAI documents publisher access and referral behavior rather than a citation-ranking formula.
Evidence & context: Google Search Central · OpenAI Help Center
Build qualities that help a source do its job
| Quality | Practical evidence | Caution |
|---|---|---|
| Relevant | Directly resolves the current question | Broad topical similarity is not enough |
| Clear | Claim, unit and conditions are explicit | Clarity cannot repair falsehood |
| Specific | Original detail, method or data is present | Precision should be supportable |
| Authoritative | Responsible author or organization and provenance | A byline alone proves little |
| Current | Time-sensitive facts show maintenance | Newer is not always better |
| Accessible | Important content can be fetched and parsed | Access does not ensure selection |
Observe citations without converting them into factors
- Define a small set of important questions and contexts.
- Record interface, date, locale, wording and visible sources.
- Check whether the response represents each source accurately.
- Compare repeated observations and relevant search results.
- Use gaps to improve facts or structure, not to reverse-engineer a secret score.
A sample can support a content decision or detect a harmful error. It cannot establish market share or causal uplift unless the method supports those claims.
Give the system and reader a reason to retain the source
Primary data, a transparent method, an expert explanation, a maintained specification or a well-scoped comparison can add information that a generic summary cannot replace. Make the contribution identifiable and independently checkable.
Sources & further reading
- AI features and your website
Google Search Central. Official documentation for AI Overviews and AI Mode checked 13 September 2026. Product behavior, availability and reporting can change; eligibility never guarantees crawling, indexing or appearance.
- Optimizing your website for generative AI features on Google Search
Google Search Central. Official Google guidance checked 13 September 2026. It treats AEO and GEO work for Google as part of SEO, recommends established search foundations and unique people-first information, and does not describe other assistants' systems.
- Publishers and Developers — FAQ
OpenAI Help Center. Official publisher guidance checked 13 September 2026. It distinguishes OAI-SearchBot discovery from GPTBot training controls and documents ChatGPT referral tagging; crawler roles can change.
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Lewis and colleagues, 2020. Research combining generation with retrieved material. Retrieval should not be interpreted as proof of factual accuracy.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