Retrieved does not mean cited

A search-enabled provider may expose retrieved URLs, grounding supports, or user-visible citations. These are different records. A visible citation is evidence that the provider surfaced a source in its answer; a retrieved URL alone does not establish this. If citation output was unavailable or incomplete, the system cannot confidently label the own domain uncited.

Answer counts and citation distributions

One answer can cite multiple domains, and a domain can occur more than once in that answer. A per-domain count of distinct citing answers is therefore not necessarily a mutually exclusive distribution that sums to 100%. A list of source shares must name whether its unit is answers, citations, or URLs. Check domain normalisation and the actual linked destination before interpreting a familiar-looking name.

Competitor evidence

A competitor recommendation gap needs an assessable own-shop negative, a supported tracked-competitor positive, and usable answer evidence. An unresolved own-shop classification is not proof of a loss. Similarly named merchants need identity review. Businesses appearing in answers can be observed entities without automatically becoming the tracked competitor cohort used by a headline comparison.

How content opportunities are grouped

Saved questions and answer evidence can be grouped by topic, intent, category, and compatible collection context. Repeated storage of the same answer must not make an opportunity look independently confirmed. A useful opportunity explains which questions exposed it, which providers and runs observed it, what the original answer said, and which merchant pages could be relevant. AI-generated wording is an interpretation of this evidence rather than an additional observation.

Matching your existing pages

A website page with related words can be a candidate destination for an improvement, but it is not proof that the page caused the observed answer. Check its actual subject, audience, product identity, locale, technical accessibility, and freshness. A missing matched page does not prove no suitable page exists; the crawl could be incomplete or the page could use different terminology.

A useful finding has three layers

Keep the measured gap, proposed explanation, and proposed action distinct. For example: 'three assessable answers recommend another shop' is an observation; 'our compatibility page may not cover this use case' is a hypothesis; 'review the original answers and add supported compatibility details' is an action. None establishes an improvement until a compatible follow-up measurement exists.

Prioritisation without fabricated demand

Purchase intent, repeated independent evidence, competitor strength, source coverage, and catalogue relevance can help order review work. They do not create a search-volume estimate, a traffic forecast, or a revenue opportunity. Keep small-sample and stale-source notices visible. A weak but actionable mapping issue may deserve attention before a large, poorly supported marketing claim.

Improve your shop after an AI visibility measurementFrom a finding to a task and a measured follow-upComparable measurements, repeated sampling, and uncertainty