A one-hour AI readiness check you can run by hand

Everything below can be checked with a browser and a terminal in about an hour. Ordered by consequence: the early items can silently negate everything after them, so work down rather than picking favourites.

Access — can a machine reach you at all

1. Read your robots.txt

Visit /robots.txt. Look for a broad Disallow: / that escaped from a staging environment, and confirm any rule naming a crawler says what you intend. Remember that a group naming a specific crawler replaces the wildcard group for that crawler rather than adding to it, so disallow rules must be repeated inside it.

2. Confirm your sitemap loads and is current

Visit /sitemap.xml. Pick five URLs at random and request each one. If any return errors, the sitemap is stale. Confirm robots.txt declares its location.

3. Check content is in the served HTML

Use view-source, not inspect — the inspector shows the page after scripts run, which hides exactly this problem. Search for a distinctive sentence from the page. Do this on your homepage and on a product or article page, since these often differ.

Identity — can a machine tell who you are

4. Look for Organization structured data

Search your homepage source for 'application/ld+json'. Confirm an Organization block exists, that it is in the served HTML rather than injected by script, and that the name and description match what the page visibly says.

5. Compare how you describe yourself in four places

Homepage, About page, structured data, and footer. Write the four descriptions side by side. Disagreements here are common, invisible day to day, and directly cause entity confusion.

6. Confirm About and contact pages exist and are linked

Reachable from the homepage, with real contact details rather than only a form. Their absence is conspicuous, because it is characteristic of sites built to be disposable.

Readability — can a machine use what it finds

7. Check titles and descriptions are distinct

Across five representative pages, confirm each has a title and description, and that they differ meaningfully. Templated titles differing only by a product name are a common gap.

8. Count your h1s

Exactly one per page, stating what the page is about. Zero and several are both problems, and both are common in component-driven layouts where headings are assembled from parts.

9. Apply the portability test

Take a sentence stating something important, remove its surroundings, and ask whether it still means anything. If every claim depends on the paragraph before it, the page is hard to quote and hard to skim.

Trust — is there reason to rely on you

10. Check authorship on editorial content

Is anyone identified as responsible, and does the attribution connect to something — an author page, a track record, author markup? A name attached to nothing does little.

11. Check whether you cite anything

Do your substantive pages link to sources supporting their claims? For each link, ask whether that page supports the specific sentence it is attached to. Decorative citations are transparent to anyone who follows them.

12. Check dates on time-sensitive pages

Anything stating prices, statistics, regulations, or capabilities should show when it was published and when it was substantively updated. Confirm the update date is not simply set on every deployment, which makes it meaningless.

What to do with the results

Fix in list order. Items 1 to 3 can negate everything else, so they come first regardless of how minor a specific defect seems. Items 4 to 6 are usually the highest return for the least work, because entity clarity is cheap to fix and widely neglected.

Items 7 to 12 are ongoing practice rather than one-time fixes. They describe how content gets made, which is why they cannot be completed in an afternoon and why they keep working when technical tactics stop.