Free tool · 23 checks

AI Agent Readiness Checker

Is your website ready for AI agents? Paste a URL. In about 10 seconds you'll get a score out of 100, your level from 0 to 5, 6 extra checks and a plain-English fix for every gap.

  • No signup
  • A fix prompt for every gap
  • Read-only, public pages only

My own site, scored

20ryandoser.com before any fixes, September 14, 2026
60After one day of fixes with Claude Code. This checker scores it 60 and Level 4 today.

The basics

What is AI agent readiness?

AI agent readiness is how easily an AI agent can reach your website, read it and act on it without a person clicking around for it. The agent might be ChatGPT fetching your page to answer a question, Claude researching vendors for someone, or a coding agent like Claude Code reading your docs.

For decades, websites were built for two visitors: people in browsers and search engine crawlers. Agents are a third visitor, and they read your site differently.

That third visitor is small but growing fast. HUMAN Security measured traffic from AI agents and agentic browsers up 7,851% in 2025, which I cover in my dead internet theory and marketing analysis.

Here's a common example. A site built as a single-page app looks perfect in your browser. But the HTML it actually sends is an empty shell, and the text only appears after JavaScript runs. An agent that doesn't run JavaScript gets a blank page and moves on to a competitor.

How an AI agent reads your site: the agent checks robots.txt, gets through the firewall, reads the raw HTML, then uses your page in its answer. A no at any gate and the agent leaves with nothing.

So the question this checker answers is simple. When an agent shows up, does your site help it or fight it?

The opportunity

Why agent readiness matters (and why almost nobody has it yet)

Cloudflare's Radar team measured how rare these standards are in April 2026. It started from the 200,000 most visited domains on the internet, filtered out redirects, ad servers and tunneling services, and found:

78%have a robots.txt file
4%state their AI usage preferences in it
3.9%serve markdown when an agent asks
<15sites publish MCP Server Cards or API Catalogs

A spot check while building this checker found the same pattern. On October 14, 2026, a scan of 82 well-known sites put 56 of them at Level 1, and 55 of those were missing just one line in robots.txt: the Content-Signal line covered in the fix order below.

That gap is the opportunity. Some fixes take five minutes. Others depend on how your site is built. Either way, the numbers show the newer agent standards are still rare, so most sites haven't adopted them yet.

Markdown is a good example of why this matters. AI models read text in tokens (small chunks of text), and every token costs processing time and room in the model's memory. Cloudflare measured up to an 80% token reduction in some cases when an agent gets markdown instead of HTML. Fewer tokens means an agent can read more of your site before it runs out of room.

Under the hood

What this agent readiness checker tests

The checker runs 17 core checks in four groups: discoverability, content, bot access control, and a group for APIs, logins, MCP servers and skills. It also runs six unscored checks for the problems I see most on real business sites: text hidden behind JavaScript, firewalls that quietly turn AI agents away, missing structured data, slow servers, pages too heavy for Googlebot to read in full, and llms.txt. Those six show up in your report but never change the score. Commerce protocols (x402, UCP and ACP) aren't checked.

Discoverability

Can an agent find your pages and the files that describe your site?

  • robots.txtA real robots.txt with User-agent rules, served as textEvery site
  • SitemapA valid XML sitemap, ideally listed in robots.txtEvery site
  • Link headersHTTP headers on your homepage pointing agents to llms.txt, a markdown copy or API docsEvery site
  • DNS-AIDSigned agent discovery records in DNSLarge agent services

Content

Can an agent get a clean, token-light copy of your page?

  • Markdown for agentsA real markdown version when an agent asks for text/markdownEvery site

Bot access control

Do you tell AI bots what they may do with your site?

  • AI bot rulesrobots.txt rules for 20 AI bots, or a "User-agent: *" group that covers themEvery site
  • Content SignalsYour search, ai-input and ai-train choicesEvery site
  • Web Bot AuthA key directory for signed bot requests (only scored if you have one)Sites that run bots

API, auth, MCP and skills

Can an agent find your tools, APIs and login without reading your docs?

  • Agent Skills indexInstruction files that teach an agent your siteAny site
  • WebMCPTools your page hands to browser agentsSites with actions to offer
  • AI catalog (ARD)One manifest listing your agent-facing filesAny site
  • API catalogRFC 9727 list of your APIsAPI owners
  • MCP Server CardHow agents find your MCP serverMCP server owners
  • OAuth discoveryWhere an agent logs in to your appApps with logins
  • OAuth protected resourceWhich login server your API trustsAPI owners
  • auth.mdPlain-language signup steps for agentsApps with logins
  • A2A Agent CardA profile for an AI agent you run (shown, not scored)Agent operators

Extra checks (not scored)

Six checks the score leaves out, for the problems that lock agents out of real business sites or slow them down.

