Key takeaways
- AI agents are tools that browse and act for a person: compare, book, buy. They fail on sites built only for human eyes.
- llms.txt is a plain-text guide to your site for AI systems. Cheap to add, and increasingly read.
- Structured data tells an agent your services, hours, prices and availability without guessing from layout.
- WebMCP is an emerging standard for exposing actions (book, quote, buy) to agents directly.
- The test is simple: can an agent get from your homepage to a completed task without a human? Most sites cannot yet.
Most websites were built for a person with a mouse. An AI agent is a different visitor. It does not see your hero image. It reads your markup, looks for the facts it needs, and tries to complete a task: find a price, check availability, book a slot, add to cart. When it hits a form it cannot parse or a menu that only works on hover, it gives up and tries the next site.
Agent readiness is the work of making your site one an agent can finish a task on. It is not exotic. Most of it is the tidy, well-structured version of what you already have.
What an agent actually does on your site
Picture the request: "Find me a physio in Mississauga with openings this week and book the earliest one." An agent will:
- Search, using the same engines we cover elsewhere, and shortlist a few clinics.
- Open each site and look for services, location, hours and a way to see availability.
- Try to book, filling a form or using a booking tool.
- Report back with what it did, or why it could not.
Every step has a failure point that has nothing to do with how good the clinic is. Hours in an image instead of text. Services in a PDF. A booking widget in an iframe from a vendor that blocks automation. A form with a required field the agent cannot identify.
llms.txt: a map for machines
llms.txt is a plain-text file at the root of your site, like robots.txt, written for AI systems. It says what the business is, links to the pages that matter most and describes them in a sentence each. It is not a ranking trick. It is a courtesy that makes your site cheaper to understand, and the systems that read it reward that.
A good one for a clinic fits on one screen: name, what you do, where, hours, then links to services, pricing, booking, FAQ and contact, each with a one-line description. We write it from the audit, and we keep it updated when pages change.
Structured data: facts, not guesses
Schema markup is a set of labels inside your pages that say, in a machine-readable way, "this is a LocalBusiness, these are its opening hours, this is a Service that costs this much, this is an FAQ." Search engines have read it for years. Agents depend on it, because it lets them take facts without inferring them from layout.
What matters for agents specifically:
- Organization and LocalBusiness with the same name, address and phone as everywhere else.
- Service or Product with real names, descriptions and, where you can, prices and availability.
- OpeningHoursSpecification that matches what is on the door.
- FAQPage for the questions people ask on the phone.
- Validity. Broken or contradictory schema is worse than none, because it teaches the agent not to trust the site.
WebMCP: letting agents act, not just read
Reading is half the job. The other half is doing. WebMCP is an emerging standard, built on the Model Context Protocol that AI tools already use to talk to software, that lets a website expose actions directly: "get a quote," "check availability," "book," "add to cart," with clear inputs and outputs. Instead of an agent guessing its way through your form, it calls the action.
It is early. Not every agent supports it yet, and the specification is still moving. But the direction is clear, and the sites that expose their key actions cleanly will be the ones agents prefer, for the same reason people prefer a shop with a clear front door. Our readiness check includes an agentic-browsing test today and a WebMCP readiness note for the actions that matter most on your site.
The failures we see most
- Blocked crawlers. A firewall or plugin quietly blocks AI user agents. Nothing else matters until this is fixed.
- Facts trapped in images and PDFs. Menus, price lists, hours, service areas.
- Booking tools that reject automation. Some vendors do, and it is worth asking yours.
- Forms without labels. A field with placeholder text only is invisible to an agent.
- Contradictions. Different hours on the site, Google and the schema. The agent picks one, and it may be wrong.
Where to start
If you do nothing else this month: confirm the AI crawlers can reach your site, put your hours, services and prices in real text with matching schema, label every form field, and publish an llms.txt. That is a week of work on most sites and it moves you from "unusable" to "usable" for the agents already out there.
The fuller version, with a WebMCP plan for your booking or quote flow, is what our agent readiness service does. But the first week is the one that matters, and you can start it today.