How the score works.
Updated 3 October 2026
Every score on this site is produced by the same nine checks, weighted the same way, on every business. This page is the whole method. It is rendered from the scanner's own configuration, so it cannot disagree with the score you were given.
The nine checks
We fetch your pages the way an AI assistant's agent would — plain HTTP, no browser, no JavaScript — and measure what it could actually learn.
| Points | What we check | Why it counts |
|---|---|---|
| 25 | AI crawler access | Whether robots.txt lets the assistants' crawlers read the site at all. Blocked here and nothing else can help you. |
| 20 | Structured data (schema.org) | Machine-readable markup saying what this business is, where it is and how to reach it. Without it an assistant is guessing from prose. |
| 15 | Content readable without JavaScript | Agents fetch raw HTML and mostly do not run scripts. Content that only appears after JavaScript runs is invisible to them. |
| 15 | Prices and availability machine-readable | An assistant asked to recommend something in a budget cannot recommend a business whose prices it cannot find. |
| 10 | Automated access (WAF / bot protection) | Bot protection that challenges every automated request also challenges the assistant fetching the page for a real customer. |
| 10 | Policies legible (terms, cancellation) | Assistants are cautious about recommending a business whose terms they cannot read. |
| 5 | llms.txt guidance file | A short file telling assistants what the site is. Not yet load-bearing, and free to add. |
| 5 | Sitemap discoverable | Lets a crawler find every page instead of guessing at URLs. |
| 5 | Agentic commerce endpoints | Whether an agent could transact directly rather than scraping a form. Forward-looking; almost nobody scores here yet. |
| 110 | Total available | Your score is the points you earn as a percentage of the points we could actually measure. |
The bands
The score is a percentage, and the letter is shorthand for where it sits.
An agent can reach the site, identify the business, and find prices and terms.
Readable and identifiable, with gaps an agent has to work around.
An agent can see the business exists but cannot answer much about it.
Most of what an assistant needs is missing or unreadable.
An agent fetching these pages learns almost nothing.
What the score is not
It measures what your website makes available to an agent. It does not measure whether assistants recommend you, and the two can legitimately disagree: a business with strong editorial coverage can be recommended constantly despite an unreadable site, and a perfectly marked-up site can go unmentioned because nobody writes about it.
That is why the report has two halves. The readiness score is this page. The other half asks the four assistants directly and counts how often they name you — and that number is observed, not inferred from your HTML.
When we cannot measure something
If a check cannot be measured — the site times out, or bot protection refuses us — it is excluded from the score rather than marked zero, and the report shows a confidence figure: the share of the 110 points we were actually able to test. A high score at low confidence means we saw too little to judge, and the report says so instead of quietly rounding in your favour.
The agents we check for
Agents that answer a live question (7)
These fetch a page because someone asked a question just now. They decide whether you are in today’s answer.
- OAI-SearchBot OpenAI
- ChatGPT-User OpenAI
- Claude-User Anthropic
- Claude-SearchBot Anthropic
- PerplexityBot Perplexity
- Perplexity-User Perplexity
- Amazonbot Amazon
Crawlers that gather training data (8)
These feed what a model knows in general. Whether to allow them is a publishing decision, not a mistake — plenty of businesses block them on purpose, and we never count that against you as an error without saying so.
- GPTBot OpenAI
- ClaudeBot Anthropic
- anthropic-ai Anthropic
- Google-Extended Google
- Applebot-Extended Apple
- CCBot Common Crawl
- meta-externalagent Meta
- Bytespider ByteDance
How often it changes
Weights change rarely, and when they do we date the change on this page. The checks themselves run fresh every time — nothing here is cached from a previous scan, so fixing something and re-running will show the new score within minutes.