AI Visibility Infrastructure
Visibility Mesh Measurement Methodology
This page explains how the Visibility Mesh score is produced, what it measures and where its limits are. The Visibility Mesh score measures how completely machines can find, read, follow, quote and trust the public pages of a website at the time of the scan. It runs from 0 to 100, and every score names the methodology version it was produced under.
The Five Gates
Five questions decide whether a machine can use a website.
The Five Gates describe the order in which a machine meets a website: it has to find a page before it can read it, follow its links, quote it or trust it. Each gate asks one diagnostic question. The Visibility Score is not a pass or fail sequence through the gates, because the score is the sum of the points a site earns across the five score categories, and a weak category lowers the total without cancelling the others.
-
01
Find
Can machines reliably discover the right pages?
-
02
Read
Can they understand what the business, its products and its services are?
-
03
Follow
Can they follow the relationships across the site?
-
04
Quote
Is there specific, attributable information an answer can use?
-
05
Trust
Can the important claims be corroborated?
01 FIND: what the scan reads
- Access is checked for 8 AI crawlers, one at a time.
- robots.txt rules are read for each page path.
- The scan obeys a noindex in the meta tag or the X-Robots-Tag header.
- Sitemaps and sitemap indexes are followed.
- Only pages with real, readable content are scored.
02 READ: what the scan reads
- Every JSON-LD block is read in full.
- Organization, WebSite and Person markup is checked.
- Product markup and identifiers are checked.
- Headings, language and visible text are extracted.
- The platform is detected before any fix is written.
03 FOLLOW: what the scan reads
- Page structure and site architecture are scored.
- Internal links in the body of each page are counted.
- Breadcrumb markup and visible trails are read.
- Lists and collections that group pages are read.
- Entities are compared across the whole scan.
04 QUOTE: what the scan reads
- Copy is scored on whether it states the answer early.
- Visible FAQs and FAQPage markup are read.
- Article and author markup is read.
- Visible published and updated dates are read.
05 TRUST: what the scan reads
- sameAs links to real profiles are checked.
- About, contact and press pages are located.
- Authorship and organization identity are compared.
- Ratings in markup must match what the page shows.
- Markup that contradicts the page is flagged.
| Score category | AI Legibility Framework v1.0 metric |
|---|---|
| Technical Foundations | Front Door |
| Schema and Structured Data | Entity Integrity |
| Semantic Architecture | Mesh Integrity |
| AI Readability | AI Readability |
| Authority and Operational Discipline | Authority Signal |
Signal families
What the engine collects before it scores anything.
- A
Access signals
The scan reads robots.txt and tests the rules for each AI crawler of OpenAI, Anthropic and Perplexity separately. It also fetches content pages with those crawlers' user agents and compares what they receive with what a browser receives.
- B
Validity signals
Error pages, soft 404 pages, bot walls, password pages, empty renders and noindex pages are recognized and kept out of the score. A site with too few readable pages receives a Crawl Diagnosis that names the confirmed cause, so an unreadable site is never scored as a zero.
- C
Structure signals
For every readable page the engine extracts the title, the meta tags, the canonical link, the heading order, every JSON-LD block and the visible text. It also records image alt coverage, body links, the page language, hreflang and microdata types from the raw HTML.
- D
Site signals
Sitemaps, llms.txt, agents.md, the platform the site runs on and whether it sells online are recorded once per scan. Homepage performance comes from Google PageSpeed Insights.
- E
Grounding facts
The engine keeps a list of facts it has proven from the page itself, such as which schema types are present across the whole scan. A finding that says something is missing is checked against those facts before it can reach a report.
Measurement process
How one scan runs, from the first request to the report.
- 01
Select pages
The free scan reads 5 key pages. The paid scans read 100, 150 or 400 pages, depending on the plan, and the Catalog Assessment reads every product in a store's export.
- 02
Load each page
Each page is loaded in a browser, with a plain fetch as a fallback, and the crawler waits for the page to settle before it captures it.
