Article Schema Done Right: Authorship, Dates, Trust

Published April 22, 2026 · ClickRadius Institute

Every AI engine composing an answer performs a quiet act of source criticism: who wrote this, when, and on whose authority? Article schema is where your site answers those questions in a form machines can verify. Most sites technically have it — a theme or plugin stamped something out years ago — and most implementations fail at exactly the properties that matter: authorship as a real entity, dates that are honest, and a publisher that resolves to a known organization. This guide covers the full implementation, the policy constraints, and the failure patterns we find most often when scoring sites.

Why article-level metadata is having a second life

In the ten-blue-links era, article markup mostly earned cosmetic benefits — a headline in a Top Stories carousel, a date under a snippet. The economics have changed. Industry trackers estimated that roughly 60% of searches were already ending without a click by 2024, and generative result surfaces have kept pushing in that direction through early 2026. When an engine composes the answer itself and attributes only a handful of sources, source-level credibility assessment stops being a tiebreaker and becomes the selection mechanism.

Provenance metadata is the machine-checkable part of credibility. According to Google's Article structured data documentation, article markup helps Google understand headlines, publication dates and author information — and Google explicitly recommends marking up authors with as much identifying detail as possible. Google's own quality frameworks point the same direction: the E-E-A-T concept (Experience, Expertise, Authoritativeness, Trustworthiness) in its helpful content guidance is fundamentally a provenance question, and structured data is how provenance becomes legible to software.

An unattributed page asks the engine to trust prose. An attributed page gives the engine a chain it can follow: this claim, by this author, with these credentials, published by this organization, on this date. Machines cite chains, not vibes.— ClickRadius Institute analysis

Choosing the type: Article, BlogPosting, NewsArticle

Schema.org arranges these as a small hierarchy — Article is the parent; BlogPosting and NewsArticle specialize it — and Google's article feature accepts all three. The rule is the same one that governs all schema selection: the most specific type that is true.

• NewsArticle — genuine news reporting with a dateline mentality. Using it on evergreen marketing content is misclassification.
• BlogPosting — posts in a blog context, including most company editorial content.
• Article — the safe general default for guides, research pieces and resources like this one.

Because the properties are inherited, switching among them later is a one-word change. Do not agonize here; spend the effort on the properties below.

The properties that do the work

headline and description

The headline should match the visible <h1>. A mismatch between marked-up headline, title tag and on-page heading is a small inconsistency, but provenance checking is a game of consistency, and you are being compared against sources with none of these discrepancies.

datePublished and dateModified

Use full ISO 8601 with timezone where possible. Two policy points from Google's documentation deserve emphasis: the dates should match what the page shows, and dateModified should change only when the content meaningfully changes. Retrieval systems demonstrably favor fresher sources for time-sensitive topics, which creates a temptation to bump dates cosmetically — resist it. Engines and crawlers retain prior copies of your pages; a dateModified that changes weekly while the content diff is empty is a detectable dishonesty pattern. We cover the strategy side of this in Content Freshness and Decay for AI.

author — the property most sites get wrong

The single most common defect we find: "author": "Admin" or a bare name string. Google's documentation explicitly recommends the author be a Person or Organization node, and recommends including an author url pointing to a page that identifies them. The strong pattern is:

"author": {
  "@type": "Person",
  "name": "Jane Rivera",
  "url": "https://example.com/team/jane-rivera",
  "jobTitle": "Licensed Structural Engineer",
  "sameAs": [
    "https://www.linkedin.com/in/janerivera",
    "https://www.example-registry.org/license/12345"
  ]
}

The url should resolve to a real author page — bio, credentials, list of their articles — and the sameAs array should triangulate the person across independent platforms. This turns "a name" into "an entity," which is the difference between decoration and verifiable provenance. If content is genuinely produced editorially by the company, an Organization author is honest and fully supported — this article carries one. What you must never do is invent a persona with a stock photo: fabricated authorship is the exact trust failure these systems are being built to detect, and it converts your entire domain's metadata into suspect testimony.

publisher

A publisher node with name and logo, ideally referencing your site's canonical Organization entity by @id rather than redeclaring it. One canonical publisher node, referenced from every article, is both cleaner and more consistent than hundreds of inline copies that drift apart over time — the @id pattern from Schema Markup for AI Citation.

image and mainEntityOfPage

Google recommends high-resolution images (it has asked for images of at least 50,000 pixels total, in multiple aspect ratios, in its article guidance) — but any honest representative image beats none. mainEntityOfPage should carry the canonical URL, closing the loop with your rel=canonical.

