Both Visibility Mesh studies of 2026 found the same weak spot: the layer of structured facts that tells a machine who a business is and what it sells. We scanned 216 Shopify storefronts on June 28, 2026, and 72 percent of them failed Entity Integrity. They earned less than half of its points. We scored 303 non ecommerce websites in July 2026. Their structured data pillar averaged 35.3 percent of its available points, the lowest of the five pillars.
Some websites failed before any of that could be measured. The engine could not read 13 of the 229 storefronts in the June study. In the July study, 47 of 350 websites produced no score. Every figure on this page keeps its study's population and date. None of it shows that a problem caused a loss of traffic or of AI mentions.
Margareta Petrovic, who founded Visibility Mesh, compiled this report from two frozen studies, and we published it on . We publish every figure with its population and date, and we do not update a frozen study with newer data.
What we measured, when and on which websites
Population, sample size and methodology version
| Study | Population | Targeted or scanned | Scored | Date |
|---|---|---|---|---|
| The State of AI Visibility 2026 | Live Shopify storefronts | 229 scanned | 216 scored | June 28, 2026 |
| The State of AI Visibility on the Non Ecommerce Web | Non ecommerce, non government websites, 14 per vertical across 25 verticals | 350 targeted | 303 scored | July 12 and 13, 2026, published July 13, 2026 |
Both studies used the same version of our scoring method, the AI Visibility Score 1.0. Our AI Legibility Framework v1.0 describes it. The two studies report their results under different category names. The storefront study reports five metrics, and the non ecommerce study reports five pillars. We quote each one in its own terms.
What the scan can and cannot see
The scan reads each page as it is served. That is the way a crawler receives it. It then scores five categories. The non ecommerce study read homepages only. Neither study read Search Console, analytics or AI answers. So neither one can say whether a site was indexed, visited or cited. The figures describe what a crawler could read on the day of the scan.
The problems that most often stop a website from being read
In the storefront study, a large share of stores failed every one of the five metrics. Failing means earning less than half of a metric's points.
| Metric | What it tests | Share that failed |
|---|---|---|
| Entity Integrity | The brand and product facts in structured data, and whether they match the page | 72 percent |
| AI Readability | Whether key facts are stated plainly or buried in marketing prose | 71 percent |
| Authority Signal | Freshness, identity consistency and trust markers | 68 percent |
| Mesh Integrity | Whether the pages link into one connected map | 58 percent |
| Front Door | Whether a crawler can retrieve and read the page at all | 41 percent |
None of the 216 storefronts earned 80 percent of the points on any of the five metrics.
The non ecommerce study points at the same layer. Across the 303 websites, the engine wrote 11,633 findings. Thin structured data and a missing block of questions and answers made up 54.4 percent of them. Google stopped showing FAQ rich results on May 7, 2026. So we read that finding as missing answers, which still matter as content a machine can read.
We also audit our own findings, and that audit confirmed more than 99.9 percent of the 11,633 findings against the live pages.
How common entity and structured data problems are
Entity problems were the most common failure in the storefront study. Of the 216 storefronts, 72 percent failed Entity Integrity. Entity Integrity checks whether the brand, its products and their attributes are stated consistently in the data AI systems read.
The non ecommerce web shows the same gap in a different form. Of 299 homepages the scanner could reach in July 2026, 52.2 percent carried any JSON-LD. Organization markup appeared on 42.5 percent, and FAQPage markup on 1.3 percent. The 1.3 percent describes homepages only.
Among the 275 non ecommerce sites with both a score and a homepage check, sites with any JSON-LD averaged 53.7. Sites without it averaged 41.0. That is a 12.7 point difference. Part of it is built into the score, since the score rewards structured data. We state that plainly so nobody reads it as proof that markup lifts a website.
Websites we could not read at all
In the storefront study, the engine could not read 13 of 229 storefronts scanned on June 28, 2026. Bot walls, empty JavaScript renders and error pages stopped it. We left those 13 out of the scoring instead of counting them as zeros.
