Should I rewrite the title first?

Only when search-query and page evidence show intent or snippet mismatch. A new title cannot repair blocked access, an incorrect canonical, weak indexing, or a page with no distinct value.

How many answer-engine checks prove a problem?

No universal number does. Use a fixed question set and repeated, dated runs, report all zero results, and avoid treating one interface response as a stable engine-level rate.

Question-led guide · diagnostic

My page is not cited by AI—where should I diagnose first?

A staged diagnostic that checks access, indexing, retrieval, source selection, claim support, citation behavior, and measurement before rewriting a page.

Author: Leo J. Li
Published: Aug 25, 2026
Updated: Aug 25, 2026
Review by: Nov 25, 2026

Direct answer

Start with access, not wording. Confirm the exact canonical URL returns usable public HTML, is allowed for the relevant crawler purpose, is internally linked, and is eligible for indexing. Then test whether the page actually answers the observed question with a distinct, supported claim. Only after access, indexing, and retrieval evidence are sound should you investigate source selection and visible citation. Repeat the exact question under recorded conditions before concluding there is a pattern.

Scope

Use this for a specific public page and a specific non-branded or branded question. Record engine, interface, locale, login state, date, exact question, whether web search was visible, answer archive reference, and visible sources. Do not begin with an aggregate “AI visibility” score that hides those conditions.

Why it happens

Teams collapse several failure modes into “the model does not like our content.” The page may be blocked, canonicalized elsewhere, unindexed, irrelevant to the actual question, retrievable but weaker than competing sources, absorbed without visible citation, or never considered because the interface did not search the web. Each stage needs different evidence and action.

Diagnosis

Access: Fetch the canonical page and robots policy with the relevant user-agent class. Check HTTP status, WAF, rendering, and public assets.
Canonical and indexing: Confirm one HTTPS canonical, internal links, sitemap presence, no noindex, and available search-console evidence.
Intent: Compare the exact question with the page’s direct answer, audience, scope, and decision.
Retrievability: Identify the passage that should answer the question without relying on navigation or hidden client state.
Evidence advantage: Name the original artifact, first-party observation, bounded framework, or synthesis the page contributes.
Selection and citation: Record visible competing sources and whether your claim appears with or without attribution.
Behavior: Separate source selection from referral, resource use, book-page view, and outbound click.

Solution

Fix the earliest demonstrated failure. Repair access and canonical issues before editing prose. Merge duplicate pages before creating more variants. If intent is weak, rewrite the direct answer and scope. If the page is commodity summary, add an independently useful artifact, test, comparison, or evidence map. If selection remains volatile, continue controlled observation rather than repeatedly changing a page after every run.

Keep derivatives traceable. Medium or DEV versions should serve a different purpose and point to the canonical when supported. Multiple copies of your own claim are distribution, not independent corroboration.

Artifact

Stage	Evidence to capture	Pass or next action
Access	Status, rendered HTML, robots/WAF result, crawler purpose	Repair blocks or continue
Canonical/indexing	Canonical, sitemap, internal links, index evidence	Consolidate or continue
Intent	Exact question, direct answer, scope, target reader	Rewrite/merge or continue
Retrieval	Candidate answer passage and surrounding support	Improve information structure
Evidence	Unique artifact, source map, test, or bounded framework	Add real value, not filler
Selection	Visible sources and repeated run conditions	Observe competition and volatility
Citation	Exact supported claim and attribution	Correct unsupported or missing context
Behavior	Referral and allowed first-party events	Improve journey without inferring purchase

Common mistakes

Testing only branded questions that contain the book or publisher name.
Changing several technical and editorial variables at once.
Using a user-agent string as proof that a verified crawler can pass the CDN.
Counting self-published derivatives as independent source diversity.
Reporting only successful citations and discarding zero-result runs.

Evidence

OpenAI documents separate controls for search discovery and model-training crawlers.
OpenAI identifies OAI-SearchBot for search and GPTBot for training and states that the robots settings are independent.
Primary source · official-doc · checked Aug 25, 2026
Limit: Allowing a crawler does not guarantee a visit, indexing, source selection, citation, or traffic.
Indexing and serving are not guaranteed even when a page meets Google's technical and content guidance.
Google's guide connects generative search eligibility to crawlability and indexing and explicitly notes that indexing and serving are not guaranteed.
Primary source · official-doc · checked Aug 25, 2026
Limit: Google documentation cannot establish behavior in other answer engines.
Citation diagnosis should stop at the earliest stage contradicted by evidence.
The staged sheet separates technical access failures from content, selection, citation, and behavior observations.
Signal Studio author framework · reviewed Aug 25, 2026
Limit: Retrieval and selection are often opaque; a diagnostic stage can remain unknown rather than proven.

Limitations

Publishers cannot observe every retrieval and ranking stage inside an answer engine. A missing citation may reflect query interpretation, source diversity, freshness, interface experiments, personalization, or no web search. The method narrows possibilities; it does not guarantee a causal diagnosis.

FAQ

Should I rewrite the title first?: Only when search-query and page evidence show intent or snippet mismatch. A new title cannot repair blocked access, an incorrect canonical, weak indexing, or a page with no distinct value.
How many answer-engine checks prove a problem?: No universal number does. Use a fixed question set and repeated, dated runs, report all zero results, and avoid treating one interface response as a stable engine-level rate.

Continue within SEO and GEO, or use one of these adjacent diagnostics:

English editorial review: Codex native-English editorial review, Aug 25, 2026.