Where to start with an AI SEO problem

Start with the earliest demonstrated failure

Choose one concrete symptom, preserve the evidence, and test the earliest point in the information path that could explain it. A missing mention is not automatically a crawl problem. A crawler visit is not proof of indexing or citation. A false answer needs an audit before anyone can know whether the cause is an outdated page, conflicting entity record, unsupported source, or the answer system itself.

Use this tool for the first route, not a final diagnosis. Many cases cross several pillars after the initial check.

Diagnostic decision tree

What can you demonstrate now?
├─ An important URL cannot be fetched, rendered, or indexed as intended
│  └─ Start with Crawlability, Rendering & Indexability.
├─ The visible page is correct, but its machine-readable description is invalid or mismatched
│  └─ Start with Structured Data.
├─ Systems confuse the organization, person, product, or location with another entity
│  └─ Start with Entity identity & authority.
├─ The source page lacks evidence, expert review, necessary context, or a maintained claim
│  └─ Start with Content strategy & quality.
├─ You cannot answer what appears, where, how often, or under which tested conditions
│  └─ Start with Measurement.
└─ You have a preserved false, stale, misleading, misattributed, or manipulated answer
   └─ Start with Auditing & Correcting, then route the confirmed cause to its owner.

If none of these is demonstrated, begin with Measurement. Capture the observation before changing the site so the team can distinguish a real failure from an assumption.

Choose between competing tasks

Address credible urgent harm through the appropriate response path without delaying escalation. Otherwise, resolve a demonstrated blocker to the intended outcome before adding enhancements that rely on it. When no failure is demonstrated, run one bounded diagnostic check before prescribing changes. Among independent tasks of similar consequence, prefer the one affecting more important pages with a more confident diagnosis; use effort and maintenance burden as tie-breakers.

For example, a team sees that an important product page returns a crawler challenge and also proposes adding alternate content formats. Start by preserving and testing the access failure because it blocks the intended page. Add formats later only if an identified consumer or use case supports them; recheck the repaired response before treating the access work as complete.

Route common symptoms

Three symptoms that need careful routing

“The answer engine does not mention us”

Start with Measurement, not a site change. Confirm the prompt set, product, market, conditions, and whether the result was a true absence, refusal, error, or wrong-entity match. Then investigate technical access, entity clarity, content fit, and source context only where evidence supports those branches.

A missing mention does not reveal why the entity was omitted, and no compliant site change can guarantee inclusion.

“The answer engine quotes an outdated price”

Start with Auditing & Correcting because there is a material observed claim. Preserve the answer and sources, establish the price and effective date, and classify the case as stale if the former price was once accurate.

Then trace every controlled representation: canonical product page, visible offer, JSON-LD, merchant feed, API, cached document, profile, and syndicated source. The corrective work may move through Human-Led Content, Structured Data, Entity Authority, and Crawlability, but the audit remains the record that connects the symptom to the verified cause and recheck.

“The answer engine cannot see our client-side app”

First test the premise through Crawlability. Compare the initial response, rendered DOM, network failures, server logs, index controls, and the specific crawler’s documented capabilities. Check light DOM and shadow DOM separately when web components are involved.

Do not infer inability from one missing answer. A page can render correctly and still not be retrieved, selected, used, or cited.

The 30-minute starting protocol

1. Write the symptom as one observable statement, including URL, entity, answer, or report.
2. Preserve the current evidence before making a change.
3. Choose the first route from the decision tree.
4. Perform one check that can confirm or reject that route.
5. Record the result, owner, next action, and recheck condition.

Examples of useful first checks are an HTTP response, an original-versus-rendered comparison, a deployed markup validation, a controlled-source reconciliation, a claim-and-source review, a captured prompt run, or a verified audit claim. “Improve authority” and “optimize for AI” are not verifiable first actions.

Know when to stop and escalate

Pause routine SEO work and use the applicable specialist process when evidence suggests imminent safety risk, fraud, impersonation, privacy exposure, legal rights, security compromise, regulated advice, or other high-consequence harm. Preserve only the information needed for the case and restrict access appropriately.

For ordinary cases, stop expanding the diagnosis when the next step belongs to a named owner and has a verification method. The purpose of routing is to produce accountable work, not to attach every pillar to every problem.