Blog · Machine readability

Author and publisher markup that survives scrutiny

Published 8 October 2026 · Updated 12 October 2026 · 8 min read

Author and publisher markup that survives scrutiny
Photograph: R Mo / unsplash

The short answer: Author and publisher markup survives scrutiny when every claim in it is true somewhere else as well. The name in author.name matches the visible byline exactly, author.url resolves to a page about that same author, the publisher is one stable organisation node reused across the site, and nothing in the markup describes a person who does not exist. Markup that only exists inside the JSON-LD is a claim with no backing, and it is the first thing a careful check pulls apart.

Most schema advice treats author and publisher as boxes to fill. They are not boxes. They are two assertions about identity, and both of them can be checked against the visible page and against whatever else the web says about the same entity.

What the markup is actually for

Google's Article documentation, last updated 8 September 2026, is blunt about the status of these fields: there are no required properties at all, and you add what applies to your content. Author and publisher sit in the recommended set, not the mandatory one. So nothing in Google's Article guidance breaks if you omit them. That is a statement about Google. No other engine publishes a comparable property list, so for the rest we cannot say what omitting these fields costs, only that we have no evidence it costs anything.

So why bother. Because the purpose is disambiguation, not eligibility. An engine that reads a page has to decide which of the several thousand people with a common name wrote it, and which company published it, and whether those two facts agree with everything else it already holds. The markup is the shortest route to that answer. Fill it carelessly and you have handed the engine a contradiction instead, which is worse than silence.

That logic is the same one behind the rest of structured data for AI search. Markup is evidence about an entity. Evidence that conflicts with other evidence gets discounted.

The five failure modes we see most

The byline that does not match. The page says "By [name], Head of Research". The markup says "name": "By [name], Head of Research". Google's author markup guidance is explicit that author.name carries the name and nothing else: no introductory word, no job title, no honorific, no publisher name folded in. Separate properties exist for the rest, including jobTitle, honorificPrefix and honorificSuffix. A name field containing the word "By" is a reliable tell that nobody checked the output.

Two authors mashed into one string. One name string holding two people creates one entity who does not exist. Google's author best practices show this explicitly: list each author in its own author field, and do not merge two names into one. This matters more than it sounds, because the merged string will never match a profile anywhere, so the author never resolves.

An author.url that goes nowhere useful. The property is meant to point at a page that uniquely identifies that author: a bio page, a profile, an about page. Pointing it at the site home page, the blog index, or a tag archive listing forty posts tells an engine nothing. Where the target is an internal profile page, the documentation asks you to mark that page up as a ProfilePage, with mainEntity set to the Person or Organization it is about. We rarely find it in the sites we audit, and we have not counted how often, so take that as an impression from our own sample rather than a figure. It is cheap to add.

The publisher described twice, differently. One template writes the publisher as Citrel AI, another as Citrel, a third carries a logo URL that 404s. Three descriptions, three weak claims. One organisation node with a stable @id, referenced from every page, is one strong claim. If the author and publisher genuinely are the same organisation, reference the same node from both properties rather than writing the object out twice with small differences between the copies.

The person who was invented. Google's guidance on creating helpful content, updated 5 October 2026, names this directly: fabricating creator profiles, with generated headshots, made-up names or false credentials, is deception. The same document asks whether a reader can tell who wrote the page, and whether bylines lead anywhere. A fabricated author is not a neutral shortcut. It is a claim you cannot defend when somebody looks.

When the author is the company

Plenty of pages have no individual behind them. A changelog, a policy page, a pricing explanation, a research note written by four people over three weeks. For those, Organization is the honest author type and the markup should say so.

This blog does exactly that. Author and publisher are both Citrel AI, because no single person owns these posts. The distinction that matters is intent. Using the organisation because the organisation wrote the page is accurate. Using the organisation to paper over a person who did write it, so that nobody has to stand behind the claims, is the same avoidance that the helpful-content guidance is pointing at.

An organisation author needs the same resolution work a person does: name, url, and a sameAs array pointing at profiles that unambiguously identify the company. That is the bridge between this markup and sameAs and the identity links engines check, and it is the part that tends to go stale after a rebrand.

