A clear purpose
One paragraph in plain English: who you are, who you serve, what kind of information lives here. The first thing an agent reads.
llms.txt tracker
— of — state & territory websites publish an llms.txt for AI agents.
A short directory file agents use to find authoritative pages on a government site — instead of guessing from search results. Building an agent? Read the spec →
Publishing llms.txt
Not yet
Publishers over time
Each new publisher strengthens the case for agent builders to make GET /llms.txt
their first move. Today the file is read mostly by dev tools — Cursor, MCP servers, and the
agents built on them.
For state digital teams
Why a state should publish an llms.txt.
People are already asking AI assistants for help with benefits, taxes, voting, and licenses.
When an agent — Cursor, an MCP server, anything browsing on their behalf — visits your site,
llms.txt is the first thing it can read. It tells the agent what your site is,
which URLs are authoritative, and how to handle time-sensitive content.
Authoritative URLs
Point agents at your canonical task-flow pages — “apply for SNAP,” “check unemployment status” — by name. So the agent lands on the right page, not the closest-looking search result.
Usage guidance
Tell agents what to do with emergency alerts, eligibility rules, and legal notices. Maryland’s file is a good template: it explicitly excludes time-sensitive alerts.
A way to reach you
A machine-readable contact for the engineering team — not the general helpdesk. Turns one-way scraping into a feedback loop with AI builders.