Methodology
How the score works
Seventeen checks across four pillars. Every one of them is something we actually observed on your public pages, and every finding shows the evidence so you can check our work.
Weighting
Reach 25, Comprehend 25, Act 35, Complete 15. Act carries the most because it is the only pillar that decides whether a transaction can complete, and because it is the one almost nobody has done, which makes it the available work.
Every check
| Pillar | Check | Points | What it means |
|---|---|---|---|
| Reach | AI crawlers allowed in robots.txt | 10 | Checks all 12 major AI crawlers. Blocking a live browsing agent such as ChatGPT-User or Claude-User is scored as critical, because that is a customer being turned away, not a training scraper. |
| Reach | Site serves a request from an agent | 7 | We request your homepage with an agent user agent. A 403 or 401 means your bot protection is refusing agents while humans see the page normally. |
| Reach | Secure and predictable address | 4 | HTTPS, and whether the final URL matches what was asked for. |
| Reach | Sitemap an agent can follow | 4 | A sitemap.xml, and whether robots.txt points to it. |
| Comprehend | Machine readable business facts | 10 | JSON-LD structured data: business identity, what you sell, FAQ answers, phone and address, opening hours. Each is scored separately. |
| Comprehend | Content readable without JavaScript | 7 | Counts visible words in the raw HTML before any script runs. A page that is empty until React hydrates is empty to a large share of agents. |
| Comprehend | llms.txt briefing for AI | 4 | A plain language file at /llms.txt telling an AI what your business is. The cheapest item on the list. |
| Comprehend | Page states what it is | 4 | Title, meta description, exactly one h1, and a lang attribute. |
| Act | WebMCP tools an agent can call | 14 | The heaviest single check. Tools registered through navigator.modelContext. A static scan can only find the call site in your bundle, so a confirmed live enumeration scores higher than a signature match, and the report says which one you got. |
| Act | Machine readable capability manifest | 5 | A manifest at one of /.well-known/mcp.json or /.well-known/model-context.json that lets an agent discover what you can do before loading a page. |
| Act | Forms a machine can fill in | 9 | Every visible field scored for a name attribute, a real label tied by id, an autocomplete token and a specific input type. A captcha on the page costs points here as well. |
| Act | Phone and email as real links | 4 | tel: and mailto: links. A number set in an image or assembled by script is invisible to an agent. |
| Act | Booking or checkout the agent can walk through | 3 | Whether your conversion runs inside a third party iframe the agent cannot see into. |
| Complete | No captcha blocking the last step | 6 | reCAPTCHA, hCaptcha, Turnstile and others. An agent that has done everything right still cannot submit. |
| Complete | Public information stays public | 3 | Whether a login wall sits in front of what you offer. |
| Complete | Second request still works | 3 | We ask twice. Comparing three products is three requests, which is shopping, not an attack. |
| Complete | Few third party frames in the way | 3 | Each embedded window is a place an agent can lose the thread. |
What we do not do
We fetch public pages only. We do not log in, submit forms, place orders, or change anything. We request your homepage a small number of times and stop. If a check could not be run, the report says so rather than scoring it as a pass or inventing a result.
Want the fixes done rather than listed?
Sneeze It builds the WebMCP tools, the structured data and the agent readable forms. Scan your site, then talk to us about the ones worth doing first.
Run a scan