Google now calls fabricated authorship deception: what changed, and what to do with your bylines

On October 1, Google added a paragraph to its helpful content guidance that names the trick without hedging: using "AI-generated headshots, made-up names, or false credentials to make content appear as if it was written by human experts" is "a form of deception," and deception makes a page "untrustworthy to both users and our automated quality systems." The same day, on a different page, Google made it "critical" to manually fact-check all AI content before publishing. Same direction, two documents: answer for what you publish.

This is documentation, not an algorithm update. Nobody got a message in Search Console over it. But where the paragraph landed says more than the paragraph itself: inside the Who/How/Why framework, right after Google "strongly encourages" accurate authorship information. The prize and the fine in the same paragraph.

What Google added to the page

The Creating helpful content guide already carried the E-E-A-T questions and the Who/How/Why framework. The version dated October 1 added definitions of effort, originality, talent or skill, and accuracy, and closed the "Who" section with the new warning. The full text, so nobody has to rely on summaries:

"However, avoid using deceptive authorship information. Fabricating creator profiles (such as by using AI-generated headshots, made-up names, or false credentials to make content appear as if it was written by human experts) is a form of deception. Any form of deception makes a page untrustworthy to both users and our automated quality systems, and is a signal of a low-quality page."

Search Engine Journal reported the change and noted what most people missed: the quality rater guidelines already covered this for a year and a half. Section 4.5.3 describes "deliberately inaccurate or misleading information about the website or content creators" as a type of deception, and rules that any page using deception of any type gets rated Lowest. The criterion is not new. The address is: it is now told to you, not only to the rater.

What did not change, to keep this honest

• No penalty or algorithm was announced. Google did not explain how it would detect a fabricated profile or promise any specific punishment. Engage Web made the same point when it covered the news on October 7.
• Bylines still do not rank. The Search Liaison account clarified in January 2024 that bylines do not help a page rank better, as Search Engine Journal reported at the time. Anyone selling "author SEO" on the back of this news is selling noise.
• The asymmetry is the story. The badge does not lift you; the lie around it sinks you. And Google wrote "our automated quality systems" on purpose: it is telling you the lie is machine-readable policy, whether or not any detector has been announced. The record is being written, same as it was on October 1 with AI content verification (we took that apart here).

The nuance the headlines skip: you did not need the fake expert

Google added an example in the same page that almost nobody quoted: not all content requires specialized expertise. A person sharing how they found a better way to clear their driveway of snow does not need a degree. Next to it, the new accuracy definition: informational pages must be factually accurate, and YMYL topics must be consistent with established expert consensus.

Read together, the message is uncomfortable for sellers of shortcuts: most business content did not need an invented expert, it needed effort and honesty. The fabricated persona almost always solved a problem that did not exist and created one that does.

Why this hurts first in AI search (our reading)

This is our analysis, not a Google statement. AI search systems resolve entities by cross-checking sources before citing. The Washington University in St. Louis audit of 7,583 AI Overviews, which we broke down on Saturday, found that 11% of cited claims had no support on the page Google linked: those systems drop what they cannot back up, and they are getting sharper at it.

An author who exists only on your site cannot be resolved anywhere else. No profiles to cross-check, no publishing history, no external corroboration. In an interface that rewards what it can attribute with confidence, an unresolvable author is exactly the kind of claim a model prefers to drop rather than risk. The same calculus applies to a model's own claims: it is why we verified by hand when Wan 3.0 showed up everywhere with false ones (the protocol, here).

How we handle it ourselves

This blog publishes almost daily in two languages, largely AI-assisted, and the byline says "Mintec" because that is the truth of the pipeline: a company runs it and answers for its claims. No invented expert has ever appeared here. That is not virtue, it is strategy: every number in every article has a source we opened before publishing, and when Google updated its AI content guidance we wrote what that review actually looks like in practice (the workflow, here).

The questions Google added to the "How" section describe our process exactly: is the use of AI self-evident to visitors? Is it explained how it was used? Why was it useful? Our articles answer those questions in their frontmatter and their prose. The same rule applies on client work: name the humans who actually review, or put the company's name on it and own it. The third option, inventing an expert, is the only one that just acquired a written price.

A 20-minute authorship audit

• Byline inventory. List every person named across articles, resources, and service pages. Each name must map to a real, publicly verifiable person. A name that does not survive a LinkedIn search is a liability.
• Photos. Any AI-generated headshot presented as a real person comes down today. Real photo or no photo: the second option is free and nobody penalizes it.
• Credentials. Every title and bio claim needs linkable support. "Ex-Google" with no evidence comes out. Google's new accuracy definition applies to bios too.
• Entities. Person schema with sameAs pointing at real, consistent profiles; a live author page, not a placeholder; byline consistent sitewide. This is not "author SEO": it is making the data machine-verifiable, which is the only thing Google actually asks for.
• Disclosures and archive. If AI substantially generated a page, say so in the "How." And sweep the old posts: a fake persona in the archive is the highest-risk page on the site, and deception taints reputation beyond the individual article.

This week

Today: inventory and photos. This week: schema, sameAs, and disclosures on the most AI-heavy pages. After that there is no "author optimization" work left worth doing. What is left is producing content worth signing.