MACHINE 404 / The rubric
The rubricThe 18 checks, published in full
Here is the entire scoring rubric, weights included. Publishing it costs us the mystery and buys something better: you can audit the auditor, and you can fix every one of these yourself if you have the time and the appetite. Most people read it and decide they would rather buy the afternoon back.
| Check | Layer | Worth |
|---|---|---|
| Offers are machine-readable The single heaviest check. A price inside a paragraph is invisible to a shopping agent. | Offer | 12 |
| Organization schema present The heaviest single item after offers: it is the difference between a company existing and not. | Identity | 10 |
| Unknown URLs return a real 404 A soft 404 tells a machine every invented address is a real page. It poisons everything after it. | Access | 8 |
| Feed or agent endpoint A product feed or an endpoint an agent can actually transact against. | Transaction | 8 |
| robots.txt valid and declares sitemap Valid syntax, and it names the sitemap instead of leaving it to be guessed. | Access | 6 |
| Sitemap reachable It answers, it is XML, and it lists more than the homepage. | Access | 6 |
| FAQ schema answers agent questions Questions with answers, in a form that can be quoted back to a customer. | Answers | 6 |
| Homepage serves real content A 200 that is a redirect shell, or 2KB of nothing, is an empty page to an agent. | Access | 5 |
| Canonical URL declared Without it a machine cannot tell which copy is the original. | Identity | 5 |
| Structured contact details Phone, email and address as data, not as a picture in the footer. | Identity | 5 |
| HTTPS and single canonical host One address, not four that all answer. | Access | 4 |
| Title names the business 25 to 65 characters, and the brand is in it. | Identity | 4 |
| Meta description is quotable 70 to 175 characters. This is the sentence an assistant repeats. | Identity | 4 |
| sameAs links to verify identity Profiles a machine can cross-check to confirm you are one entity, not three. | Identity | 4 |
| Explicit AI crawler policy You name GPTBot, ClaudeBot, PerplexityBot and Google-Extended, and you decide. Silence decides nothing. | Transaction | 4 |
| OpenGraph card complete Four tags or more, so a share is not a grey box. | Identity | 3 |
| Breadcrumb structure Where this page sits, so the model can describe your structure. | Answers | 3 |
| Content readable without JavaScript Most AI crawlers do not run your JavaScript. What is left is what they read. | Transaction | 3 |
Why the weights are not equal
A missing OpenGraph card is worth 3. A price that no machine can read is worth 12, because it is the one that loses a sale rather than a share. A soft 404 is worth 8, because it corrupts everything measured after it: if every invented address answers 200, a crawler believes pages exist that never did.
Nothing here is a preference. Each one is either present or absent, and the same run on the same site tomorrow produces the same number.
What the free scan covers
The scanner on the product page runs eight of these in your browser in about ten seconds and shows you what an agent sees, with no email asked for. The paid report runs all eighteen properly from a server, writes the instructions for your exact platform, and re-scans afterwards so the before and after is a number rather than a claim.
See your own number first. It is free.
Type your address and the panel fills with your own data, the way an agent reads it. No email, no call, about ten seconds.
Scan my site See the prices