This is a twenty-five point audit for making a site legible to AI search, ordered so that each stage gates the next. Working through it out of order is the most common way teams waste effort, because formatting a page for extraction achieves nothing if that page is not retrievable in the first place.
Each point is written so you can answer it yes or no about your own site today. Where a point needs more explanation than a line allows, it links to the guide covering it. Expect the first section to be less interesting and more valuable than the last.
Key takeaways
- Work the sections in order: retrieval gates extraction, and extraction gates corroboration.
- Most sites fail on the foundations rather than on anything specific to AI search.
- Twelve of the twenty-five points are conventional technical SEO, which is the honest picture.
- Fix the pages already ranking between roughly position three and fifteen first.
- Measure citation separately from ranking, because rank tracking does not capture it.
Foundations: can you be retrieved at all
Nothing downstream matters if a page cannot be fetched, indexed, and judged relevant. These eight points are ordinary technical SEO, and if any of them fails you should stop here and fix it before reading further. This is the least exciting section and the one that most often explains a total absence from AI answers.
Work through them against the specific pages you want cited rather than the site in aggregate. A site can be broadly healthy while the twenty pages you actually care about are canonicalised away, orphaned, or blocked, and aggregate metrics hide exactly that.
- 1. Every page you want cited returns a 200 and is present in the index
- 2. robots.txt does not block the crawlers you intend to be visible to
- 3. Canonical tags point at the URL you actually want ranked, not a variant
- 4. The site responds quickly and reliably to bot traffic, not only to browsers
- 5. No important page is orphaned; each has at least one internal link into it
- 6. Duplicate and near-duplicate pages have been consolidated or differentiated
- 7. The XML sitemap lists current canonical URLs and excludes redirects and errors
- 8. Core content renders without requiring client-side JavaScript execution
Structure: can your answer be lifted cleanly
These seven points govern whether a retrieved page becomes the passage that gets used. The unifying principle is self-containment: a block of text should make sense when read on its own, without the paragraphs above it.
Apply these first to pages already ranking between roughly position three and fifteen for queries you care about. Those pages are already in the candidate set, so restructuring them produces visible movement faster than work anywhere else on the site.
- 9. Each heading is phrased as the question a reader would actually ask
- 10. A direct answer appears immediately below each heading, before elaboration
- 11. Answer blocks make sense in isolation, without pronouns referring backwards
- 12. One idea per paragraph, so passages can be extracted without their neighbours
- 13. Genuine comparisons are presented as tables rather than prose
- 14. Statistics name their source in the same sentence as the number
- 15. Procedures are numbered lists with one action per step
Machine readability: is the page unambiguous
These five points remove doubt about what the page is and who published it. They are quick to implement and quick to break, so the maintenance point matters as much as the initial work.
The consistent theme is that markup must match what a reader can see. Structured data describing invisible content is a guidelines violation rather than an optimization, and it undermines the trust the rest of your markup was meant to build. Our schema markup guide covers the detail.
- 16. Organization markup states a consistent name, URL, and contact details
- 17. Editorial pages carry Article markup with a visible author and dates
- 18. FAQPage markup is used only on genuine question-and-answer content
- 19. Local businesses publish LocalBusiness markup with address, hours, and service area
- 20. Markup is validated after every template change, not only at launch
Measurement: will you know whether it worked
The final two points exist because rank tracking does not measure citation, and a programme judged only on rankings will misread its own results. Impressions holding while clicks fall is the expected outcome of being summarised, not evidence of failure.
Set the baseline before making changes, because retrospective comparison is impossible otherwise. A structured starting point is what an audit produces, and tracking AI visibility covers building the prompt set you will monitor against.
- 24. A fixed set of buyer-phrased prompts is checked on a schedule across assistants
- 25. Reporting separates informational from commercial queries and tracks qualified conversions
FAQ
Questions about this guide
How long does this audit take to complete?
The foundations section is typically a day of investigation and a variable amount of remediation. Structure work is roughly an hour per page and should be prioritised rather than applied everywhere. Authority and measurement are ongoing rather than one-off.
Which points matter most if I can only do a few?
Points one through eight, because they gate everything else. A site failing retrieval cannot benefit from any of the later work, however well executed it is.
Is this different from a normal SEO audit?
Twelve of the twenty-five points are conventional technical SEO, which is the honest picture of the discipline. The additions are passage-level structure, corroboration, and measuring citation separately from ranking.
Do I need a tool to run this?
No, though a crawler makes the foundations section considerably faster and manual prompt checking across assistants becomes tedious at any real scale. The audit is defined by the questions rather than by the tooling, though our free SEO checklist tool covers the same ground interactively.