How to Check Your Agent Accessibility Score in AmICited

AI agents don’t read your site the way people do, and if you’ve never checked whether yours is built for them, the answer is probably “not yet.”

At a Glance

• The Agent Readiness Summary at the top of Audit → Agent Accessibility shows six pass/fail tiles at a glance.
• It covers llms.txt, accessibility (the accessibility tree), crawler access, sitemap URLs, WebMCP, and agentic commerce readiness.
• Any weak or blank tile is a concrete gap; each links to a detailed section further down the page for the fix.
• Crawler access and llms.txt matter most: if a crawler can’t reach or understand your site, the other checks won’t move the needle.
• Re-check after fixes, since these are live technical signals that can change as soon as you ship an update.

What is agent accessibility?

Agent accessibility is a measure of how easily an autonomous AI system (a shopping assistant, a research agent, a chatbot answering a user’s question, a crawler feeding a large language model) can read, understand, and act on your website without a human clicking through it. It’s a different discipline from traditional web accessibility (which serves screen readers and assistive tech) and from classic SEO (which serves search engine crawlers ranking pages in a list of blue links). Agentic AI systems don’t browse; they parse. They fetch a page, extract structured meaning from it, decide whether to trust and cite it, and sometimes take an action (like completing a purchase) on a user’s behalf, a pattern known as agentic commerce .

That parsing step is where most sites quietly fail. A page that looks perfect in a browser can be functionally invisible to an agent if its content only renders after JavaScript executes, if a robots.txt file blocks the crawler’s user-agent, if there’s no sitemap declaring which URLs exist, or if the site gives agents no machine-readable way to discover what they’re allowed to do. None of these are things a human visitor would ever notice. All of them are things an AI agent checks before it decides whether your brand is a citable, actionable source.

This is why generative engine optimization (GEO) treats technical accessibility as a foundation layer rather than an afterthought. You can publish the most authoritative content in your category, but if the underlying infrastructure prevents an agent from reading it, none of that authority reaches ChatGPT, Perplexity, Gemini, or Google’s AI Overviews. An AI accessibility audit is the technical review that surfaces exactly where those gaps are: checking crawler permissions, structured data, file discoverability, and the newer agent-specific protocols that are emerging as AI shopping and AI-driven research become mainstream.

The signals that matter fall into a handful of categories: whether you’ve published an llms.txt file that tells language models how to read and cite your site; whether your pages are structurally readable via the accessibility tree AI systems use to parse content; whether crawlers are actually permitted in via robots.txt ; whether an XML sitemap exists and is complete; whether the site exposes callable tools for agents through emerging standards like WebMCP; and whether it advertises a machine-readable transaction protocol for agentic commerce. Each of these is binary in a practical sense (either the agent can do the thing, or it can’t), which is exactly why AmICited condenses them into a single row of pass/fail tiles rather than a long narrative report.

The Agent Readiness Summary

The Agent Accessibility audit checks whether your site is technically ready for AI agents, and the Agent Readiness Summary at the top gives you the whole picture in one row of tiles.

Where to find it

Open Audit → Agent Accessibility. The page centers on “your domain’s /llms.txt, the file that tells AI agents and LLMs how to read and cite your site,” with the readiness summary at the very top. This sits alongside AmICited’s broader AI visibility tracking, so you can move from “is my site technically readable” to “is my brand actually being cited” without switching tools.