The five failure patterns that quietly cost citations

1. String authors. "author": "Admin" provides negative information — it tells the machine that nobody stands behind the content.
2. Frozen dates. A 2021 datePublished with no dateModified on a page you have actually updated forfeits freshness credit you have legitimately earned.
3. Template drift. A redesign changes the H1 structure, and the markup now describes a page that no longer exists. Schema validates syntactically while being semantically stale — the validator will not save you; only regeneration from CMS fields will.
4. Contradictory duplicates. An SEO plugin, a theme and a hand-added block each emit their own Article node with different dates. Machines confronted with three contradictory provenance claims reasonably trust none. One block per page.
5. Markup on pages that aren't articles. Category pages, tag archives and product pages stamped with Article markup by an over-eager plugin dilute the signal on pages where it is real.

Article schema and the content itself: two layers, one goal

Provenance markup makes your content attributable; the content still has to be worth attributing. The Princeton-led GEO study (Aggarwal et al., KDD 2024) found that adding quotations, statistics and source citations to content raised its visibility in generative engine responses by up to roughly 40% in benchmark testing — the three signals ClickRadius's content scoring weights directly. The compound pattern, then, is an article that (a) carries dense, sourced, quotable claims and (b) wraps them in verifiable authorship and honest dates. Each layer amplifies the other: statistics from an identifiable licensed expert published last month outrank the same words from "Admin, date unknown" in any sane source-selection model.

Think of Article schema as the byline, and the byline as an argument: here is why this source, of all sources, should be the one quoted. Most of the web still declines to make that argument.— ClickRadius Institute analysis

That last point is the opportunity. Industry estimates suggest a large majority of brands today have zero presence in AI-generated answers. In most niches, simply being the site whose articles carry complete, honest, verifiable provenance puts you in a small minority — a advantage that costs a template change.

Building author entities that compound

The author property rewards a small amount of infrastructure most sites never build: real author pages. If your articles name experts, each expert needs a stable URL — /team/jane-rivera — that functions as the canonical home of that person's professional identity on your domain. What belongs on it:

• A substantive bio stating credentials, years of practice, licenses and specializations in plain prose — the facts an engine would need to justify treating this person as an authority on the topics they write about.
• Person markup on the page itself, with jobTitle, worksFor referencing the organization's @id, knowsAbout for expertise areas, and sameAs links to LinkedIn, professional registries and anywhere else the person verifiably exists.
• A list of their articles on your site, linked both ways — article credits author, author page lists articles. This bidirectional linking is what lets a machine assemble "this person has written twelve sourced pieces on this topic over three years," which is an authority argument no single article can make alone.

Once the page exists, every article's author.url points at it, and each new piece the expert publishes strengthens all the previous ones. This is the compounding that string authors forfeit: fifty articles by "Admin" are fifty orphans, while fifty articles by a resolvable entity are a body of work.

Two honest boundary cases. First, ghostwritten content reviewed by an expert: the defensible pattern is crediting the reviewer truthfully — schema.org and Google both support a reviewedBy arrangement, and "medically reviewed by Dr. X" style attribution is an established, honest convention — rather than pretending the expert typed every word. Second, AI-assisted content: Google's public guidance since 2023 has been that it rewards helpful content "however it is produced," while its spam policies target content generated primarily to manipulate rankings. The provenance layer does not require concealing your production process; it requires that a named, real party stands behind the accuracy of what is published.

Implementation checklist

1. One Article/BlogPosting JSON-LD block per article page, generated from CMS fields — never hand-edited per post.
2. headline mirrors the H1; description mirrors the meta description.
3. datePublished fixed at first publication; dateModified wired to real content revisions only.
4. author as Person (with URL, title, sameAs) or honest Organization; author pages exist and are indexable.
5. publisher references the canonical Organization @id.
6. mainEntityOfPage equals the canonical URL; representative image present.
7. Validate with the Schema.org validator and Google's Rich Results Test; re-run after every template change.
8. Audit quarterly for duplicate or contradictory blocks from plugins.

Frequently asked questions

Should the author be a Person or an Organization?

Whichever is true. Named human experts as Person nodes with credentials and profile links are the strongest pattern for expertise-sensitive topics. If content is genuinely produced editorially by the company, an Organization author is honest and acceptable — Google's documentation explicitly supports both. What hurts you is a fake persona, or a bare name string with no entity data behind it.

Is it OK to update dateModified whenever I touch a page?

Only when the content meaningfully changes. Google's guidance says not to bump dates for trivial edits, and mass-updating dates without changing substance is a detectable pattern — engines can compare stored copies of your page across crawls. Honest dateModified is a freshness signal; dishonest dateModified is a credibility risk.

Article, BlogPosting or NewsArticle — which type should I use?

Use NewsArticle only for genuine news reporting, BlogPosting for blog-style posts, and Article as the general default. They inherit from each other, so the properties are nearly identical; pick the most specific type that truthfully describes the content.

ClickRadius audits every article on your site for provenance completeness — authorship entities, date honesty, duplicate markup — as part of its six-category AI-citation score, and can fix what it finds. Get your free AI Readiness Score, or see plans on the pricing page.