The publisher logo, the one hard number

Publisher is where a checkable requirement still lives. Google's logo documentation, also updated 8 September 2026, sets a minimum image size of 112x112 pixels, requires the image URL to be crawlable and indexable, and asks that the logo look right on a white background. The same page also recommends placing the organisation markup on the home page or on a single page that describes the company, such as an about page, and says you do not need it on every page of the site.

Three things to verify on the logo today. It returns a 200 to a plain request. It is at least 112 pixels on both sides. And it is not blocked in robots.txt, which happens when a CDN path is disallowed wholesale.

Four checks you can run this afternoon

  1. Fetch the page as a bot would. curl -sL <url> | grep -A40 'application/ld+json'.

    2. If the JSON-LD is missing from that response, it is being injected by client-side JavaScript. Google documents that it renders pages before extracting structured data. Most other crawlers publish no such commitment, and we have no way to see from outside whether they execute your scripts, so markup that only appears after JavaScript is a claim you cannot confirm has been read. Server rendering removes the question.

  2. Diff the markup against the visible page. Copy author.name out of the JSON and

    3. search for that exact string in the rendered text. Not a close match. The same string.

  3. Open every URL in the author and publisher objects. Each one should return 200 and

    4. should be about the entity that named it. A redirect to the home page is a fail.

  4. Count your publisher descriptions. Pull the JSON-LD from ten pages across templates

    5. and compare the publisher objects. More than one spelling, one URL or one logo means you are making several claims where you meant to make one.

The self-check has ten checks in this shape, covering access and identity as well as markup.

What this does and does not buy you in AI answers

Being honest about the ceiling here matters. We have found no published weighting for structured data in answer assembly from any of the engines we track, which is not the same as knowing there is none, and the public experiments we have read point in different directions depending on the engine and the site. What is documented is narrower: Google describes the Search features that this markup feeds, and the assistants that read a rendered page describe no equivalent use of it. That asymmetry in documentation, rather than a measured difference in behaviour, is one of the practical differences behind Google AI Mode: what changes for a publisher.

Our working position comes from audits rather than from a controlled study, and audits change several things at once. On that basis: we have not seen clean author and publisher markup produce a visible jump on its own, and we treat contradictory markup as a risk worth removing rather than as a cost we have measured. On our own evidence the upside looks modest and the downside is unquantified, which makes this cheap insurance rather than a growth lever. It belongs in the identity layer of answer engine optimisation, next to naming consistency and the links between your profiles.

Honest limits

We cannot isolate the effect of author markup alone. Audits change several things at once, and the engines shift underneath the work, so attributing a citation gain to one property would be guesswork dressed up as measurement. The failure modes above come from the sites we happen to audit, which are sites that already suspected a problem, so they are a sample and a biased one, not a census of the web. We have not kept a tally of how often each one appears, so this post carries no frequency figure, and an earlier claim that they show up on most sites has been removed for that reason. Every documented requirement here is Google's; where other engines are concerned we are reporting the absence of published guidance, which is a gap in what we can see rather than evidence that the markup does not matter to them.

FAQ

Is author markup required for an article to appear in Google? No. The Article documentation states there are no required properties, and author sits in the recommended set. It helps with disambiguation rather than eligibility.

Can an organisation be the author rather than a person? Yes, when the organisation genuinely wrote the page. Use Organization, give it a url and a sameAs array, and reference the same node as publisher if they are the same entity.

What goes in author.name? The name only. No "By", no job title, no honorific, no company name. Job titles and honorifics have their own properties.

Does author markup help in ChatGPT or Perplexity answers? Neither engine publishes a weighting for it, and we cannot see one from outside, so we do not know. Treat it as identity hygiene that prevents confusion, not as a lever we can show moves citations on its own.

Sources

Related reading

Structured data for AI search: what to mark up firstsameAs and the identity links engines checkGoogle AI Mode: what changes for a publisherWhat is answer engine optimisationDuplicates, canonicals and confused AI answersBrand entity resolution: making AI engines sure who you are