AI Visibility Infrastructure

Visibility Mesh Research

Visibility Mesh Research publishes measurements of how well machines can find, read, follow, quote and trust real websites. Every figure names its population, its date and the methodology version behind it, and every study stays frozen on the day it was published.

Explore the methodology Open the dataset

The living benchmark

Two populations, stated separately.

The benchmark keeps growing as we scan. Total sites analyzed counts every site the engine has scored under any version of the method. The current benchmark population counts only the sites scored on the current version, and the median comes from that population alone.

900+ Total sites analyzed Every distinct site our engine has scored, across all versions of the methodology.
750+ Current benchmark population Sites scored on the current methodology version, one latest score per site.
43.8 Median score in the benchmark Out of 100, across the whole benchmark population.
67% Share of the benchmark below 50 Sites that earned less than half of the available points.

Measured October 2, 2026 under rubric aivs-1.0. Our own websites are left out of every figure. The platform figures and the online store sub cohort are on the Research and Press page.

Published research

Studies, each frozen on its own date.

A study figure and a benchmark figure are two separate measurements. Read them side by side, and never as a trend.

Study 01

The State of AI Visibility 2026

Sample
216 live Shopify storefronts scored, from 229 scanned
Date
Method
AI Legibility Framework v1.0
Data
Per store data available on request
46Median AI Visibility Score out of 100
27Stores of 216 fully legible to AI
72%Stores that failed Entity Integrity

The median Shopify storefront in the study scored 46 out of 100, and Entity Integrity, the machine readable product and brand layer, was the weakest of the five metrics.

Limitations: one platform and one point in time. The 13 stores the engine diagnosed as unreadable were excluded from scoring and were not counted as zeros.

Read the studyFramework v1.0

Study 02

The State of AI Visibility on the Non Ecommerce Web

Sample
303 websites scored across 25 verticals, from 350 assessed
Date
Method
AI Legibility Framework v1.0
License
CC BY 4.0
48.1Mean score out of 100
35.3%Share of schema points earned, the weakest category
1.3%Homepages exposing FAQ schema

Across 303 non ecommerce websites, the schema category averaged 35.3% of its available points, well below every other category.

Limitations: homepages only, read on July 13, 2026. The 47 sites that could not be scored are reported as coverage gaps, and the feature comparisons show correlation, not proof of cause.

Read the study

Study 03

We Scored 245 Shopify Stores on What AI Can Read

Sample
245 Shopify storefronts, compared with 613 websites of all kinds
Date
Method
Rubric aivs-1.0
Source
The living benchmark on the day of publication
46.0Median Shopify score
47.2Median score of all 613 websites
67%Shopify stores below 50

Shopify storefronts scored a median of 46.0 against 47.2 for the wider web, and they lost points on structured data and answer ready copy rather than on technical basics.

Limitations: the figures record one reading of the benchmark on August 22, 2026, and they are not updated after that day.

Read the study

Study 04

Agentic Commerce Readiness

Sample
245 Shopify storefronts
Date
Method
Rubric aivs-1.0
Scope
Readability, the precondition for an agent purchase
6Stores in the Established band
0Stores in the Leading band

Of 245 Shopify storefronts read on August 22, 2026, 6 reached the Established band and none reached Leading, the level where an AI agent could work with a catalog confidently.

Limitations: the study measured readability and placed no purchase attempts, so it shows whether an agent could reasonably try, not completed transactions.

Read the study

Open data

AI Legibility Scores

Sample
596 websites from the living benchmark
Date
Method
Rubric aivs-1.0
License
CC BY 4.0
596Websites published by name
47.4Median score in the file

The AI Legibility Scores dataset publishes one row per website with its score, band, detected platform and scan date, so anyone can check the benchmark against the sites behind it.

Limitations: sites scanned at the request of a merchant, an app user, an agency or a customer count in every benchmark figure and are never published by name.

Dataset pageDownload the CSV

Method and editions

The method behind every number is public.

The methodology page explains how a score is produced and where its limits are, and the changelog records every change to the method with its date.

See where my site stands