Verifiable Authorship: Why Provenance Became an Authority Signal

In 2009 a small piece of markup called hRights briefly became a public argument. It was part of hNews, a format for news articles, and it promised to say in machine-readable form who could reuse a story and on what terms. Critics read it as a digital lock. Its defenders read it as a label. The label reading turned out to be correct, and the idea underneath it never went away: a page should be able to state, in a form machines can read, who wrote it, who stands behind it and what rules it was produced under. That idea now sits at the centre of how authority is assessed, by search engines and by the AI systems that decide which sources to quote.

What hNews and hRights were

hNews was a microformat drafted from 2008 by the Associated Press and the Media Standards Trust, a British charity working on transparency in news. A microformat is a set of agreed class names added to ordinary HTML, so that the same page a reader sees also tells software what each part of it is. hNews built on hAtom, the existing microformat for blog entries, and added the fields that a news story needs and a blog post usually does not.

The draft specification made one of those fields compulsory: source-org, the organisation the story came from, encoded as an hCard. A story without it was not valid hNews. It recommended a dateline saying where the story was filed, an item-license link stating the terms of reuse, and a principles link pointing to the publisher’s statement of editorial standards and ethics. The author came from hAtom underneath. Put together, a story carried its own provenance: who wrote it, which organisation published it, where it was reported, what standards it was held to and how it could be used.

hRights was the rights part of that picture. In July 2009 the AP announced a news registry that would follow its content across the web, and described hNews and hRights as the framework that would make it possible. The announcement was worded in a way that suggested enforcement, and the reaction was immediate.

Expression is not enforcement

The clearest answer came from Creative Commons, in a post published on 1 August 2009 that set out the difference between expressing rights and enforcing them. Its point was technical and simple. Markup cannot track, monitor or enforce anything. In the post’s own image, microformats are labels, Post-it notes attached to a document, not locked boxes blocking access to it. Creative Commons had built its own rights language, ccREL, on exactly that understanding: it expresses and communicates the terms of a licence and leaves enforcement to the law. The post also said something that is easy to miss in the controversy. Tagging content with rights, origin and means of attribution, it wrote, was a good proposal.

That distinction is the one that matters for authority. Nothing on a page can force a search engine or a language model to believe it. What a page can do is declare, clearly and consistently, facts that the outside world can then check. A declaration that holds up when checked is a signal. A declaration that does not is noise, and over time it becomes a liability.

What happened next

hNews did not fail. It spread further than most microformats ever did. With support from the Knight Foundation and the MacArthur Foundation, and in partnership with Tim Berners-Lee’s Web Science Trust, the Media Standards Trust reported more than 577 US news sites using it. Then the ground moved under every microformat at once.

In 2011 the major search engines launched schema.org, a shared vocabulary for structured data, and on 27 September of that year schema.org announced extended support for news, aligned with rNews, a standard from the International Press Telecommunications Council. The people who did that work came from the New York Times, Getty Images and the Associated Press. The questions hNews had asked were carried into the vocabulary that search engines actually committed to reading. The syntax changed. The questions stayed.

How provenance is declared today

Each field of the 2009 draft has a direct descendant in the markup a site uses now.

• The author. The author property of an Article, typed as a Person, with a name and a link to a profile page that describes who that person is and what they have done.
• The source organisation. The publisher property, typed as an Organization, with its name, logo and address. This is the old source-org, and it is the field most sites still get wrong, by leaving it empty or by declaring a different name on every page.
• The principles. publishingPrinciples, a schema.org property that points to a document describing the editorial principles of the publisher. It is the direct successor of the principles link in hNews. Around it, the trust indicators mapped by the IPTC with The Trust Project add a masthead, a corrections policy, an ethics policy and ownership and funding information.
• The licence. license on the work, the descendant of item-license. Still a label, still not a lock.
• The identity links. sameAs on the Person and the Organization, listing the other places where the same author and the same publisher can be found and verified.

Google’s own guidance on article markup, expanded in 2022, asks for the same discipline in plain terms: include every author, put only the author’s name in the name field, choose the right type, Person or Organization, and use the additional fields available, such as the profile URL. None of this is new in spirit. It is hNews with a different syntax and a much larger audience.

Why a declaration is only half of the signal

Markup is the declaration. The signal comes from what the declaration connects to. An author entity that points to a profile page, and from there to a consistent presence on LinkedIn, on Reddit, in published books and on the forums of the field, can be checked. An author entity that points to nothing is a name in a field.

This is what AMS calls the Entity Stack: the complete set of platforms where an identity as a credible source is established and verified. The structured data on the site is the index of that stack. The sameAs links are its table of contents. And the confirmation that those external profiles are real, active and recognised by others is Cognitive Validation, authority confirmed through mentions in places that search engines and AI systems already trust.

The 2009 argument maps onto this exactly. hRights could not enforce anything, and author markup cannot either. Declaring an author does not make that author credible. It makes the author identifiable, which is the precondition for credibility being assessed at all. A model choosing between two pages that say the same thing cannot weigh the reputation of a source it cannot identify. Provenance is what turns content into Contextual Authority instead of anonymous text.

Where this sits in execution

In the AMS sequence, provenance belongs to Build, the last of the three modes of Authority Execution, next to schema and the Entity Stack. It comes after Repair and Design for a practical reason. Declaring a publisher and an author is cheap. Declaring them consistently across a site that has already been repaired and structured is what makes the declaration worth reading. A site that announces a clear editorial identity on top of orphaned pages, conflicting bylines and three versions of its own name is declaring confusion in machine-readable form.

The order of work that follows from this is short. One name for the publisher, used identically everywhere. One profile page per author, saying who they are and linking out to where they can be verified. An editorial policy that exists, says something specific and is linked from the markup. A licence stated where it applies. Then, and only then, the external presence that gives all of those declarations something to point at.

What the 2009 debate got right

Everyone involved in the hRights argument was right about something. The AP was right that a story should carry its origin with it. Creative Commons was right that markup expresses and never enforces. Both were describing the same mechanism from different sides: a label is only as strong as the reality it describes and the ability of others to check it.

That is still how verifiable authorship works. The systems reading the label have changed, from news aggregators to search engines to language models, and they are better at checking than anything that existed in 2009. They read the declaration, follow the links and compare what they find. A site that says clearly who wrote it, who publishes it and what standards it keeps, and can be shown to mean it, gives them something to cite. A site that does not leaves them guessing, and a system that is guessing will quote someone else.

Frequently Asked Questions

Sources

• Creative Commons, Rights Expression vs. Rights Enforcement: Clarifying the Associated Press Story, 1 August 2009.
• Microformats wiki, hNews 0.1 draft specification (Associated Press and Media Standards Trust).
• Knight Foundation, Tool That Increases Transparency in News Content Now Being Used on 577 Sites.
• schema.org blog, Extended schema.org news support, 27 September 2011.
• schema.org, publishingPrinciples.
• IPTC, Trust Indicators guidelines.