Before asking whether the site "uses AI", it is worth asking something simpler: can an AI agent read, understand, and trust what is on this site? The answer depends on a list of technical items, most of which should already exist as SEO good practice — only now they gain an additional purpose.

The checklist

1. Valid robots.txt with explicit signals Every site should have a correct robots.txt, controlling what may be crawled. It is worth going further and explicitly declaring how content may be used by AI systems, with the Content-Signal line proposed by contentsignals.org — for example search=yes, ai-input=yes, ai-train=yes. That removes ambiguity about the site's intent.

2. Updated sitemap.xml A simple, up-to-date site map remains the most direct way to tell any crawler, human or AI, which pages exist and when they last changed.

3. Unique titles and descriptions per page Each page needs its own <title> and meta description, specific to that page's content. Generic or duplicate pages are harder to cite correctly, for both search engines and AI assistants.

4. Canonical tags The canonical tag avoids ambiguity when more than one URL points to the same content, something that confuses both traditional crawlers and systems that need to decide which version of a page to cite.

5. Structured data (JSON-LD) Structured markup describing organization, services, articles, and other site entities helps machines extract facts without interpreting free text. It is one of the elements most cited in Google's official SEO guidelines for content understandable by automated systems.

6. An llms.txt file A plain-text summary of the site, meant to be read by language models, covering who the company is, what it does, and where to find more information. It works as an executive summary for agents.

7. Open Graph image and a proper 404 page Details that seem brand-only but matter: a well-configured OG image helps sharing and preview by automated tools, and a custom 404 page keeps agents and people from getting stuck in dead ends.

8. For companies with real data: a discoverable MCP server This is the most advanced item on the list, and not every company needs it now — but for those with systems with changing data (catalog, availability, status), exposing an MCP (Model Context Protocol) server lets agents query real information instead of interpreting a static page. To be discovered without friction, the server should have a server card (/.well-known/mcp/server-card.json) and appear in an API catalog (/.well-known/api-catalog, following RFC 9727), ideally referenced by Link headers (RFC 8288) on the homepage.

A reference example

yanmantovani.com implements all of the above: robots.txt with Content-Signal, llms.txt and llms-full.txt, JSON-LD on main pages, unique title and description per page, canonical tags, its own OG image, custom 404, and an MCP server at /api/mcp discoverable via server card and API catalog. None of these items alone solves everything, but together they form a consistent base.

Summary

This checklist is not about chasing trends. It is about reducing friction between what your company knows about itself and what an automated system can actually discover and confirm. The lower that friction, the greater the chance of appearing correctly in the answer an AI assistant gives someone who never visited your site.