What the tiles cover

• llms.txt: whether you have a complete, valid /llms.txt. This file is the closest thing AI agents have to a table of contents for your site: it points them to the pages, docs, and data you want cited, in a format built for machine consumption rather than a browser. A missing or malformed file doesn’t necessarily block a crawler, but it does mean the agent is left to guess at what matters on your domain.
• Accessibility: how readable your pages are via the accessibility tree, the same structural representation of a page that screen readers rely on. Agents that render pages (rather than just fetching raw HTML) often walk this tree to extract content, so a page built entirely from unlabeled <div> soup or interactive widgets with no semantic markup reads as noise even when the visible text is fine.
• Crawler access: whether AI and search crawlers are allowed to read the site at all. This tile reflects your robots.txt configuration and any server-level blocks. It’s the most consequential tile on the page: if a crawler is disallowed, everything downstream (citations, structured data, agent tool access) is moot, because the agent never gets past the front door.
• Sitemap URLs: whether sitemaps are declared and how many URLs they list. A sitemap doesn’t guarantee citation, but it is the fastest way to tell an agent “here is the full set of pages worth considering,” rather than relying on it to discover everything through internal links alone.
• WebMCP: whether your site exposes callable tools for agents. This is the newest and most forward-looking tile: as agents move from reading content to taking actions, sites that expose defined, machine-callable capabilities (rather than expecting an agent to reverse-engineer a UI) are positioned to participate in that shift instead of being locked out of it.
• Agentic commerce: whether your site advertises a machine-readable transaction protocol. For any brand with a checkout flow, this tile is a leading indicator of whether AI shopping agents can complete a purchase on a customer’s behalf, or whether they’ll simply skip you in favor of a competitor that supports it.

How to use it

1. Scan for weak tiles. Anything failing or blank is an accessibility gap for AI agents. Because the tiles map to independent technical systems, a single weak tile is usually a contained, fixable problem rather than a symptom of something broader.
2. Drill into the detail. Each tile maps to a section further down the page (llms.txt review, robots & sitemaps, accessibility tree, WebMCP, agentic commerce) with specifics and re-check buttons. This is where you get the concrete “what’s missing” detail rather than just the pass/fail signal: the difference between “llms.txt: needs work” and knowing precisely which pages it’s missing or which directive is malformed.
3. Re-run after fixes. Update your site, then re-check to watch the tiles turn healthy. Because these are structural checks rather than crawl-and-wait metrics, most tiles update as soon as AmICited re-fetches the relevant files, which makes the summary a fast feedback loop for engineering work rather than something you check once a quarter.

Why this matters beyond the audit page

A healthy readiness summary is a leading indicator, not a vanity score. Sites that pass all six checks are the ones most likely to show up when you go looking for citation share in AI answers, because the agent could actually reach, parse, and trust the content in the first place. This is the same logic behind treating your robots.txt and sitemap as first-class SEO assets rather than boilerplate: they’re no longer just for Googlebot, they’re the on-ramp for every AI crawler now shaping what gets recommended, summarized, or purchased on a user’s behalf. If you want the deeper mechanics of any one tile (for example, a step-by-step walkthrough of writing a compliant file), AmICited’s guide on how to implement llms.txt covers the syntax and common mistakes in more depth than the audit page alone.

Getting from a score to a fix

The summary is the scorecard; the sections beneath it, each with its own Academy guide, are where you act. Treat a red or missing tile the way you’d treat a failed test in a CI pipeline: it names the problem precisely enough that fixing it is usually a matter of engineering time rather than guesswork. llms.txt gaps are typically a content and file-format problem you can close in an afternoon; crawler-access and sitemap issues are usually a server-configuration fix; accessibility-tree problems tend to trace back to how a front-end framework renders content and may take longer to resolve. WebMCP and agentic commerce are the newest checks, and for many sites today they’ll simply be unimplemented, which isn’t a failure so much as an opportunity to get ahead of a category most competitors haven’t touched yet.

It’s worth running this check on a recurring basis rather than once. Sites change: a new deploy can silently reintroduce a robots.txt block, a CMS migration can drop your sitemap, a redesign can strip the semantic markup an accessibility tree depends on. Because AI agents don’t send you an error message when they can’t read your site, the readiness summary is often the only place you’ll see the regression before it shows up as a drop in citations. Pairing it with AmICited’s AI rank tracker , which watches your actual citation position across ChatGPT, Perplexity, Gemini and AI Overviews, gives you both sides of the picture: the technical readiness score that predicts whether you can be cited, and the live tracking that confirms whether you are being cited. If you’re auditing a site for the first time and want the full picture rather than just the agent-readiness slice, AmICited’s guide to how to conduct an AI visibility audit walks through the complete methodology, and teams managing this across multiple client sites will find the workspace and reporting structure in AmICited for agencies built specifically for running the same checklist at scale.