One article, one signatory: what an expert signature changes for Google and for AI

On most company blogs, articles are signed "the team" or not signed at all. On media properties that count, every article carries a person's name, their role and a link to a page saying who they are. The difference shows in the rankings. It changes how Google assesses a source, how a generative engine picks who to cite, and, since August 2026, how a publisher proves that content produced with an AI was reviewed under editorial responsibility.

What Google asks for, word for word

Google's search quality rater guidelines, a public document of over 170 pages, tell raters to find who is responsible for a page and who created its main content. They define what Google calls experience, expertise, authoritativeness and trustworthiness, and state that trust is the central criterion: the other three exist to establish it. For content touching health, money or safety, a page whose author is unknown is explicitly rated as less trustworthy.

These guidelines describe not an algorithm but what Google is trying to measure, and the public guide on creating helpful content puts the question bluntly: "is the content presented in a way that makes readers trust it, with clear sourcing and information about the author?"

Declaring the author so a machine can read it

A visible byline is not enough: a robot has to read it without ambiguity. Google's Article structured data documentation describes the expected properties; three of them bear directly on the byline:

• author, with a name and preferably a URL pointing to the author's page;
• datePublished and dateModified, which date the responsibility;
• a Person object in the schema.org sense, which can carry the role, the organization and links to other profiles of the same person.

Google recommends a stable author URL, the same one across every article by that person. That is what ties all of one person's articles to a single profile, and so builds authority for a person rather than only for a site.

What a generative engine does with the byline

A generative engine does not rank pages, it picks sources to cite. To pick, it has to attach the passage to a source: which site does this come from, written by whom, with what standing on the subject? An article signed by someone whose author page shows ten years of practice in the field is a source the system can name. An article signed "the team" is an anonymous source. We set out that mechanism in the four signals AI tools look at: the identifiable source is the fourth, and it cannot be separated from consistency across one signatory's articles.

The byline has another, less visible effect: a named signatory holds a consistent line. When the same author signs two articles, a mismatch in figures between them comes back to them, and they fix it, because somebody carries the responsibility. On a media property publishing dozens of pieces a month, the byline is what stops the drift.

The byline as proof of review

Since August 2, 2026, article 50 of the EU AI Act requires AI-generated text to be disclosed when it is published to inform the public on matters of public interest. The same article carves out an exception, and it has two routes: no disclosure is due where the content has undergone human review or editorial control, and a natural or legal person holds editorial responsibility for its publication. Reviewing every article is therefore not the only way to meet it; it is, however, the easiest to prove, and the byline is the most readable proof of all. We set out the scope in our AI Act timeline for a publisher.

The byline is the visible piece of that arrangement. An article reviewed and signed by an identified person, inside a traced approval circuit, is content whose editorial responsibility is owned. A "reviewed" box checked by anyone, with no name, proves nothing. The same author page then serves two purposes: it reassures the reader and it documents compliance.

How to organize signatories on a media property

The three roles are not always needed. A general how-to guide can have an author and nothing else. Health, finance or legal content gains from showing an expert, because that is what quality raters look for first on those subjects, and because editorial responsibility carries more weight there.

What the author page becomes

On a structured media property, the author page stops being a contact card and becomes a hub: it carries the role, the background, the fields covered, the list of signed articles, and links to verifiable outside profiles. That page is what Google ties to the author URL declared in the structured data, and it is what a generative engine uses to decide whether "Jane Smith, structural engineer" is a legitimate source on roofing. A thin author page loses most of the benefit of signing; a rich one accumulates the credit of every article published.

The mistakes that cancel the effect

• The fictional signatory. An invented name with a generated photograph is worse than no byline: it is a false statement of provenance, and it is precisely the kind of practice Google's quality guidelines ask raters to spot.
• The omniscient signatory. Someone who signs a tax article on Monday, a plumbing article on Tuesday and a nutrition article on Wednesday has expertise in nothing. A signatory holds one field.
• The empty author page. A name pointing to a page with no role, no background and no article list allows no attribution. The author page is content in its own right.
• The byline with no review. Signing a text you have not read exposes both the person and the publisher. A byline comes at the end of a circuit, it is not applied by default.

On a media property publishing in volume, these rules only hold if the tool enforces them: an article should not be publishable without a signatory, the author page should exist before the first article, and the structured data should be generated automatically from the profile. That is what separates an editorial system from a plain generation tool, and what makes a byline credible to a reader and to an engine alike.