When a page is not visible in an AI-search experience, teams often jump to a special schema or a new content batch. The constraint may be more ordinary: robots, rendering, noindex, canonical choice, orphaned links, unclear intent, or unsupported claims. Troubleshoot the layers in order and keep absence of one citation separate from crawlability.
1. Define the observation
Record query, audience, market, language, device, date, page, search experience, and evidence. Separate not cited, not retrieved, not indexed, not eligible, and not useful. State whether the decision is repair, consolidate, create, monitor, or hold.
Google’s AI features guidance says existing foundational SEO practices remain relevant and there are no additional technical requirements for eligibility. Use that to reject a shortcut promise.
Keep an observation log with the exact query, wording, market, language, device, date, and result type. AI-search outputs can vary, so a single absent citation is a weak basis for a site-wide change. Look for a repeated pattern across a defined query set and compare it with the pages that actually answer those questions.
2. Check crawler access
Test robots, CDN, WAF, authentication, status, redirects, rate limits, JavaScript, and server errors. Compare source HTML, rendered DOM, browser view, and crawler inspection. Record time, version, host, and response.
Check whether answer-bearing text is available without interaction, hidden in an image, loaded after a failure, or blocked by a component. A page that loads for a human may still be incomplete for a crawler.
Capture the response headers and the rendered text for a representative template, not just one URL. Check alternate hosts, trailing-slash variants, language paths, and parameters that may expose a different response. Escalate infrastructure findings with the exact timestamp and request path so an engineer can reproduce them without relying on a screenshot.
3. Check indexability and snippet eligibility
Inspect noindex, canonical, redirect chain, duplicate variants, sitemap, language, and selected URL. Keep indexing, snippet eligibility, and citation selection as separate fields. More pages do not repair an index directive or a selected canonical elsewhere.
Use the Search Console Performance report to observe queries, pages, devices, countries, clicks, impressions, CTR, and position within its scope. It does not prove AI selection or buyer intent.
4. Check canonical and duplicate intent
Build a URL matrix with page job, query family, audience, title, evidence, links, canonical, owner, status, and template. Mark exact, near, support, obsolete, and unknown relationships. Decide which URL should answer the question before editing tags.
Google’s canonicalization documentation treats declared canonical as a hint. Compare declared and selected states and preserve a decision note.
Ask whether the duplicate pages differ in audience, transaction, evidence, or next action. If they do not, consolidate deliberately and preserve the strongest URL. If they do, give each page a distinct job and link them as a useful sequence. A canonical tag cannot compensate for two pages that make the same promise while competing for the same question.
5. Check internal discovery
Trace hub, service, proof, comparison, article, navigation, breadcrumb, sitemap, and contextual links. Record anchor, depth, orphan status, and route to the next action. A page in a spreadsheet is not a discoverable page.
Repair links and information architecture before creating a near-duplicate. Use a supporting role when a page adds evidence but should not compete for the main query.
Check link persistence after template changes and content pruning. Record which navigation, hub, breadcrumb, and contextual links are contractual and which are optional. A page can remain technically indexable yet become practically undiscoverable when its only internal link is removed during a redesign.
6. Check intent and evidence
Write the question, audience, stage, decision, claim, source, date, limit, example, and CTA. Remove generic paragraphs, invented results, unsupported capability, and ambiguous terminology. Add original evidence and a clear boundary.
People-first content guidance is a useful quality test, but no editorial guidance guarantees indexing or citation. Require subject review for specialist, regulated, or product claims.
For each important claim, store a source, date, reviewer, confidence, qualification, and expiry trigger. Add a short explanation of method or scope where a reader could otherwise overgeneralize. This creates material that can be checked and quoted accurately instead of copy that merely contains the target phrase.
7. Check measurement and handoff
Record crawl/index state, page version, query cohort, interaction, consent, source, CRM record, accepted quality, and mature outcome. Mark unknown and attribution limits. Do not call a page view, event, or cited link revenue.
If the page has a commercial CTA, test routing, response, capacity, and retirement. A crawlable page that produces an unowned request is not operationally ready.
8. Run a bounded repair
Choose one page or intent group. Snapshot technical state, canonical, links, evidence, metrics, route, and owner. Repair one class: access, indexability, canonical, links, intent, evidence, or CTA. Recheck after a reasonable crawl and business-maturity window.
Stop if the page duplicates another job, evidence is unavailable, ownership is absent, privacy is unclear, or the proposed control cannot be tested. Preserve version and rollback.
Choose the repair owner before deployment and define the observation window. Technical changes may need a crawl interval; commercial changes may need a sales-cycle interval. Keep a before-and-after record of status, canonical, links, text, query set, route, and mature outcome so that a negative result is still informative.
9. Apply the crawlability gate
| Gate | Required evidence | Hold if | | — | — | — | | observation | query, page, date, experience, decision | one answer is called a site-wide fault | | crawl | status, robots, rendered text, infrastructure | browser view is the only evidence | | index | noindex, canonical, sitemap, eligibility | more pages mask a directive problem | | intent | question, audience, distinct job | page repeats an existing URL | | links | hub, contextual, breadcrumb, depth | page is orphaned | | evidence | source, date, limits, reviewer | claims are generic or unsupported | | operations | CTA, owner, SLA, capacity, retirement | traffic has no workable route |
The troubleshooting is complete when the team can locate the break, repair one bounded layer, and state what remains unknown. Crawlability is a precondition for discovery, not a promise of AI-search selection.
How did this article land?
Choose one reaction. You can change it anytime.
