Technical SEO and AI Readiness: A Practical Website Checklist

Check access before adding more content

Choose one important service page. Open its public URL without a login, check the response, and confirm that the important service information is present. Then inspect what a search crawler can access. A page that looks correct to its owner may still be blocked or point at the wrong canonical URL.

Use the traffic diagnostic guide if a previously useful page has lost visits. That investigation should happen before blaming an AI interface or replacing the whole site.

A practical acceptance checklist

Record the URL, date, finding and correction for each check. “Pass” means the agreed condition was verified; it does not mean the page will rank.

Worked example: the page loads, but the journey fails

Suppose a hypothetical consultation page returns HTTP 200 and says “serving North Idaho.” Its form requires a postcode but rejects the nearby town named in the service copy. The page passes the access check and fails the service-fit and inquiry checks. Adding schema would not resolve that contradiction.

The correction is to have the owner approve coverage, publish it consistently and make the form apply the same rule. Test an in-area case, an out-of-area case and a validation error locally. Only perform a delivery test when the receiving team has authorized it.

Search understanding is not agent execution

The connected business model identifies the provider and offer. An action interface must also specify inputs, validation, limits and a truthful result. A public description is documentation, not a permission grant.

For a request endpoint, test retries, changed input, delivery failures and human handoff. State whether duplicate protection is persistent or limited to a particular store or time window. An idempotency key is not proof of the customer’s identity or consent. The Luxe case study separates implementation, reported historical behavior and what was not independently verified.

Put AI-specific files in proportion

Google’s current guidance does not require special AI text files or special schema for its generative search features. Its ordinary SEO foundations still matter. Keep optional discovery files only when they serve a documented consumer or maintenance purpose.

Google: optimizing for generative AI search

A mention is a test observation, not a technical certificate

If an assistant names the business, save the question, date, answer and cited URLs. Check whether it got the service, area and next step right. Repeat the exercise with different realistic needs, including one the business should decline.

This absorbs the useful lesson from “legacy visibility” claims: a favorable answer may be sourced from a directory or an old page. It does not tell you that your own site is clear, that every assistant sees it, or that a request path works. Fix the facts you can identify rather than buying a promise to own a permanent AI position.

Keep the checks after launch

A successful launch does not freeze the business. Review the affected pages when a service, credential, form provider or domain changes. If an integration fails, the site should still offer an honest route to a person.

A website planning brief makes those responsibilities explicit. Ask for test evidence tied to the agreed scope, rather than a dashboard score presented as proof of new business.