What is an AI readiness score for a website?

What the score actually is

It is a weighted sum of six technical checks against one domain, worth 100 points when all six run.

There is no standard behind it. No vendor, standards body or search engine publishes a definition of “AI readiness”, so every tool offering a score is publishing its own opinion about what matters. That is defensible — the underlying checks are real and measurable — but it means a score is only as honest as its published weighting.

So ours is printed on the tool itself, in full, before you run anything. A score whose arithmetic you cannot inspect is a number to be suspicious of, including ours.

The six checks and what each is worth

Two checks about access carry 40 of the 100 points, because nothing else matters if a crawler is turned away.

The ordering is an argument, and you are entitled to disagree with it. Access is weighted highest because a blocked crawler makes every other check irrelevant; schema is weighted low because it helps a machine read a page it has already fetched, which is a smaller job than being allowed in.

Why a check that did not run is not a zero

A check that fails to complete leaves the total entirely, rather than scoring nothing.

If your FAQ page times out, we cannot say your FAQ markup is bad. We can only say we did not read it. Scoring that as zero would invent a finding, and it would punish a slow server by pretending to know something about its markup.

So the denominator shrinks: six checks that all run share the full 100 points; five that run are scored against whatever those five are worth, and the result says which one is missing and why. A score of 72 against the five checks that ran is more honest than a full-weight score with a guess in it.

This is the same rule the rest of this site runs on: not measured is a state of its own, never a bad result wearing a number.

Readiness is not visibility, and the gap is the whole point

A perfect 100 does not mean any assistant will name your business.

The score measures whether your site can be read. Whether an assistant mentions you depends on what it was trained on, what it retrieves at the moment someone asks, who your competitors are, and how the question was phrased — none of which lives on your server and none of which a technical check can see.

The relationship runs one way. A site that blocks crawlers, hides its content behind JavaScript and describes itself nowhere is very unlikely to be quoted accurately. A site that fixes all of that may still never be named. Readiness removes excuses; it does not create demand.

If the question you actually have is “does ChatGPT mention us”, that is asked rather than inferred — you put the questions your customers ask to the assistants and record what comes back.

Checks 1 and 2: crawler access, and what the live fetch really proves

The first check reads your rules; the second tests whether your server honours them.

Access resolves your robots.txt per AI user-agent, because the answer differs by bot: a retrieval crawler that feeds answers is worth more to you than a training crawler, and blocking them is a different decision. The crawler checker shows the same resolution bot by bot.

The live fetch then requests your homepage while identifying as each bot, and records the HTTP status. This catches the gap that robots.txt cannot: a CDN rule, a firewall or a bot-management product answering 403 to something your robots.txt explicitly allows.

Its honest limit: the request comes from our IP, not the bot’s. A site that verifies crawlers by address or by reverse DNS may answer us differently from how it answers the real thing. What this check proves is that a request carrying that user-agent was served or refused — which is enough to catch the common misconfiguration, and not enough to certify what Google’s own infrastructure receives.

Check 3: does your content exist before JavaScript runs

Twenty points, because this is the failure that looks like nothing is wrong.

Most AI crawlers fetch HTML and do not execute scripts. If your copy is injected by a framework after load, the crawler receives an empty shell while your visitors, and your own eyes, see a full page. Nothing in your analytics will ever tell you.

The check compares what arrives with no JavaScript against what a browser builds, and the render check shows both copies side by side when you want to see the difference rather than a score.

Checks 4 and 5: Organization and FAQ schema

Twenty-five points between them, and they are doing different jobs.

Organization markup states who you are in a form a machine does not have to infer: the name, the URL, the profiles that corroborate you elsewhere. That matters because assistants prefer facts they can cross-check, and an entity they can resolve is easier to name correctly. The generator writes one from what you type and invents nothing.

FAQ markup is weighted lowest of the six, deliberately. Google removed its FAQ rich result, so the search-feature reason for it is gone; what remains is that it states, machine-readably, which questions your page answers. Our check requires the answer text to be on the page, not only in the markup — what each schema finding means goes through that in detail.

Check 6: is the page shaped like an answer

Fifteen points for writing that can be lifted without being rewritten.

This reads your headings and opening sentences. A heading phrased as the question someone asks, followed by a sentence that answers it without depending on the paragraph before, can be quoted accurately by a machine. A heading reading “Our Approach” followed by “That’s why we believe…” cannot.

It is the softest of the six and the one most worth your own judgement: it measures a property of good writing that happens to also serve machines, and the content checker reports it per page rather than as a single number.

What a good score looks like

Above 80 with nothing unrun means the mechanics are not your problem.

Below that, the number matters far less than which checks failed. Losing 25 points on crawler access is a different emergency from losing 10 on FAQ markup: one means assistants are being turned away at the door, the other means a page is slightly harder to quote.

So read the rows, not the total. The score exists to rank your next hour’s work, and a tool that gives you a number without telling you which input produced it has given you a feeling rather than a brief.

Why two tools give you different numbers

Because they are scoring different things and calling both “readiness”.

One tool weights page speed heavily; another includes backlinks; a third scores llms.txt, which no vendor’s documentation says its crawler reads. None of them is lying, and none of them is measuring the same quantity. A twenty-point difference between two scores usually means a twenty-point difference of opinion.

The only useful comparison is between your own score today and your own score after you changed something — and even then, only if the weighting did not move underneath you.

What to actually do about each check

Work down the weights, because that is the order in which these cost you something.

1. Crawler access (25). Read your robots.txt per bot rather than as a whole. The common defect is a blanket rule written to stop scrapers that also turns away the retrieval crawlers behind AI answers. Decide training and retrieval separately; they are different trades.
2. Rendering (20). If the main content is injected by JavaScript, the fix is server-side rendering or static generation for the pages that matter. If that is a project rather than an afternoon, start with the pages you would most want quoted.
3. Live fetch (15). A 403 to a bot your robots.txt allows is nearly always a CDN or firewall rule, not your application. Look at the edge before the code.
4. Organization schema (15). One block, on one page, with the profiles that corroborate you. Delete the duplicates your theme and plugin both emit rather than reconciling them.
5. Answer structure (15). Phrase a heading as the question someone types, and make the sentence after it stand on its own. This is an editing habit, not a project.
6. FAQ schema (10). Keep it accurate and keep the answers visible on the page. Expect nothing in the search results from it.

Only the first three can be fixed by someone with server access alone. The last three are content work, which is why a readiness score that produces a developer ticket and nothing else has only told you half the story.

Why we print the weighting, and what would change it

Because a score is an opinion, and an opinion you cannot inspect is a brand asking for trust it has not earned.

These weights are a judgement about what currently costs a site the most, and judgements age. If retrieval crawlers started ignoring robots.txt, access would be worth less. If an assistant vendor published evidence that a particular signal changes what it quotes, that signal would earn points it does not have today. Neither has happened, so the weighting has not moved.

What will not happen is a weight changing because it makes scores look better. A score that drifts upward across a customer base without anything improving is a marketing instrument, and it would make every number this tool has ever produced worthless.

What ours cannot see

• Whether a crawler ever came. That is in your server logs, and reading them is a separate job.
• What any assistant says about you. It never asks one. Nothing in a readiness score is evidence about an answer.
• Pages other than the one it checks. It scores a domain from its homepage and the files at its root; a deep page can be far worse than the score suggests.
• Whether your content is any good. No technical check can tell you that, and a site can score 100 while deserving to be ignored.

FAQs