In the non ecommerce study, 47 of 350 websites produced no score. Most of them had heavy bot protection or a homepage built only in JavaScript. Four of the 47 never finished scanning, so we cannot say why those four failed.
A bot wall that stops our crawler does not prove that it stops Googlebot. Services such as Cloudflare keep a list of verified bots they can let through. We have not yet classified the reasons across all our scans. Until we do, we publish no single share for the wider web.
Online stores compared with other websites
Our two frozen studies do not settle this. We would rather say so than stretch them. The median storefront in the June study scored 46 out of 100. The median non ecommerce website in the July study scored 49.1. The samples differ in platform, date and depth. The July study read homepages only. We do not treat that small gap as a finding about stores.
The live benchmark mixes online stores and other websites. Its figures change every day. The Research and Press page carries the current split between the two, and a daily update keeps it current. Our dated snapshot of 245 Shopify storefronts on August 22, 2026 is another fixed point in time.
Limits of this data
- Each study is one scan per site on one date. Scores move when sites change.
- The score measures what AI systems can read on a site, and it does not forecast citations or traffic. A low score does not prove a site gets fewer visits or fewer mentions.
- Findings follow the engine's reporting rules, so the counts describe what the engine reports.
- The storefront study covers one platform, and the non ecommerce study covers homepages only.
- The published storefront percentages appear truncated rather than rounded.
How to cite this report
Cite the original study for each figure, with its population and date. Link this page when you use the comparison. For the storefront figures, cite Visibility Mesh, The State of AI Visibility 2026, which scored 216 Shopify storefronts on June 28, 2026. Both studies and the open AI legibility dataset are published under CC BY 4.0.
Questions people ask
How many websites cannot be read by crawlers because of bot protection or empty JavaScript pages?
In our frozen studies, the engine could not read 13 of 229 Shopify storefronts scanned on June 28, 2026. Bot walls, empty JavaScript renders and error pages stopped it. Of 350 non ecommerce websites we set out to measure in July 2026, 47 produced no score. Most of them had heavy bot protection or a homepage built only in JavaScript, and four never finished scanning. We publish no single share for the wider web yet. The reasons across all our scans are not classified yet.
How common are entity integrity problems on websites?
Entity problems were the most common failure in our storefront study. We scanned 216 Shopify storefronts on June 28, 2026. Of those, 72 percent failed Entity Integrity and earned less than half of its points. Entity Integrity checks whether the brand, its products and their attributes are stated consistently in the data AI systems read. On 299 non ecommerce homepages we reached in July 2026, 42.5 percent carried Organization markup. Neither figure shows that an assistant confused any of these companies.
Are online stores more or less readable to AI than other websites?
Our frozen studies do not settle it. The median Shopify storefront scored 46 out of 100 on June 28, 2026. The median non ecommerce website scored 49.1 in July 2026. The samples differ in platform, date and depth. The second study read homepages only. We do not treat the gap as a finding about stores. The Research and Press page carries the current split between online stores and other websites in our live benchmark.
Next step
The method, the populations and the live benchmark all sit on the Research and Press page. Reporters can use the Visibility Mesh research and press kit there as well. To see where your own site stands against these findings, run the free scan. It reads five of your key pages the way a crawler receives them.
Every figure we publish comes with its method, population and date.
Sources
We read every source below on September 26, 2026.
- Our storefront study is The State of AI Visibility 2026, with 216 storefronts scored on June 28, 2026.
- Our non ecommerce study is The State of AI Visibility on the Non Ecommerce Web, published July 13, 2026.
- Our AI Legibility Framework v1.0 defines the five metrics.
- Google explains rendering in Understand the JavaScript SEO basics, updated March 4, 2026.
- Google describes Organization structured data, updated September 8, 2026.
- Google's documentation updates page records the end of FAQ rich results on May 7, 2026.