A business website can be readable to a person and still leave important facts implicit for software expected to retrieve, summarize, or represent it.
Four layers of evidence
We regrouped the already-published findings from the frozen 2026 dataset into four practical layers. Each percentage below counts unique websites, so a site with several findings in the same layer is counted once.
The layers overlap. A homepage can appear in more than one. “Gap” means BoostGeo produced one of the specified findings from the captured homepage; it does not mean the business is absent from AI answers.
1. Identity was the broadest recurring gap
586 of 897 assessable homepages (65.3%) had at least one of three identity-layer findings: the checked Organization signal was absent, the checked WebSite signal was absent, or no JSON-LD was detected.
The most common individual finding was the Organization schema opportunity, present on 565 homepages. That does not mean all 565 lacked a visible business name. It means the specific machine-readable signal checked by the study was not detected.
2. Representation was often incomplete
472 homepages (52.6%) had at least one issue involving page title length, canonical URL, or Open Graph preview data. These fields help systems and sharing surfaces identify which URL is authoritative and how the page should be described when represented elsewhere.
3. Many pages did not explain themselves cleanly
390 homepages (43.5%) had at least one page-explanation finding: no meta description, a description outside the checked range, no main heading, or very thin visible copy.
A main heading or meta description is not a magic ranking lever. Together with the visible page, however, they provide compact evidence about what the page is and what the business offers.
4. Crawler blocking was real, but narrower in this sample
At least one major AI crawler was blocked on 31 assessable homepages (3.5%). That is worth fixing when intentional access is desired, but it was not the dominant class of evidence gap in this captured-homepage study.
The practical order is simple: state accurate business facts clearly, align how the page describes itself, add machine-readable identity, and then confirm that the systems you want can access it.
Most websites had more than one measured gap
Across the twelve findings included in this derivative analysis, 769 assessable homepages had at least one, 387 had three or more, and 156 had five or more. These counts describe detected findings, not the amount of work required. One accurate update can sometimes resolve several related signals.
Confidence check
The pattern remained stable when the 131 limited-confidence assessments were excluded. Among 766 high-confidence homepages, the four layer rates were 65.4%, 52.2%, 43.1%, and 3.5%, respectively. That sensitivity check reduces the chance that the headline pattern is an artifact of limited evidence.
What this analysis can—and cannot—say
This is a secondary analysis of the same frozen, quota-based dataset used in the BoostGeo 2026 AI Website Readiness Study. It measures captured homepage evidence. It does not measure public AI mentions, rankings, leads, revenue, or business quality, and it does not prove that adding any single technical signal will cause a recommendation.
Review the methodology and limitations, then use the plain-English improvement guide to turn the evidence into a sensible order of work.