  • Readable without JavaScriptReal text in the raw HTML, before any script runsEvery site
  • Firewall test (simulated)Whether a simulated ChatGPT-User request gets turned awayEvery site
  • Response speedHow fast your server sends the first byteEvery site
  • Page sizeUncompressed HTML under the 2 MB Googlebot readsEvery site
  • Structured dataschema.org JSON-LD describing your business and pagesEvery site
  • llms.txtA markdown map of your key pagesOptional

Scoring

How the score and level work

Your score is the share of checks you pass. Pass 10 of 15 and you score 67. A check shown only for context (the A2A Agent Card, and Web Bot Auth unless you have it) doesn't count either way. A check that couldn't run counts as not passed. The six extra checks never count.

The level is the part to watch. It climbs in order, and each level needs the one before it:

  1. 0
    Not ReadyFewer than 2 of robots.txt, a sitemap and Link headers.
  2. 1
    Basic Web Presence2 of 3: robots.txt, a sitemap, Link headers.
  3. 2
    Bot-AwareLevel 1, plus AI bot rules and Content Signals in robots.txt.
  4. 3
    Agent-ReadableLevel 2, plus markdown for agents.
  5. 4
    Agent-IntegratedLevel 3, plus 1 of: an Agent Skills index, an MCP Server Card, an API catalog.
  6. 5
    Agent-NativeLevel 4, plus 1 of: all of those integrations, login metadata (OAuth or auth.md), or Web Bot Auth.

Your report can show two scores. The full score counts every scored check. The content score counts only the 7 basics (robots.txt, sitemap, Link headers, DNS-AID, markdown, AI bot rules and Content Signals). If you don't run an API, an app or an MCP server, the content score is the one that matters.

Which checks your site needs: if your website runs an API or app, use the full score that adds login, API and MCP files. If not, use the content score with the basics every site needs. Never publish fake files for things you don't run.

Here's why that matters. On September 14, 2026, ryandoser.com scored 20 out of 100. After a day of fixes with Claude Code, it reached 60 and Level 4, "Agent-Integrated." The checks still failing are DNS-AID (a draft almost no site has yet) and files for an API, a login system or an MCP server, none of which this site runs. Publishing fake versions of those files to chase a number would only feed agents bad information.

Beyond the score

What you get in your agent readiness report

A score on its own doesn't fix anything. The report is built for the person who has to make the changes.

  • A fix for every gap: a plain-English explanation and a copy-paste prompt for Claude Code or any coding agent, plus one button that copies every fix as a single prompt.
  • Your next level: the exact checks between you and the next level, so you know what to do first.
  • 6 extra checks: text without JavaScript, a simulated ChatGPT-User firewall test, structured data, response speed, page size and llms.txt. The first two catch problems that can lock agents out completely, even when the score looks fine.
  • Two scores: a full score, plus a content score for sites with no API, app or MCP server, so a blog isn't marked down for files it doesn't need.
  • Honest limits: WebMCP is found by reading the page's code, not by running it in a browser, so tools that load late can be missed. If your firewall turns the scan away, the report says so instead of guessing.

Fix order

How to make your website AI agent friendly

If you run a normal business, content or ecommerce site, fix things in this order. Each fix in your report comes with a prompt you can paste into Claude Code or any coding agent.

  1. 1

    First, make sure agents can get in and read the page

    Two of the extra checks catch problems that lock agents out completely: text that only appears after JavaScript runs, and a firewall that turns AI agents away. Neither changes the score, but either failure can keep agents from reading the page at all, so check these first.

    Biggest impact
  2. 2

    Level 1: 2 of robots.txt, a sitemap and a Link header

    Most sites already have the first two, which is enough. A Link header is one line in your host's header settings that points agents to your llms.txt or a markdown copy of the page.

  3. 3

    Level 2: add a Content-Signal line

    robots.txt needs rules that cover AI bots (a "User-agent: *" group counts) and a Content-Signal line with your yes or no for search, AI answers and AI training. Most sites stall on that one missing line. It records your preference. Bots aren't forced to follow it.

    Where most sites get stuck
  4. 4

    Level 3: serve markdown to agents that ask

    Some hosts and CDNs can switch this on with a toggle. Otherwise, a small piece of middleware can usually serve a markdown copy of each page. Fewer than 4% of top sites do this.

  5. 5

    Level 4: give agents something to use

    For a content site, the honest route is an Agent Skills index: one JSON file plus a short SKILL.md that explains your site. MCP Server Cards and API catalogs are for sites that actually run those things.

