Home/AI Agent Readiness

Add-on or standalone

AI Agent Readiness

Make your site one an AI agent can read, use and finish a task on: find your hours, get a quote, book, buy. Tested with the six Lighthouse Agentic Browsing audits and a live task run, then fixed.

Get my free first look The six audits, explained

A developer's desk with a monitor showing a structured tree of labelled page elements

The shift

Agents do not see your site. They read it.

An AI agent never looks at your screen. It reads the accessibility tree, the markup and the files you publish for machines. When it hits a button with no name or a form with no labels, it gives up and tries the next site. The next site gets the booking.

Picture the request: "Find me a physio in Mississauga with openings this week and book the earliest one." The agent searches, shortlists a few clinics, opens each site looking for services, location, hours and availability, tries to book, and reports back. Every step has a failure point that has nothing to do with how good the clinic is. Hours in an image. Services in a PDF. A booking widget that blocks automation. A required field the agent cannot identify.

Agent readiness is the tidy, well-structured version of what you already have, plus the two files and one standard that tell agents where things are and what they can do.

What we test

The six Lighthouse audits, plus a live task run.

Google added the Agentic Browsing category to Lighthouse 13.3 in May 2026. We run it, read it honestly, then go further with a real agent attempting a real task on your site.

agent-accessibility-tree

Can an agent find your controls?

Agents act on the accessibility tree. A button without a programmatic name does not exist to them. Modals, drawers, chat widgets, carousels and mobile menus are the usual offenders.

We fix: every interactive element gets a real name and a valid role.

cumulative-layout-shift

Does the page hold still?

For a person a layout shift is a misclick. For an agent it is clicking where the button no longer is, which ends the task.

We fix: dimensioned images, reserved space for banners, fonts that do not reflow.

llms-txt

Is your llms.txt real?

Parsed as markdown: an H1, real descriptions, markdown links. A bare list of URLs fails, and a malformed file scores worse than none.

We fix: a validated llms.txt describing your key pages in one line each.

webmcp-registered-tools

Can an agent act, not just read?

WebMCP lets a site expose actions with a schema an agent can call. Proposed standard, Chrome origin trial.

We fix: a plan for which flows to expose first and what the schema should say.

webmcp-form-coverage

Are your forms annotated?

Declarative annotations (tool name, description, per-field descriptions) let the browser derive a tool from a form. Without them the agent guesses from labels.

We fix: annotations on booking, quote and contact forms, with autosubmit off for anything sensitive.

webmcp-schema-validity

Is the schema valid?

A registered tool with a broken JSON Schema is worse than no tool. Agents learn not to trust the site.

We fix: every registered tool validated against the spec before it ships.

Then the live task runWe give a real agent a real task on your site: find the hours, get a price, book a slot, add a product to the cart. We record where it succeeds, where it stalls and why. That transcript is the most persuasive page in the report.

What you get

Files, fixes and a plan.

Report

Readiness scorecard

The six Lighthouse audits with what each result means for you, plus the live task run transcript. Not applicable is reported as not applicable, never as a failure.

Files

llms.txt and structured data

A validated llms.txt. Organization or LocalBusiness, Service or Product, OpeningHours and FAQ schema, consistent with what is on the page and with your other profiles.

Fixes

Named controls and labelled forms

Every button, link and field an agent needs gets a programmatic name. Facts move out of images and PDFs into real text.

Plan

WebMCP roadmap

Which flows to expose first (quote, availability, booking, cart), the schema for each, and the declarative form annotations shipped where the standard already works.

Timeline

Three days to know. Weeks to fix.

3 days

Readiness check

Lighthouse run, live task run, the scorecard and the fix list.

1 week

Quick wins

Crawler access, llms.txt, labels, facts out of images. Most sites go from unusable to usable here.

2 to 3 wks

Full fixes

Structured data, form annotations, booking-flow work on an existing site.

Included

On a new build

Every site we build is agent-ready from day one.

Straight talk

What we will tell you honestly.

"Not applicable" is not a failure. The WebMCP checks report not applicable on almost every site today. Anyone selling you a fix for that is selling you nothing.
Agentic browsing is not a ranking factor. It is a readiness signal for a channel that is arriving, not an SEO emergency.
A bad llms.txt is worse than no llms.txt. Absence scores not applicable. A malformed file scores a fail.
Some vendors block agents on purpose. If your booking tool does, we will tell you, and we will tell you what the alternatives are.

Questions

Plain answers.

What is an AI agent, in this context?

Software that browses and acts on the web for a person: compares options, checks availability, fills a form, books, buys. ChatGPT's agent mode, Perplexity's assistant, browser agents and the shopping agents the big platforms are shipping. They read your site's markup and accessibility tree, not the picture on the screen.

Is agent readiness a ranking factor?

No. Google has not announced it as one and independent testing found no effect on position. It is a readiness signal for a channel that is arriving. We will not sell it to you as an SEO emergency.

What is llms.txt, and do I need one?

A plain-text file at your site root, written for AI systems, that describes your key pages in a sentence each. Lighthouse parses it as markdown: it needs an H1, real descriptions and markdown links. Absent scores not applicable. A malformed one scores a fail, so only ship one that has been validated. We write and validate it.

What is WebMCP?

An emerging standard, built on the Model Context Protocol AI tools already use, that lets a site expose actions (get a quote, check availability, book, add to cart) with a schema an agent can call directly. It is a proposed standard in a Chrome origin trial. Almost nobody has adopted it yet, so not applicable is the normal result today. We plan which of your flows to expose first, and ship the declarative form annotations that make it work.

How long does the readiness check take?

Three business days. Fixes on an existing site usually land in two to three weeks. On a new build it is included from day one.

Can you do this on my current platform?

Usually. WordPress, Shopify, Webflow, Squarespace and custom builds all allow the core fixes: crawler access, labelled controls, structured data, llms.txt. Some booking widgets block automation at the vendor level; we will tell you which ones and what to do about it.

From the blog

AI agent readiness: what llms.txt, structured data and WebMCP do for your site What is AI visibility? A plain guide for business owners

Ready for the next buyer.

Start with the free first look. It includes the agent readiness checks we can run from the outside.

Get my free first look

Or check your site in 20 seconds