When people talk about AI and websites, they usually mean one thing: can Google's AI find you? That is important, but it is only part of the picture. A bigger shift is happening that most businesses have not noticed yet.
AI agents are starting to interact with websites on behalf of users.
Not just reading content. Actually doing things. Filling out contact forms. Comparing service providers. Booking appointments. Requesting quotes. Making purchases. The shift is from "can AI find you" to "can AI do business on your site."
Agent readiness measures whether your website is structurally prepared for this shift.
What is an AI agent?
An AI agent is software that acts on behalf of a person. You tell it what you want, and it goes and does it. Today's agents include ChatGPT with browsing, Google's Gemini, Perplexity, and a growing number of specialised assistants built on top of these models.
When someone says "find me an accountant in Manchester who specialises in R&D tax credits and send them an enquiry," the agent needs to find relevant businesses, evaluate their websites, and complete actions. Every step of that process depends on the website being machine-readable.
Most websites are not.
The three waves of AI interaction
We are in the middle of a progression that is happening faster than most businesses realise.
Wave 1: AI reads your content (happening now)
Google's AI Overviews, ChatGPT, and Perplexity can read web content and synthesise answers. Clear structure can help them interpret a page, but citations also depend on relevance, authority, the question asked, and the AI service being used.
Wave 2: AI interacts with your site (starting now)
Agents browse your site, read your service pages, understand your pricing, and fill out your forms. This requires a different kind of readiness — your site needs to be navigable and actionable by machines, not just readable.
Wave 3: AI transacts on your site (coming soon)
Full end-to-end transactions. An AI agent books an appointment, pays a deposit, and confirms on behalf of the user. Protocols like WebMCP (Model Context Protocol for web) and NLWeb (Natural Language Web) are being developed to make this possible at scale.
Most websites are still catching up with Wave 1. The businesses that prepare for Waves 2 and 3 now will have a structural advantage that is very difficult to replicate later.
What is WebMCP?
WebMCP — the web extension of Anthropic's Model Context Protocol — is an emerging standard that lets AI agents interact with websites in a structured way. Instead of agents having to scrape and guess, WebMCP lets a website declare what actions are available, what data they accept, and how to complete them.
Think of it as an API for the open web. A machine-readable menu of "here is what you can do on this website."
Similarly, NLWeb (Natural Language Web) is an initiative that lets websites accept natural language queries and return structured responses. Instead of navigating through menus and forms, an AI agent could simply ask your website "do you offer R&D tax credit services for SaaS companies?" and get a direct, structured answer.
Neither protocol is widely adopted yet, but the direction is clear: the web is becoming machine-actionable, not just machine-readable.
The 9 agent-readiness checks in the current score
The current scoring framework measures agent readiness through 9 structural checks on the page you submit. They show whether the signals an agent would rely on are present. They do not run actions, prove an action is safe, or show that any agent actually uses your site.
Agent readiness checks
Why these signals matter
Content is in the fetched HTML
Many automated readers do not run JavaScript. If your content only appears after scripts run, some of them may see little or nothing. Server-side rendering or static generation usually solves this — your developer will know what this means.
Forms and action paths
For an assistant to help someone send an enquiry or book, form fields need clear labels, standard input types and semantic HTML, and the links to key actions need to be visible in the page.
Structured data
Schema.org markup states what the page is about in machine-readable form, for example an Organization with your name and contact details. The check looks at field coverage only.
Token efficiency and navigation
Pages heavy with boilerplate and scripts make the useful text harder to find. Plain HTML navigation helps automated readers discover your other pages. Both are diagnostic ratios, not hard thresholds.
Crawler policy, llms.txt and machine endpoints
An explicit robots.txt policy tells AI crawlers where they stand. A sitemap helps them find pages. llms.txt is an optional, emerging convention: it is not required by major search or AI systems and does not guarantee visibility.
Good practice the score does not measure
Machine-readable contact details, specific service descriptions, a stated pricing or engagement model and a fast response time are also worth having. They are not among the 9 scored checks.
Where to start
Agent readiness is a newer area. The free scan shows which of these signals your page already has and which are flagged for review, so you can decide with your developer what is worth doing.
Check your agent readiness score
Our free scan includes the agent-readiness checks. Scan time varies with the website.
Scan My Website