AI-Readable Content: Make Pages Easy to Find, Understand and Quote
Once AI crawlers are allowed in (see AI crawlers), the next question is whether they can make sense of what they find. Many of the habits that help here are the same ones that have always helped search engines and readers: real text in the HTML, clear structure, and honest information about who wrote what and when.
No checklist guarantees that an assistant will cite you; that also depends on how useful and trustworthy your content is. These checks remove the technical reasons it couldn’t.
Content in the HTML
Most AI crawlers fetch your HTML and don’t run JavaScript. If your main content is rendered in the browser by a framework, they see an empty shell.
We count the words of visible text in the HTML we received (scripts, styles and SVG excluded):
| Result | When |
|---|---|
| Passed | 150 words or more |
| Needs attention | fewer than 150 words |
| Failed | fewer than 50 words and an empty app root (#root, #app, #__next, #__nuxt…) |
Fix: render content on the server. Next.js, Nuxt, SvelteKit, Astro and Remix all support server rendering or static generation; for a client-only app, pre-rendering important pages is a good middle ground.
Schema for who and what
Structured data tells machines who is behind the site and what a page is. We read the JSON-LD blocks on the page:
| Result | When |
|---|---|
| Passed | an identity type (Organization, LocalBusiness, WebSite, Person) and a content type (Article, Product, FAQPage, HowTo, Event, Service, BreadcrumbList…) |
| Needs attention | only one of the two |
| Failed | no schema types |
See our structured data guide for examples.
Author and date
Readers and AI systems weigh who wrote something and how current it is. We look for an author
(<meta name="author">, article:author, rel="author", or author in JSON-LD) and a date
(article:published_time, <time datetime>, or datePublished/dateModified in JSON-LD).
| Result | When |
|---|---|
| Passed | both an author and a date |
| Needs attention | one or neither, on an article-type page |
| Info | a home page without article content, where neither is expected |
Quotable structure
Assistants quote passages that answer a question directly. Pages with descriptive sub-headings, lists and tables are easier to quote accurately.
| Result | When |
|---|---|
| Passed | at least two H2/H3 headings, plus lists or tables (or question-style headings) |
| Needs attention | one sub-heading |
| Failed | no sub-headings |
Write headings that say what the section answers (“How long does shipping take?”), put the answer in the first sentence under it, and use lists for steps and tables for comparisons.
llms.txt
/llms.txt is a proposed convention: a plain Markdown file that summarizes your site and links to
its most useful pages for language models. As of today the major AI engines aren’t known to use
it, so we report whether you have one for information only, and it never affects your score. If you
add one, keep it short and accurate; it doesn’t replace any of the points above.
Further reading: Google: AI features and your website.