The one check almost every site fails, and why it matters
Pillar percentages hide which single check is doing the damage. Re-running the engine check by check across the same cohort surfaces it.
Published · updated · by Julian Joseph, founder
Get an Agent Trust Gap BriefWhich single check does almost every site in the cohort fail?
sd.article — "Article / TechArticle / Dataset for content pages." Re-running the ARB engine against the same 30-site cohort defined in cohort.json (2026-08-24), it failed on 30 of 30 sites (100%) and passed on none.
What exactly does this check look for?
A JSON-LD block on the page typed Article, TechArticle, BlogPosting, Dataset or Report. Per the published rubric, its weight is 2 of 18 points in the Structured Data pillar — small alone, but never earned across the cohort.
Why does almost every homepage fail this by default?
The cohort was measured at each site's root URL — an Organization or WebSite page, not a blog post. The check targets content pages; a homepage is not one unless it deliberately adds an Article-typed node of its own.
Which checks are the next most commonly failed?
FAQPage/QAPage markup and BreadcrumbList each fail on 96.7% of the cohort (29 of 30), tied with a stated AI/agent access policy. A named author or reviewer fails on 93.3% (28 of 30) — all recomputed the same way.
| Check | Title | Weight | Fail rate | Pass/Partial/Fail |
|---|---|---|---|---|
sd.article | Article / TechArticle / Dataset for content pages | 2 | 100.0% | 0/0/30 |
sd.faq | FAQPage or QAPage markup | 3 | 96.7% | 1/0/29 |
sd.breadcrumb | BreadcrumbList markup | 2 | 96.7% | 1/0/29 |
gv.ai_policy | Stated policy for AI/agent access | 3 | 96.7% | 1/0/29 |
ev.author | Named author or reviewer | 3 | 93.3% | 2/0/28 |
ag.api | Machine endpoint or API discoverable | 3 | 80.0% | 6/0/24 |
ev.dated | Content carries a machine-readable date | 4 | 73.3% | 8/0/22 |
ag.contact_ld | Contact details in structured data | 3 | 66.7% | 10/0/20 |
ex.question_heads | Question-form headings | 4 | 56.7% | 3/10/17 |
sd.sameas | sameAs identity links in JSON-LD | 3 | 46.7% | 16/0/14 |
Source: re-ran the ARB engine (methodology at methodology.json) against the 30-site cohort in cohort.json/cohort_run.py. Recomputed 2026-08-24T23:09:21Z UTC.
Why does missing Article schema matter for AI answer engines specifically?
Article/TechArticle typing is one of the strongest signals an answer engine uses to decide a page is dated, authored, citable content rather than a navigation shell. Without it, an otherwise strong page is structurally invisible as a citable claim.
How was this measured, exactly?
By importing the unmodified ARB/1.0 engine and calling engine.audit() against each of the same 30 URLs cohort_run.py defines, then counting each check's verdict across all 30 runs — not inferred from the pillar percentages already published in cohort.json.
What the rubric itself says to do about it
The published fix text for sd.article is direct: “Wrap research and article pages in Article/TechArticle with datePublished and author.” — taken verbatim from methodology.json, not paraphrased. It names two things a content page needs that a homepage usually lacks: a datePublished, and a named author.
- Identify which pages are actually content. A pricing page or a homepage is not an article; a guide, a report, or a dated explainer is. Apply the fix only where it is true.
- Add the datePublished and dateModified. Both dates need to be real — the day the page went live, and the day it last substantively changed, not today's date copied onto every page indiscriminately.
- Name a real author, with a bio URL. A generic "ApexClaw" organization author still satisfies a schema validator; it does not satisfy a reader, or an answer engine, asking who stands behind the claim.
- Keep the Article-family type alongside whatever type is already there. Adding a second JSON-LD node does not require removing an existing WebPage, AboutPage or ContactPage node.
- Don't stop at the check that fails. A page can add Article/TechArticle and still miss the datePublished or the author the check is really asking about; re-check that both properties are present, not only the top-level type.
- Expect this to be the cheapest fix on the list. Unlike a booking endpoint or a contact API, adding one JSON-LD node to an existing page requires no new infrastructure at all.
Method note
cohort.json and methodology.json publish pillar-level percentages, not per-check pass/fail counts. The table above required re-running the same, unmodified engine against the same named 30-site cohort to get check-level detail — the cohort membership and the rubric are both drawn from the two authorized files; only the granularity is new, and it is dated 2026-08-24.
Standards behind Structured Data checks: schema.org Article/TechArticle types are cross-referenced in this rubric's Governance pillar against OWASP Agentic Top 10, NIST AI RMF, ISO/IEC 42001 and the EU AI Act.
By Julian Joseph, Founder, ApexClaw. Every figure on this page is recomputed at build time from a named source file and date — see the method notes above. Reviewed against the claims policy: sourced, first-party, or labelled.