TL;DR
- Access, indexing, retrieval and visible citations are different observations. A brand mention is not necessarily a citation or recommendation.
- Check whether each linked source supports the nearby claim; distinguish owned pages from third-party sources.
- Save the exact question, answer, source URLs and test conditions. Investigate possible causes and repeat the check before drawing a conclusion.
Separate observations before explaining causes.
Accessible means a page could be reached under a particular test condition. Indexed means it entered a specified index. Retrieved means information was obtained for a task; external observers may not have enough evidence to confirm that step. Cited means the final answer visibly attributes a source. These are analytical distinctions, not a universal pipeline every product follows.
Mentioned means the brand appears in the text. Recommended requires a rule that the brand is presented as suitable for the stated need. A link to an independent comparison is a third-party citation, even when it mentions you; it is not an owned-domain citation.
Check the cited passage, not just the domain. A link may be real but fail to support the adjacent claim. Label that as an unsupported attribution rather than counting it as a valid endorsement.
Questions and product modes change the evidence.
Google describes query fan-out: its AI search features may run related searches on subtopics. It also notes that models and links can differ across features. Google’s AI feature explanation. That documented example does not reveal the actual subqueries used by another platform.
Save the exact question, platform, mode and time before comparing sources. A follow-up question in a long conversation is not the same test as a fresh session. One favorable answer is not a stable position across a market.
Crawler rules answer only part of the question.
OpenAI documents OAI-SearchBot for search and GPTBot for potential model training, with independent controls. ChatGPT-User handles certain user-triggered visits; it is not the automatic search crawler or the setting that controls search inclusion. OpenAI crawler documentation. Allowing training is therefore not a prerequisite for search participation.
Inspect the relevant policy and hosting behavior together. A robots rule does not tell you whether a specific request met a challenge, reached the expected HTML or used an index built earlier. Investigate with the site owner; do not replace working crawler rules with a generic list.
Avoid blanket claims about JavaScript. A direct fetch, rendered browser visit and indexed representation can expose different content. Record the product and access mode you actually tested.
A citation diagnostic you can reuse.
Use possible explanations as hypotheses. Each row identifies the next check and the limit of that check. Multiple explanations may fit the same symptom.
| Observation | Possible explanation | What to check | What this does not prove |
|---|---|---|---|
| A test gets 403 or a challenge | Hosting policy, rate limits or request context | Status, response body, legitimate request details and owner-accessible logs | That every platform request is blocked |
| Page opens, retrieval is unknown | Accessible does not mean selected for this question | Relevant index evidence where available; repeat questions and inspect source links | That the platform has retrieved this page |
| Brand appears without a source link | Mention without visible attribution | Save exact wording; label mention and recommendation separately | A citation to your site or evidence of a visit |
| A third-party page is cited | That source may address the question or supply context | Read its passage, date and brand references | That your site is technically broken |
| Link does not support the nearby claim | Citation placement or answer synthesis may be wrong | Match each claim to the actual linked text | That a valid URL makes the claim accurate |
| An old source creates a brand error | Outdated material remains available or influential | Approved current fact, source dates, owned copies and correction channels | That editing one page forces an immediate answer update |
| Citation appears once but not again | Output variability or changed test conditions | Repeat the same documented prompt and retain every result | A permanent rank, gain or loss |
Investigate an incorrect claim.
Illustrative example: an answer cites an old distributor page to describe a discontinued service as current. Save the full answer, exact prompt, platform and mode, language, market if known, timestamp and cited URL. Preserve the relevant source passage and the approved current service record.
Confirm the conflict with the business owner. Update outdated owned pages and request a correction from the third party through an authorized contact. Record what was changed and when. Do not claim the external correction is complete until you can read it back.
Repeat the same observation after a suitable review interval and report both the source status and answer status. If the answer remains wrong, retain that outcome. An inaccessible source should be marked unavailable, not automatically classified as false.
Turn source checks into a useful report.
Report supported, unsupported and unverified citations separately from brand mentions. Keep the saved evidence and the classification rule so another reviewer can reproduce the judgment. Use the measurement guide to define the denominator.
Markup alone does not settle source selection; see the schema evidence comparison for the relevant research boundaries. When the source path is unclear, review the scope of an AI visibility audit with the question and saved answer ready.