Scored here
- Whether named agents are blocked in robots.txt
- Whether the page parses without running JavaScript
- Whether a model can say what the business does
- Whether it answers fast enough and over HTTPS
Before an agent can do anything on your behalf it has to get in and understand what it is looking at. An agent readiness check scores that.
Agent readiness has two halves. This scores the first one properly and does not touch the second.
They happen in order, and the third is where nearly every site we check loses its points.
We read your robots file against twelve AI crawlers by name, then your sitemap and llms.txt.
We take the page before JavaScript runs: headings, structured data, and whether the words are there.
HTTPS, a redirect that resolves, and a page that arrives before a crawler gives up.
Five categories, weighted. Every check behind the number is set out on the methodology page.
Your robots file, read against twelve crawlers
Schema, heading order, sitemap, llms.txt
What a model can name from the page alone
Mobile performance, and how fast the page answers
HTTPS, and a redirect that resolves
All of it on screen the moment the scan finishes. No summary version, no upgrade.
AI readiness score
An AI engine gets about half of what it needs from this page.
Also in it:
0 to 100, from 5 weighted signals, with a grade and a band.
Five of them: crawler access, structure, understanding, speed, security.
12 agents across 6 engines, checked one at a time.
Five facts: your business, services, products, industry and location.
Every one, with what it is worth and how much work it is.
Four: ChatGPT, Claude, Gemini and Perplexity, weighted separately.
One page, read live, with your AI readiness score at the top and the full audit under it.
Start at the five category scores.
They tell you in one look whether this is a writing problem or an engineering one. Usually it is writing.
Start at the ranked fixes.
Each carries its points and its effort, so you know what you are buying before you ask.
Send the report link.
It has its own address and shows its arithmetic, so it survives a sceptical reader.
Whether an AI agent, something acting on a person’s behalf rather than just answering them, can use your website. It has two halves: whether the agent can read the page at all, and whether you publish the files that tell it what actions are available. This checks the first half.
No. We do not fetch agents.json, anything under /.well-known/ or an MCP card, and we would rather say so than let the page imply it. The guide to agent protocols covers what those files are and who is actually reading them, which today is fewer systems than the discussion suggests.
Run the check and you will have the foundation answered in under a minute: whether 12 named crawlers across 6 engines can reach you, whether your structure parses, and whether a model can say what you sell. A manifest on top of a page an agent cannot read is worth nothing, so this is the half to settle first.
In what happens after the reading. An answer engine reads your page and quotes it. An agent reads your page and then tries to do something: compare, book, buy, fill a form. The reading requirement is identical, which is why one scan covers both, and the acting requirement is where agents need more from you.
Usually the same three things that stop any machine. The crawler is blocked in robots.txt. The content only exists after JavaScript runs. Or the page never states plainly what the business is, so the agent cannot tell whether it is even in the right place. Those are all in the report.
The named crawlers from the large labs generally do, and that is what we test against. An agent operating on behalf of a signed-in person may identify differently or not at all, so robots.txt is the floor rather than the whole picture. Blocking a crawler reliably costs you the engine behind it.