- 03
Score each page
Each readable page is scored on the five categories under one fixed rubric version. A category's score is its average across the scored pages, and the total is the share of available points, from 0 to 100.
- 04
Check every finding
Every finding must rest on content read from the live page. Findings are grounded against the whole scan, so a fact that holds for one page is never reported as a fact about the whole site.
- 05
Order the fixes
Findings come back in plain English with a severity and the page they were read from, and the fix plan puts them in the order we would make them.
- 06
Stamp the result
Each scan stores the rubric version and the build of the engine that produced it, so any score can be traced to the exact method behind it.
Score interpretation
The band tells you more than the last decimal.
The score is the share of available points a website earns across the five categories, from 0 to 100. A perfect 100 is not the goal. AI systems and the standards they read keep changing, so the band a site sits in and the direction it is moving matter more than the last point.
| Band | Score | What it means |
|---|---|---|
| Starting Point | 0 to 39 | AI struggles to read the website or place it as a trustworthy source. |
| Developing | 40 to 59 | AI crawlers can read parts of the website, and important gaps remain. |
| Established | 60 to 79 | Most of what we measure is readable to AI crawlers. |
| Leading | 80 to 100 | Nearly everything we measure is readable to AI crawlers. |
Rescan methodology
A rescan compares like with like.
- 01
Same method
A rescan is compared with a baseline produced under the same rubric version. For Managed AI Visibility, the rescan also uses the same depth as the baseline.
- 02
Comparable pages
The comparison uses a baseline that read a comparable set of pages, so a change in which pages were sampled is not reported as a change in the site.
- 03
A threshold for change
Two scans of an unchanged website often land within about 2 points of each other. Our reports treat a move of less than 2 points as no measurable change, so a bigger move usually means something on the site changed.
- 04
Traffic is read separately
A rescan shows what changed in what a crawler can read. Traffic from AI assistants and organic search is read from the site's own analytics and reported next to the score, never folded into it.
Reading the score correctly
What the score means, and what it does not mean.
The score means
- It measures how much of a website an AI crawler could read and use on the day of the scan.
- It is comparable with any other score produced under the same methodology version.
- For the issues the scan detects, the report lists the findings and a fix for each one.
- A change between two comparable scans shows a change in what machines can read.
The score does not mean
- It does not count how often ChatGPT or another assistant names a business. A separate AI mentions check records that.
- It is not a ranking in Google or in any AI answer.
- It does not include traffic, rankings or backlink data, which is why a well known site can score low and a small one high.
- It is not a promise that any assistant will cite, recommend or rank a website. Search engines and AI assistants decide what they show.
Limitations
Where the measurement stops.
- 01
A score is a snapshot of the public pages on the day of the scan. Websites change, so an old score describes an old website.
- 02
The free scan reads a sample of 5 key pages. A deeper scan reads more pages and can find problems a sample does not reach.
- 03
The scan reads only what any visitor or crawler can see. Pages behind a login or a password are outside its reach.
- 04
Scores are repeatable but not identical. Two scans of an unchanged site often land within about 2 points of each other.
- 05
Being readable is a precondition for being cited. Authority, freshness, what other sources say and how closely a page matches the question also decide what an assistant says.
- 06
Each published study was measured once, on its own date and population, and is not revised when the method changes. A study figure and a benchmark figure are two separate measurements.
What Visibility Mesh is not
Some things we will not do to a website.
- We do not stuff keywords into pages.
- Mass produced AI content is no part of the work.
- Schema goes only where the page supports it, so there is no schema spam.
- Links are never bought, and backlink packages are not for sale.
- Nobody can promise a number 1 spot in ChatGPT, and we do not.
Visibility Mesh improves the structure and evidence that machines and humans use to understand a business.
Measure your own website on the same method.
The free scan uses the same engine and the same methodology version as every figure on this page, so your score is directly comparable.