Can an AI agent get a job done on your site?
A visitor’s AI assistant will soon try to do things on your site for them. This check shows what it can do today: reach you, find your price and hours, book or order, and where it fails. Only what your site offers is scored.
By AI agent we mean the assistants that act for a customer: ChatGPT’s agent mode, Perplexity Comet, Claude in Chrome, and the fetchers behind every AI answer. They all read the page without JavaScript and find their way by the names on buttons and fields.
no signup · we fetch only public files · nothing about you is stored
Why these eight, and not a checklist of everything
Vercel ran the controlled experiment: when the answer was already in server-rendered HTML, adding llms.txt, JSON-LD or sitemaps changed nothing. When it was not, nothing else saved it. Agents do not run JavaScript.
Ahrefs looked at 137,000 domains: 28% publish an llms.txt and 97% of those files are never fetched by anything. Google calls it a temporary crutch, not a search signal. Lighthouse scores a missing one Not Applicable, not fail.
Of the six audits in Chrome’s Agentic Browsing category, the two that apply to a normal site are the accessibility tree and layout stability. An unlabelled icon button is invisible to an agent, the same way it is to a screen reader.
If the only path to a booking runs through a script, an agent cannot press it. That is why the four jobs are scored on whether a machine can follow them, not on whether a human can see them.
What we fetch
/, the homepage as a plain fetch, as an AI crawler, and asking for markdownrobots.txt, agent user-agents and Content Signalssitemap.xml · sitemap.md, real routes an agent can deep-link to.well-known/mcp.json, callable tools, and how many of them do anything.well-known/agent-card.json · ucp, A2A and commerce discoveryllms.txt · AGENTS.md, the optional readmes, weighted honestly