What It Took to Make Our Sites Agent-Ready

More and more of the visits to a website aren't people. They're AI agents reading pages to answer someone's question, or trying to use a product on someone's behalf. This week we went through this site and our product sites, SawirPost, Sawirly, Rindraw, Math Daily and others, to make them easier for agents to read and use.

We used isitagentready.com as the checklist. It grades a site in levels, and here's what each step took.

Markdown copies of every page

An agent reading HTML wades through navigation, scripts and footers to find a few paragraphs. So every page on this site now has a Markdown copy next to it, at index.md, with only the page's main content. A small script builds them from the HTML, so they never drift.

Agents get them two ways: by adding index.md to a URL, or by asking for Accept: text/markdown. A tiny Cloudflare Worker in front of the static files checks that header and serves the Markdown copy, with a rough token count in an x-markdown-tokens header. Browsers still get the normal page.

curl -H "Accept: text/markdown" https://sawirstudio.com/shopify/

Link headers

Every HTML page now says where its Markdown copy is, in an HTTP Link header with rel="alternate". The home page also points to llms.txt and the sitemap. An agent can find everything from one request without guessing URLs.

llms.txt

llms.txt is a plain-text index of the site for language models: a one-line summary and a list of every page with its description, linking to the Markdown copies. Ours is generated from the sitemap by the same script.

Content Signals

robots.txt says who may crawl, not what they may do with what they read. Content Signals adds that. Ours is one line, and allows all three uses, because we want to be found and quoted:

Content-Signal: ai-train=yes, search=yes, ai-input=yes

Those three steps took every site to Level 3.

An API catalog and Agent Skills

SawirPost and Sawirly have APIs, so they went further. Each publishes an API catalog at /.well-known/api-catalog (RFC 9727): a small JSON document pointing to the OpenAPI description and the docs. An agent that lands on the site can find the API without reading a single page.

They also publish Agent Skills, short instructions for agents on how to do common tasks with the product. That took both to Level 4.

Where we stopped

Level 5 wants agents to be able to sign up and authenticate on their own, through an OAuth agent registration flow. A page describing how to sign up with an email and password doesn't count. That's a real feature to build, with real security questions, so we're thinking it through instead of rushing it.

Was it worth it?

For a static site, most of this was an afternoon: one script, one small Worker and a line in robots.txt. Nothing changed for people visiting the site. We think more questions will be answered by agents reading sites like ours, and a clean Markdown page is the easiest way to be quoted correctly.

We set up the same for clients as part of Launch & growth.

Keep reading: Make Your WordPress Plugin Usable by AI Agents with the Abilities API, or The Throwable App Icons on Our Home Page.