A client forwarded me the Search Engine Journal write-up on Friday with three words attached: do we care. The write-up was about fake author photos. The answer is yes, but not for the reason the headline gave.
Here is what happened. On October 1, Google updated its page on creating helpful, reliable, people-first content. Three things went in. A definition of main content. Four named attributes that Search Quality raters assess. And a paragraph saying that inventing a creator profile is deception.
The coverage picked the third one. You can see why. AI-generated headshots and made-up credentials are vivid, a little sordid, and easy to write 600 words about. Matt G. Southern covered it at Search Engine Journal on October 2, and most of what followed covered his coverage.
I want to argue the opposite priority. The fake profile clause is the least consequential thing Google added. The taxonomy change underneath it is the one that reclassifies work your team is already doing.
What Google changed in its people-first content guidance
Google updated its people-first content page on October 1, 2026, adding a definition of main content, a list of four quality attributes its raters assess, and a paragraph that classifies fabricated creator profiles as deception. None of the three is a new policy.
That last point matters, so hold onto it. Every one of these ideas already lived in the Search Quality Rater Guidelines, the long internal document Google gives the people who score results by hand. Raters have been scoring effort and originality for years. They have been told to treat deceptive author profiles as a low-quality signal for years.
What changed is the audience. Google moved the model from the document it trains raters with into the document it points site owners at. A version of the page saved on June 8 carries none of it. You can read the current text in Google's own guidance on creating helpful, reliable, people-first content.
Publishing an internal standard is not a small act. It is Google telling you which yardstick is about to be used out loud, which it tends to do shortly before the yardstick gets used harder.
The fake headshot clause is the least enforceable part
Fabricated creator profiles are now named in Google's public documentation as a form of deception, but spotting an invented byline and a synthetic portrait across the open web is the weakest detection problem on Google's list.
“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.”
Read that and the instruction is unambiguous. Do not do it. I am not arguing with the rule. I am arguing with the panic.
Think about what detection requires here. A name has to be shown not to belong to a person. A headshot has to be shown to be generated, against image models that stopped producing the obvious tells some time ago. A credential has to be shown to be false, which usually means checking a register that is not machine-readable. Each of those is a judgement call on a single page, and Google is ranking billions of them.
Which is why this clause will mostly be enforced the way the industry's other reputational rules get enforced. Somebody notices. The notice spreads. A manual reviewer looks. We walked through how that sequence actually plays out in the manual action that followed a scaled AI content program, and the trigger there was human attention, not a classifier.
So yes, go and check your author pages. It will take an afternoon and the downside risk is reputational as much as algorithmic. The press has already shown what that looks like: an invented art therapist was quoted more than 30 times in real outlets before anyone checked whether she existed. Nobody wants to be the agency that supplied her.
But if replacing headshots is the whole response to the October 1 update, the response missed the update.
People-first content now has four named attributes
The four attributes Google now names for assessing main content are effort, originality, talent or skill, and accuracy, and each one points at a different failure mode in a content operation rather than at a different kind of writing.
These are not vibes. They are the four questions a rater is asked to answer about the substance of a page, and they are now the four questions you should expect a stakeholder to ask you about yours.
| ATTRIBUTE | WHAT GOOGLE DESCRIBES IT AS MEASURING | WHAT FAILS IT | WHERE IT BREAKS IN PRACTICE |
|---|---|---|---|
| Effort | The extent of human work invested, from original analysis through to custom interactive tools, as against content that was generated automatically | Large volumes of generated text shipped without manual oversight, which the guidance says represents little to no effort | Programs where the brief, the draft and the publish are all automated and the only human step is approval |
| Originality | Whether the page offers information or a perspective that is not already available elsewhere online | Summarising the existing top ten results more neatly than they summarise themselves | Briefs built from competitor outlines, which guarantees the output is a composite of what already ranks |
| Talent or skill | Whether the content shows the expertise a visitor needs to be satisfied, including clear writing, production quality and tools that work | Accurate content that is badly written, badly built, or broken on the device most people read it on | Pages that pass an editorial review and fail a usability one, because no single owner holds both |
| Accuracy | Factual correctness, with the bar raised for topics touching health, finance or safety | Numbers with no source, sources that do not say what the page claims, figures that were true two years ago | Fact-checking treated as a copy edit rather than a separate pass with its own sign-off |
Look at the effort row again, because that is the sentence with teeth. The guidance states that using generative AI to produce large amounts of text without manual oversight represents little to no effort. Not low quality. Not risky. Little to no effort, which is the bottom of the scale on the first of the four attributes.
Alongside that, Google's guidance on AI-generated content now tells publishers it is critical to fact-check that output by hand before it goes live. Put those two together and you have a standard that is indifferent to how the draft was produced and extremely interested in what a human did to it afterwards.
That is consistent with everything Google has said since 2023, which is the part the hand-wringing keeps getting wrong. The position has never been that machine-written text is forbidden. We argued that case at length in why Google never actually punished AI content and nothing on the October 1 page contradicts it. Generation was never the offence. Shipping generated text that nobody worked on is the offence, and now the offence has a name and a scale position.
Lily Ray published a prediction on October 4 that reads the same change the same way, and points out the context around it: Google shipped four spam updates in 2026 against one in all of 2025. Her argument is that scaled AI content is the dominant cause of the demotions she has seen this year, and that the October 1 edits are Google saying so in advance.
