Free AI readiness check

Can ChatGPT, Claude and AI agents actually use your website?

A free check of what AI assistants and browser agents see when they visit your site. We test what they really do, not just which new standards you've adopted.

Scored by evidence: firewall blocks and JavaScript-only pages carry the weight. llms.txt and new agent protocols are shown, but barely count.

How we score

17 checks, weighted by what actually stops AI using your site

Each check passes, needs work, fails, or is marked “couldn't check” when our request was blocked or timed out. Checks we couldn't run never count against you. If AI assistants are blocked outright the score is capped at 30; if the page is empty without JavaScript it's capped at 45.

Check	Why it matters	Points
Can AI reach you?		35
AI assistants get the same page as browsers	When someone asks ChatGPT or Claude about your site, the assistant fetches the page itself. If your firewall or bot protection turns it away, the assistant has nothing to read and may say it can't access you. Source	15
robots.txt lets AI search and assistants in	AI search engines and assistants obey robots.txt. Blocking OAI-SearchBot keeps you out of ChatGPT search; blocking ChatGPT-User or Claude-User stops assistants reading your page when a user asks. Blocking training bots (GPTBot, ClaudeBot) is a business choice and isn't scored. Source	15
Page is indexable	AI search features are built on search indexes. A page that errors, redirects in loops or says noindex won't be shown or cited. Source	5
Can AI read you?		30
Main content is in the server HTML	ChatGPT, Claude and Perplexity read the HTML your server sends and don't run JavaScript. Content that only appears after JavaScript runs is invisible to them. Source	20
Key links are plain HTML links	AI crawlers find your other pages by following <a href> links in the HTML. Navigation built only with JavaScript leaves them stuck on the homepage. Source	5
Reasonable page speed and weight	AI fetchers work to tight time and size budgets. Slow or very heavy pages get cut off or skipped.	5
Can AI understand you?		15
Clear title, description, one H1 and a language	Titles, descriptions and headings are the first things a model reads to work out what a page is about. The lang attribute tells it which language to answer in.	6
Organization details in structured data	Organization or LocalBusiness JSON-LD states your name, website, logo and contact details unambiguously. Studies show little effect on AI citations, so this carries a small weight. Source	4
Sitemap present and fresh	A sitemap with recent lastmod dates tells crawlers which pages exist and which changed, so new content gets picked up.	5
Can AI agents act on your site?		10
Form fields are labelled	Browser agents such as ChatGPT Atlas and Claude in Chrome fill forms through the accessibility tree. A field without a label is a guess for them. Source	4
Actions are real buttons and links	Agents click things they can identify as buttons or links. Links without an href and buttons with no text or label are hard for them to find or understand. Source	3
A contact path is easy to find	The most common thing an agent does for a customer is get in touch: call, email or book. A plain contact link or phone number on the page makes that possible.	3
CAPTCHA on forms	A CAPTCHA can stop an agent acting for a real customer. That is often the right call for spam, so this is information only.	info
Ready for what's next?		10
Markdown for coding agents, tested with their real headers	Claude Code, Cursor, OpenCode and Copilot ask for markdown, which can use up to 80% fewer tokens than HTML. ChatGPT, Claude.ai and Perplexity don't ask for it yet, so this matters most for developer-facing sites. Source	5
Content Signals declared	Content Signals in robots.txt state whether your content may be used for search, AI answers and training. It is a statement of preference, not enforcement. Source	2
Agent integrations (MCP, API catalog, agent card)	MCP server cards, API catalogs and agent cards let agents use your product directly. They matter for SaaS, API and commerce sites; adoption is still below 1%, and a brochure site doesn't need them. Source	3
llms.txt	Harmless, but log studies show AI fetchers almost never request it and Google says its search ignores it. Information only. Source	info

Agent integrations count only for e-commerce and SaaS sites. For a business site they're shown for information and their points go to the checks that matter.

What this check can't tell you

Readiness isn't visibility

A good score means nothing technical stops AI using your site. It doesn't promise rankings or citations; that depends on your content and reputation.
Our AI user agent requests don't come from OpenAI's or Anthropic's IP ranges, so a block is a strong hint to check your firewall, not proof.
We check one page per scan. Run your key pages (pricing, contact, top product) separately.
Gemini and Copilot render JavaScript; ChatGPT, Claude and Perplexity don't. We test the stricter case.

For developers and agents

Use the checker as a JSON API

Send a GET request with the page URL and, optionally, the site type (business, ecommerce or saas). The response has the score, grade, each check with its evidence, and a fix prompt for every failure.

GET https://www.atilab.io/api/agent-readiness?url=example.com&siteType=business

It's rate limited per IP and cached for 15 minutes per URL. AI agents browsing this site can also call it as the WebMCP tool atilab_scan_site.

FAQ

Questions about AI agent readiness

What does "agent-ready" mean for a business website?

It means AI assistants (ChatGPT, Claude, Perplexity) can reach your pages, read the content without running JavaScript, understand who you are and how to contact you, and that browser agents such as ChatGPT Atlas or Claude in Chrome can use your forms and buttons. It does not mean you'll rank or be cited; it means nothing technical is stopping that.

How is this different from Cloudflare's isitagentready.com?

Cloudflare's scanner counts how many new agent standards a site has adopted. We weight checks by evidence of real impact: firewall blocks, robots.txt rules and JavaScript-only content carry the score, while llms.txt and agent protocols are shown but barely scored. We also replay the exact headers real coding agents send, and show what an AI can actually tell about your business from the page.

Should I block GPTBot and other AI training bots?

That's a business decision, so we don't score it. Blocking training bots (GPTBot, ClaudeBot, Google-Extended, CCBot) keeps your content out of model training. Blocking search and user bots (OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot) is different: it keeps you out of AI search results and stops assistants reading your page when a customer asks about you.

Do I need an llms.txt file?

It's harmless but there is little evidence it matters. Ahrefs found 97% of llms.txt files got no requests in a month, and Google says its search ignores them. We show whether you have one but give it no points; fixing JavaScript-only content or a firewall block will do far more.

Why does the result say a bot was blocked when my settings allow AI?

Our requests use the real AI user agents but come from our servers, not OpenAI's or Anthropic's IP ranges. Some firewalls verify the IP as well as the user agent, so a block here means your firewall treats AI agents differently and is worth checking, not proof that ChatGPT is blocked. Upstream WAF rules and Bot Fight Mode on Cloudflare often override AI allow settings without owners noticing.

Is the check free, and what do you store?

The score, every check and the AI reading are free with no signup. We keep results for 15 minutes so share links load quickly. If you ask for the PDF fix plan we record your email and score so we can follow up; nothing else is stored.

Want AI agents to do more than read your site?

We build MCP servers, WebMCP tools and AI agents that let customers book, quote and buy through the assistants they already use.

Book a 20-min review AI agent services

More free tools

All free tools