An AI-readable website is one whose content is accessible without executing JavaScript, structured in semantic HTML, described by Schema.org structured data and open to AI crawlers in its robots.txt. The llms.txt file, an emerging standard, completes the setup by giving language models a clear map of your site. Here is the 2026 technical checklist, verifiable point by point on your own site.
Semantic HTML: the foundation of everything
AI crawlers do not execute JavaScript, or do so poorly. A site whose content only appears after browser-side rendering — the case of classic single-page applications — is, for many of them, an empty page. The first check is therefore simple: view the source code of your key pages (right-click, "view source") and verify that the important text actually appears there. If it does not, server-side rendering (SSR) or static generation (SSG) is required; modern frameworks such as Next.js or Astro offer it natively.
Beyond that, semantics matter: a single H1 per page, a logical H2/H3 hierarchy, short paragraphs, lists for enumerations and appropriate tags (article, nav, main). Models chunk your pages into fragments: the cleaner the structure, the more coherent and quotable the extracted fragments.
Also think about non-text content: descriptive alt attributes on images, transcripts for videos, HTML tables rather than screenshots. Anything that is not accessible text remains invisible to most extraction systems.
Schema.org and JSON-LD: the four essential schemas
Structured data describes your content in a format machines understand without ambiguity. Four schemas cover the essential needs of an SME:
- Organization: who you are — name, logo, contact details, social profiles.
- Article: your editorial content, with author and publication dates.
- FAQPage: your questions and answers, directly usable by answer engines.
- BreadcrumbList: each page's position in the site's structure.
A minimal example of an Organization schema, to place inside a script tag of type application/ld+json:
{ "@context": "https://schema.org", "@type": "Organization", "name": "Your Company", "url": "https://www.example.com", "description": "IT expertise for SMEs", "sameAs": ["https://www.linkedin.com/company/example"] }Then validate every schema with the official Schema.org validator or Google's rich results test: invalid structured data is simply ignored.
llms.txt: a business card for the models
Proposed in 2024, llms.txt is a Markdown file placed at the root of the site (yourdomain.com/llms.txt) that introduces your site to language models: who you are, what you offer, where to find the important content. Its adoption by AI providers remains partial to date, but its implementation cost is close to zero and it structures information usefully. A typical llms.txt contains:
- A level-1 heading with the company name, followed by a one-sentence summary of your activity.
- One section per major content group (services, blog, pricing) listing the essential URLs, each with a short description.
- Optionally a companion llms-full.txt file containing the full content of key pages in Markdown format.
The bet is simple: if the standard takes off, you are ready; if it stays marginal, you will at least have clarified how your offering is presented — rarely a wasted exercise.
robots.txt: who do you let in?
The main AI crawlers identify themselves by their user-agent: GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and Google-Extended, which controls the use of your content by Gemini. The decision to allow them is strategic: a media outlet living off its content may want to block training bots; a services company seeking visibility has every interest in welcoming them, or it disappears from generated answers. Check your current file: some CDNs and robots.txt templates block these bots by default, without anyone having actually decided it.
Sitemap and hreflang: the signage
An up-to-date sitemap.xml, declared in robots.txt, helps all bots — classic and AI — discover your pages and their updates quickly. If your site exists in several languages, hreflang tags indicate which version matches which audience: without them, an AI may quote your English page to a French-speaking user, or simply ignore one of the two versions. Take the opportunity to check the declared last-modified dates: a sitemap announcing pages untouched for years gives no bot a reason to come back often.
Testing concretely: ask the AIs
- Ask ChatGPT, Perplexity and Gemini: "What do you know about [your company]?" and "What does [your company] offer?". The answers reveal what the models have actually retained — and the errors to fix.
- Ask the business questions your customers would ask and check whether your site appears among the cited sources.
- Fetch your pages with a simple command-line tool such as curl, to see exactly what a bot gets without JavaScript.
- Monitor your server logs: regular visits from GPTBot or ClaudeBot confirm that your content is being collected.
- Repeat the exercise after every major site change: a redesign can silently break the AI readability you had gained.
AI readability requires neither a redesign nor a large budget: it is a series of precise technical checks, most of which come down to basic web hygiene. But every neglected point is a place where your business becomes invisible to assistants your customers already use daily. A quick technical audit, carried out with a partner who works with these standards every day, generally identifies and fixes the essentials within a few days.