Those last two figures are hers rather than ours, and they are worth holding at arm's length as a single analyst's read. The direction they point in, though, matches what we see in client reporting: the cheapest content format of the last three years is being cited less, and when it is cited, the publisher is not the beneficiary.
Why the main content definition is the real reclassification
Google's new definition of main content covers any part of a webpage that directly helps the page achieve its purpose, and it explicitly names page titles and headings, interactive features, tabbed sections and user-generated contributions as part of that.
Sit with the title and heading clause for a second, because it is doing more work than it looks like it is doing.
In most content operations, titles and headings are not content. They are a separate workflow owned by a separate person, usually generated in bulk from a keyword sheet, usually the last thing anyone writes and the first thing anyone automates. Thousands of pages are sitting right now with a hand-written body and a machine-assembled title tag and H1 set that nobody read.
Under the published definition, those titles are main content. Which means they are assessed on effort, originality, talent or skill, and accuracy, like everything else. A page with a careful body and a templated heading stack is not a well-made page with cosmetic metadata. It is a page whose main content is partly unexamined.
The same logic reaches further than metadata. Interactive features count, so a calculator that returns the wrong number is now an accuracy failure in main content rather than a broken widget in the sidebar. User contributions count, so an unmoderated comment thread full of spam is part of what gets judged. Tabbed sections count, so the content you hid behind a tab to keep the page tidy is in scope, and hiding it does not remove it from assessment.
This is also where the fake profile clause and the main content definition stop being separate stories. A byline is a trust signal attached to a page. A title tag is a claim about what the page contains. Both are statements about the content that are made outside the content, and Google has just confirmed it reads both as part of the thing being judged. Our running argument about the trust gap that opens when content signals and content quality diverge is the same argument in a different register.
What the results for AI content questions actually look like
The results for the question practitioners actually type, whether Google penalises AI content, are held by pages with almost no page-level link equity, with community threads sitting inside the top ten and an AI Overview above all of it.
We pulled the result set on October 5, 2026. The demand across the cluster is modest and steady rather than enormous, which is normal for a question that gets asked by people who do this for a living: 150 searches a month in the United States for the older helpful content update phrasing, 100 for the direct penalty question, 80 for the AI content and SEO pairing, 70 each for scaled content abuse and Google's AI content policy, 50 for people-first content itself.
Monthly US search volume across the Google AI content guidance cluster, indexed to the largest term at 150 searches a month (Ahrefs, October 5, 2026)
The shape of the result set is the more interesting half. Domain Rating across the ranking pages runs from 68 up to 100, which is as authoritative a set as you will find on any question. Page-level equity is almost absent from all of it.
| PAGE IN THE RESULT SET | DOMAIN RATING | URL RATING | WHAT THE PAIRING TELLS YOU |
|---|---|---|---|
| Google's own 2023 guidance post on AI-generated content | 100 | 23 | The only page in the set with real page-level equity, and the one everybody links to when the question comes up |
| A Google support forum thread on the same question | 100 | 4 | Position three held by a user thread on the strength of the domain alone |
| A major SEO platform's data study | 92 | 4 | Domain Rating 92 and effectively no page equity, ranking on structure and trust |
| A business press write-up of a third-party study | 91 | 4 | Same pattern again, one position lower |
| An agency explainer | 68 | 4 | The lowest authority in the set, holding a top ten position against all of the above |
| Community and question-and-answer threads | Not rated at page level | Not rated at page level | Reddit at position five and two Quora threads at seven, which is demand the publisher set is not satisfying |
Two readings follow. The first is the familiar one: when every page in a result set carries a URL Rating of 4, links are not the sorting mechanism, so structure and trust are, and the barrier to being the best answer is editorial rather than financial.
The second is the one I would act on. Google's related questions on this result set are whether AI content is allowed at all, whether Google penalises it in 2026, and whether Google can even detect it. Three questions, and the October 1 page answers all three more precisely than any of the ranking pages do, because it reframes the question from detection to effort. The publisher set has not caught up, and community threads are occupying the gap.
What to do before the next spam update
The work worth doing this month is an audit of effort and accuracy evidence on your highest-value pages, plus a pass over the metadata the new main content definition just absorbed, rather than a sweep to swap out author photographs.
Order matters here, because the cheap task is the one that feels urgent and the expensive task is the one that moves the number.
Then there is the sequencing question, which is the part clients get wrong most often. The temptation after a guidance change is to start a site-wide remediation project. That is almost always the wrong first move, because the next ranking event will not hit the site evenly and you want to know which attribute is exposed before you spend against it.
That last play is the one that pays for the rest. We built the same instrumentation into the read on the September 2026 spam update's phased rollout, and it is the difference between a diagnosis and a committee. It is also the first thing we stand up in our SEO and GEO audits, for the plain reason that you cannot score four attributes across a site without a view of which templates carry the value.
None of this requires a new tool, a new vendor or a budget conversation. It requires the quality process you already run to cover the parts of the page Google has just told you it counts, and it requires somebody to own the word accuracy. That is roughly a fortnight of attention inside a normal content marketing program.
You are not behind on this. The guidance moved four days ago and almost nobody has acted on anything except the headshots. What you have is a short window where the obvious response is also the shallow one, and the useful response is sitting in a paragraph the trade press skipped.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.