The State of AI Visibility on the Non-Ecommerce Web

We assessed 350 non ecommerce, non government websites across 25 verticals for one thing: how clearly an AI answer engine can read them. 303 returned a score. The picture is middling and machine opaque.

01 What we found

The non ecommerce web is middling and machine opaque. Across 303 sites the mean AI visibility score is 48.1 out of 100 (median 49.1, SD 11.0), and 53.1% score below 50. Not one site cleared 72. These are recognizable institutions, universities, museums, national newsrooms, research labs, yet most present themselves to an answer engine as a thin, under structured page.

The clearest signal: structured data is the bottleneck. The Schema pillar averages just 35.3% of its available points, well below every other pillar. Only 52.2% of reachable homepages carry any JSON-LD, and only 1.3% expose FAQ schema, the format answer engines quote most directly. Sites that do add schema score 12.7 points higher.

02 Every vertical, ranked

Sectors that live by search, marketing and SEO publishers, news, health, lead. The laggards are institutions whose authority is offline: sports bodies, reference sites, museums, and universities, which tend to bury content behind JavaScript and skip schema.

03 Where the points are lost: five pillars

Four pillars cluster near half marks. Schema collapses to 35.3%. The web gets readable text and technical hygiene passable, then fails to translate any of it into the machine readable structure an answer engine needs to quote it.

04 The structural facts (deterministic)

Separately from the score, we fetched each homepage and its robots.txt directly and parsed them, no model involved, so every figure below is exactly reproducible. The basics are common; the AI specific structure is rare.

05 AI crawler posture

Almost every site serves a robots.txt (97.7%), but only 28.4% name a single AI crawler, and most of those entries are blocks (25.8% disallow at least one AI bot). 22.4% publish an llms.txt. The web has not yet formed an intentional posture toward AI crawlers; most access happens by default, neither invited nor refused.

06 The most common gaps

Aggregated across 11,633 specific findings, the same gaps recur everywhere. Thin structured data and the absence of an FAQ block together account for more than half of everything flagged.

07 What moves the score

Grouping sites by a structural fact isolates what correlates with AI visibility. Structured data dominates: any JSON-LD is worth 12.7 points, Organization markup 12.3. Even naming AI bots in robots.txt tracks a 6.9 point lift.

08 What to do about it

The findings double as a checklist. If you want AI engines to read and cite your site, these are the moves that separate the top of this sample from the bottom, in rough order of impact.

1. Add structured data. Mark up your Organization and WebSite site wide, and every content page with the right type: Article, FAQPage, BreadcrumbList, Product where it applies. This is the single biggest lever we measured, worth 12.7 points on average.
2. Expose an FAQ block with FAQPage schema. Real questions, real answers, visible on the page and mirrored in schema. It is the format answer engines quote back most often, and only 1.3% of sites do it. This very page is an example.
3. Make core content readable without JavaScript. Server render the key text. If a crawler sees an empty shell or a paywall gate, so does the AI.
4. State your entity clearly. Organization schema with a logo and sameAs links to your real profiles, plus a named author with visible credentials. Ambiguous identity is un citeable identity.
5. Show freshness. A visible published and updated date, and dateModified in schema. AI systems discount content they cannot date.
6. Set an intentional AI crawler posture. Decide in robots.txt whether you welcome or refuse the major AI crawlers, and consider an llms.txt. Do not leave it to default.
7. Keep the structure clean. One H1, a logical heading order, a BreadcrumbList, and internal links between related pages so a machine can map your site.

09 Leaders and laggards

10 Frequently asked questions

11 Method and integrity

• Sample. 350 non ecommerce, non government sites, across 25 verticals. 303 returned a score; 47 could not be scored (heavily bot protected or JavaScript only homepages). Unscannable sites are reported as coverage gaps, never as a zero.
• Scoring. Each homepage is scored against a fixed, structured five pillar rubric with an explicit, versioned set of sub criteria, calibrated to a reproducibility target so the same site re assessed returns within a tight band. Applied identically to every site.
• Deterministic verification. The structural prevalence and AI crawler figures involve no scoring model at all: each is a direct fetch and parse of the live homepage and its robots.txt, exactly reproducible by anyone.
• Integrity. Every finding must be grounded in content actually read from the live page. Across 11,633 findings, the audited false claim rate is 0%. Anything a crawler could not read is labelled not verifiable, never asserted as absent.
• Point in time. Findings reflect each site only at its crawl timestamp (2026-07-13). Re assess for current state.