Agent-ready website: what it is and how an AI agent sees your site
An agent-ready website is a site an AI agent can read and use without help, for example to book an appointment on someone's behalf. An agent can look at a screenshot, at the HTML and at the accessibility tree of your page. What you arrange for that is largely what a good, accessible website already asks for: real buttons, labelled fields and a page that does not shift. There is no official standard, and Chrome calls its own check experimental.
Someone tells an AI assistant: find an accounting firm nearby and book an introduction for next week. The assistant opens a few websites, reads what is there and tries to fill in the contact form. Whether that works on your site depends on how the page is built. That is what agent-ready is about. Below is how an agent sees your site, what you can arrange and what we measured on our own site.
What is an agent-ready website?
An agent-ready website is a site that AI systems can read and use without help. It concerns two kinds of visitors: AI search engines that fetch your pages to answer a question, and AI agents that do something on someone's behalf. In its guide to generative AI in Search Google describes those agents as systems that perform "tasks on behalf of people", such as booking a reservation or comparing products.
Agent-ready is an umbrella term. There is no certificate and no official standard that decides when a site qualifies. There are guidelines from Google and Chrome, and this article is based on them.
How does an AI agent see your website?
An AI agent can view your website in three primary ways: as a screenshot, as HTML and as an accessibility tree. That is how Kasper Kulikowski and Omkar More describe it in Build agent-friendly websites on web.dev, a developer site with articles by the Chrome team and external experts.
- Screenshot. The agent captures an image of the page and has a vision model recognise the parts. According to web.dev that can be slow and expensive, which makes it mainly a backup when the structure is unclear.
- HTML. The agent reads the code: which parts sit inside which, and what data they hold. If a buy button sits inside the block of a product, the agent assumes the button belongs to that product.
- Accessibility tree. This is a summary the browser makes of your page, with the role, the name and the state of every interactive element. Assistive technology uses it too; a screen reader is one example.
Modern agents combine the three. They take a list of what can be operated from the code and the accessibility tree and lay it next to the image to understand the layout.
The consequence: a button that looks like a button to a person and is a styled block without a role in the code can be missed by an agent.
What makes a website usable for an agent?
A website is usable for an agent when every action a visitor has to take is clearly reflected in the page itself. web.dev gives guidelines for that. The most important ones:
- No invisible layers. A transparent layer over a button can cause the agent to skip the button.
- Real buttons and links. Use
<button>and<a>. If that is not possible, give the element a role and make it reachable by keyboard. - A label on every form field, linked to the field with the
forattribute. That tells the agent what to fill in. - A pointer cursor (
cursor: pointer) on everything that is clickable. - Large enough. Interactive elements required to complete an action should be visible and larger than 8 square pixels.
- A stable layout. A button that sits in a different place on every page will likely confuse an agent that works with screenshots.
Under "Next steps" web.dev writes: "Everything we suggest to make a site 'agent-ready' also makes sites better for humans." If you built your site to be accessible, you have already done a large part.
What is WebMCP?
WebMCP is a proposal for a web standard with which a website itself states which tasks an agent can perform. Chrome describes it in its documentation: the site registers a task as a "tool", with a description of the input, and the agent calls that tool instead of guessing where to click.
It is not a standard yet. The specification says so itself: "It is not a W3C Standard nor is it on the W3C Standards Track." Chrome calls it "a proposed web standard", you can try it through an origin trial from Chrome 149, and web.dev warns that it is "in active development" and may change. For sensitive actions, such as a purchase, you can have the agent ask the user for confirmation first.
For most business sites this is, in our view, the last step. There is more about it in WebMCP: what it is and how it works.
How AI-ready is your business?
Take the free AI Readiness Quiz: 7 questions, 2 minutes.
Take the quiz →How do you measure whether your site is agent-ready?
You can test it with Lighthouse, Chrome's testing tool, which has a category for agentic browsing. Chrome writes that the category currently focuses on "actionable signals rather than a definitive ranking". According to Chrome's explanation the category looks at four things:
- whether the site registers WebMCP tools;
- whether the accessibility tree is sound, among other things: does every interactive element have a name, are the roles right and is nothing hidden that can still be operated;
- whether the page stays stable (the measure for that is Cumulative Layout Shift);
- whether there is an llms.txt.
The category gives no score from 0 to 100. You see a fraction: how many of the checks you pass. Chrome calls the category experimental, "based on proposed standards", and it needs Chrome 150 or later.
There is a simple test of your own as well: go through your contact form using only the Tab key and Enter. If that does not work, there is probably something in the code that hinders an agent too. It is a rule of thumb; a full test takes more.
Do you have to act on this now?
Arrange the basics now and wait with the rest until you have a reason for it. Google writes in its guide that you can look into agents if this is "relevant to your business and you have extra time", and calls protocols such as the Universal Commerce Protocol "emerging".
The basics are what is listed above: buttons, labels and a stable page. That work is not wasted: web.dev writes that the same advice also makes a site better for humans. Whether your pages get found comes before that. How those layers relate is in SEO vs GEO vs agent-ready.
What we cannot promise you: that an agent-ready site brings in more customers. We have no measurement of that.
How does mindsora.io score?
What we measured on our own site:
- Lighthouse, agentic browsing. On 13 November 2026 Lighthouse failed our llms.txt file, because it did not contain a markdown link. The accessibility tree was fine. After the fix we passed 3 of the 3 checks that applied on 14 November (own measurement through PageSpeed Insights, homepage, mobile). How that error came about is in our llms.txt guide.
- WebMCP. We have not built that in. The checks for it did not apply to us.
- Service catalogue at /api/services. Our packages and add-on services are there with prices as JSON, so an agent can fetch them without parsing the page.
- Our Agent-Ready Grader gave mindsora.io 98 out of 100 points on 17 October 2026. Semantic HTML scored 8 out of 10. That grader is our own instrument, so it does not replace an independent assessment.
What we did not measure: whether agents visit our site and whether they complete a task there.
What does it cost to make your site agent-ready?
We offer two services, with prices excluding VAT. The Agent-Ready Audit costs €697 one-off: an assessment of llms.txt, schema markup and semantic HTML. The Agent-Ready Implementation costs €2,500 setup and €350 per month: llms.txt, schema.org markup, semantic HTML and a booking endpoint. Neither includes WebMCP. Both are in our services overview.
You can go through the ten points that matter now yourself with our agent-ready website checklist.
Frequently asked questions
Is agent-ready the same as accessible? Largely. An agent uses the accessibility tree, the same layer as a screen reader. WebMCP and llms.txt come on top as extras.
Do I need an API? No. An agent can fill in a well-built form. A fixed entry point for machines is an extra.
Do I need an llms.txt? For Google Search, no. If the file exists, Lighthouse tests its content: an H1, at least one markdown link and at least 50 characters. If it is missing, the audit does not apply; Chrome calls the file "optional at the moment".
Our free Agent-Ready Grader tests other points, among them crawler access, llms.txt, schema.org and page structure (header, main, nav), with a score from 0 to 100. It does not test buttons, labels, layout stability or WebMCP.
Want to work out how many hours a recurring task costs you? The free AI Readiness Scan does that: 7 questions about your business processes.
Ready to automate?
Book a free consultation and discover what AI automation can do for your business. Or take the 2-minute AI Readiness Quiz first.
Book a call