  6. 6

    Level 5: only if it fits what you run

    Level 5 takes one of three things: every integration file, login metadata (OAuth or auth.md), or Web Bot Auth for bots you operate. Those are for products with accounts, APIs or their own agents. A blog or local business can stop at Level 4. Never publish fake files to chase the number.

The AI SEO strategy guide covers the content side, and this breakdown of how Claude searches the web shows which bots fetch what.

Reality check

What agent readiness will not do for you

A perfect score won't get you recommended by ChatGPT.

Readiness covers what happens once an agent arrives. Getting named in an AI answer in the first place is a different job, and it runs on your content, your brand and what other sites say about you. No search engine has said a readiness score is a ranking factor, so be wary of anyone selling it as one.

Same goes for llms.txt. Google says you don't need AI text files to appear in its AI features, and there's no demonstrated citation benefit. The score here doesn't count it. It shows up as an optional extra. The evidence is in the AI search myths post. If AI search visibility is the actual goal, start by tracking where you show up today, then look at the AI visibility tools worth paying for.

Readiness is the floor. It makes sure agents can get to the information you publish and understand it when they do.

Next step

How to get named in AI search, not just agent-ready

Once agents can read your site, the next question is whether AI answers pick you. That depends on pages worth quoting, clear signs of who runs the business and other sites that mention you.

FAQ

Frequently asked questions

Is my website optimized for AI agents?

Run the checker and start with your level, not the number. Level 2 (Bot-Aware) means your site passes the basic discovery checks and publishes AI bot rules and Content Signals, which covers the basics for most business sites. Then look at the two extra checks that can lock agents out completely: text readable without JavaScript and the firewall test. If both pass, the page we tested returned readable text to our scanner, including its simulated ChatGPT-User request.

What is a good agent readiness score?

The level tells you more than the number. Level 2 is a solid floor for any business site, and Level 3 puts you ahead of most of the web, since fewer than 4% of top sites serve markdown to agents. The number is the share of checks you pass, so a site with no API or login loses points for files it doesn't need. Switch your report to the content score to see how you do on the basics alone.

Do I need llms.txt or an MCP server to be agent ready?

No. The score doesn't count llms.txt at all, and you don't need an MCP server to reach Level 4. Once you've cleared Levels 1 to 3, a content site can get there with an Agent Skills index, a small file that explains your site to agents. Only publish MCP, API or login files if you actually run those things.

Does agent readiness help my SEO or AI search rankings?

Not directly. Whether ChatGPT or Google AI Overviews recommend you comes down to your content and your reputation across the web. The overlap is access. If you block AI search bots or hide your text behind JavaScript, AI search tools can't cite pages they can't read. If getting named in AI search is the goal, my AI SEO System ($199, one time) audits your content and trust signals and walks you through the fixes.

How do I add Content Signals to robots.txt?

Add one line to the "User-agent: *" group in your robots.txt, for example: Content-Signal: search=yes, ai-input=yes, ai-train=no. That says your pages can show up in search and be used to answer AI questions, but not to train AI models. Set each yes or no to fit your business. It records your preference, and bots aren't forced to follow it. In a spot check of 82 well-known sites, 55 of the 56 stuck at Level 1 were missing only this line.

How long does it take to make a website agent-ready?

It depends on your host. ryandoser.com went from a score of 20 to 60 and Level 4 in one day of fixes with Claude Code. On many sites, Levels 1 and 2 come down to a robots.txt edit and one header setting. Level 3 usually takes the longest on hosts with no built-in markdown option.

Should I block AI bots?

Blocking training bots like GPTBot and ClaudeBot is a fair business call. Think hard before blocking the search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot), because they decide whether your pages can show up in AI search results. The user bots (ChatGPT-User, Claude-User, Perplexity-User) visit a page when a person asks an assistant to read it. OpenAI says robots.txt rules may not apply to those user-requested visits, so a firewall rule is the reliable way to stop them.

Is the checker free? Do you save my results?

Free, no signup, no email. Each result is cached for 10 minutes so repeat scans and share links load fast, then it expires. The scanner only makes read-only requests to public URLs on the site you enter.

Adoption stats are April 2026 figures from Cloudflare Radar. Radar's chart updates weekly, so current numbers may differ. Page last updated October 22, 2026.

The AI SEO System I use to rank in Google & AI search. Runs in Claude or any AI setup.

Get my system

Free AI Marketing Guide

Get the free AI Marketing Guide

The systems I use to run my agency solo. No coding needed to start:

Join over 10,000 subscribers getting my weekly AI marketing newsletter. Unsubscribe anytime. Privacy policy.