AI Readiness Score API: What the 0-100 Grade Measures

An AI readiness score API returns a 0 to 100 grade for how readable a website is to AI crawlers and answer engines. BusinessMCP’s version scores five weighted checks: crawler access in robots.txt, structured data, llms.txt, metadata and sitemap discoverability. It measures whether machines can read the site, not whether any assistant will cite it.

The five checks and their weights

BusinessMCP publishes the weights for its AI readiness grader, and the API endpoint runs the same grade for $0.03 per call through its website checks API.

Check	Points	What it looks at
AI crawler access	30	Whether robots.txt lets crawlers such as GPTBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot read the site
Structured data	25	Organization and WebSite identity, plus types like FAQPage, Product, Article and BreadcrumbList
Metadata basics	20	Title, description, canonical and Open Graph
llms.txt	15	Whether an llms.txt file exists
Discoverability	10	A reachable sitemap

The grader page does not explain how partial credit is awarded within each category, so treat the score as a heuristic summary and read the per-check detail when you act on it.

Key takeaway: nearly a third of the score is simply whether crawlers are allowed in. A beautiful site that blocks every AI crawler cannot score well, whatever else it does.

What the grader fetches

Per the vendor, the browser tool makes four small requests, all to the domain: robots.txt, llms.txt (and llms-full.txt), the homepage HTML capped at 500KB, and a check that sitemap.xml is reachable. The API documentation adds that readiness reads served HTML, robots.txt and llms.txt. It runs no JavaScript.

That has a practical consequence. A single-page app that renders everything client-side may present an almost empty document to the check. The grade then reflects the empty shell, which, to be fair, is also what a non-rendering crawler sees. The API says results flag when this is likely, so read those flags before drawing conclusions.

What a score cannot tell you

Be wary of these misreadings:

1. “90 means we will be cited.” No. Citation depends on content quality, authority, the query and the product. The vendor states this explicitly.
2. “llms.txt moves the score, so it moves rankings.” The 15 points are for presence of the file. BusinessMCP’s own guidance calls llms.txt an emerging convention, not a ranking factor, and says no AI vendor has committed to reading it. Our note on checking llms.txt covers this in detail.
3. “Same score, same site quality.” Two sites can both reach 70 by different routes, one strong on structure and one on access.

A better use of the score is as a regression alarm. If a site drops from 82 to 52 after a deploy, something broke, probably robots.txt, and that is worth knowing the same day.

Using the grade in a workflow

Three patterns work well.

Portfolio triage. Grade every domain, sort ascending, and open the lowest ten. BusinessMCP lists “auditing a portfolio” as a primary use, with a fix list per site. Our guide to bulk domain audits shows the loop.

Pre-launch gate. Grade a staging or new domain before launch and fail the pipeline below a threshold you choose. Pick the threshold from your own baseline, not from a generic number.

Change detection. Store each score and alert on a drop. Combine it with the finer checks: AI crawler access tells you which crawler changed, which a single number cannot.

Acting on a low score

Work in order of weight.

1. Fix robots.txt first. Remove accidental blanket disallows and decide deliberately which AI crawlers you want.
2. Add or repair structured data for your organisation and key content types.
3. Complete metadata: unique titles, descriptions, canonical tags and Open Graph.
4. Publish a short, curated llms.txt if it suits you; do not expect miracles.
5. Make sure a sitemap exists and is referenced.

Deciding which crawlers to allow is a business choice, not a technical default. Some publishers block training crawlers and allow search ones; the right call depends on your goals, and this site makes no recommendation.

Cost of grading

At $0.03 per call, grading 100 domains is $3. Weekly runs on a 25-domain portfolio cost under $40 a year. The free credit covers experimentation: $5 for 7 days per the vendor’s data API page. For fuller scenarios, see website data API pricing.

Scoring rules and prices can change, so check the vendor pages for the current weights before you build thresholds on them.

Reading a score alongside the finer checks

A single number hides which part is weak, so combine it with the narrower checks. If the grade drops, query crawler access first, because it carries the most weight. If crawler access is fine, inspect structured data and metadata by hand. Add the WebMCP result separately; it is not part of the 0 to 100 score as the grader page describes it, and the WebMCP checker article explains why it is a different question.

Symptom	First thing to check
Score fell by 25 or more overnight	robots.txt, likely a blanket disallow
Score stuck low despite good content	Structured data and metadata gaps
Client-rendered site scores very low	Whether the check flagged a rendering limit
Good score, no AI citations	Content authority, which the score does not measure

Setting thresholds

Choose thresholds from your own data. Grade your best three sites and your worst three, note the range, and set a “review” line a little below your typical good score and an “alarm” line well below that. A generic number such as 80 means little without that baseline. Revisit thresholds quarterly, and whenever the vendor changes scoring rules, which you can detect by an unexplained shift across all of your domains at once.

Where this fits in a wider strategy

Readiness is a precondition, not a strategy. BusinessMCP’s guides on generative engine optimization discuss the broader work of being cited, which depends on content and authority as much as machine readability. Use the grade to remove avoidable technical obstacles, then judge your results by the outcomes you care about rather than the score itself.

Questions to ask before trusting any readiness grade

• Are the category weights published? BusinessMCP’s are.
• Is partial credit explained? On the grader page it is not, so treat sub-scores as approximate.
• Does the tool fetch only your own domain, and store nothing? The vendor says the browser tool does both.
• Does the grade claim outcomes it cannot measure? A good one says plainly that it does not.