Did Google finally go after AI slop? It never said so – but on 1 October it quietly added two passages to the English Helpful Content guidelines. Several translated versions, including Polish, still show last year’s text.
The September spam update is the fourth this year. Google said the rollout may take up to two weeks and, as I write this, has not announced its completion. Several people I know saw heavy drops on a very characteristic type of content. So when the new paragraphs appeared on 1 October, I read them through the question that keeps coming up in those conversations: what are those pages actually for?
In the cases I have seen and in the public case studies, the picture is consistent: a lot of URLs, text topped up by AI or assembled from a data feed, and very few reasons for a human to open that particular page. Google does not call this “AI slop” – and I suspect the definition is less clear-cut than it may seem. What the new documentation does describe is concrete: no original effort, no originality, and main content that fails to fulfil the page’s purpose.
To be clear: I have not reverse-engineered any spam update. But I see a recurring pattern, and the common denominator is a page that cannot answer one question – why should the user stay here?
What changed in Google’s helpful content guide
The English version of “Creating helpful, reliable, people-first content” was updated on 1 October 2026. It gained a section defining main content with four quality attributes, and a paragraph warning against fabricated author profiles. I consider both important for where search is heading.
Here is the part that rarely gets mentioned: Google’s documentation is written in English, and every other language version is a translation that arrives later. Sometimes months later. I work in Polish, so I can show you exactly what that looks like – the Polish version still carries the date 18 December 2025.
| Passage | English (1 Oct 2026) | Polish (18 Dec 2025) |
|---|---|---|
| “Main content” section with the definition | present | missing |
| Four attributes quality raters assess | present | missing |
| Warning about fabricated author profiles | present | missing |
| The question about automated contenta pre-existing difference, not an October addition | “substantially generate” | “na dużą skalę” – at scale |
If your team runs on a localised version of Google’s documentation, this is the practical cost: you are working without clarifications that have been sitting in the English original for months. The criteria themselves may well be older than the documentation – but the explanation of how raters apply them exists in one language only. And the last row shows a subtler problem than lag: a translation can quietly narrow what the original asks. I come back to that one later.
One more thing about the timing. The Search Central changelog logged an update to the separate guidance on generative AI content on 1 October. The helpful content page carries the very same date in its footer, but got no changelog entry of its own. There is also no statement connecting the new paragraphs to the August or September spam update. A curiosity worth noting, because the chronology tempts me to write a far more dramatic article than this one.
What is in the generative AI guidance? It is a short document and, despite the title, it does not prohibit using AI. It opens by saying generative AI can be particularly useful when researching a topic and adding structure to original content; the caveat is about generating many pages without adding value, which may violate the scaled content abuse policy. Google then points directly at the quality rater guidelines and names the sections: 4.6.5 on scaled content abuse and 4.6.6 on main content created with little to no effort, originality or added value. The core of the document is one predictable instruction: factcheck manually before publishing, because a language model predicts a likely sequence of words rather than retrieving facts. That review covers titles, meta descriptions, structured data and image alt text – everything that can surface in search results. Finally, Google encourages explaining to readers how the content was made, and for e-commerce defers to Merchant Center policy: AI-generated images must carry IPTC DigitalSourceType metadata with the value TrainedAlgorithmicMedia, and AI-generated product data submitted to Merchant Center must be specified separately and labelled as such.
Page purpose: what can the user actually do here?
Google defines main content as the part of the page that directly helps it achieve its purpose. That can be an article or a video, but equally a calculator, a search function, an interactive map, user reviews, or information sitting behind a tab.
Google Search defines “main content” as any part of the webpage that directly helps the page achieve its purpose.
Google Search Central, Creating helpful, reliable, people-first content, “Main content”
I like this definition considerably more than another round of debate about optimal word count or chunking. If a page promises to calculate your mortgage payment, its most important content is a working calculator. If it promises a comparison, I need an actual comparison. Three paragraphs of preamble about how important an informed choice is will not deliver the purpose on the page’s behalf.
Which makes the useful question for an article: after reading it, do I know something I would not have got from the five results above it? And for the next location page in a set: what does it give me beyond a swapped-in city name?
Asking about purpose is not a new rule. A page can simply be weak without violating any policy. But when you mass-produce near-identical URLs mainly to capture query variants, and each one gives the user roughly the same thing, you are in the territory the policies call scaled content abuse. Three things matter there at once: scale, little value for the user, and a purpose aimed at manipulating rankings – not the mere fact that somebody dislikes an article. There is a digression waiting here about how SEO rests on influencing rankings and how organic results were never really free, but that is material for another piece, or for a conference corridor.
Why this update reminded me of the spam updates
The August spam update ran from 18 to 21 August. The next one started on 24 September and is still rolling out as I write. Google has not said what kind of spam the second one targets.
After the August update, Glenn Gabe documented four sites with heavy drops:
- a site in an ultra-YMYL niche combining mass-produced programmatic pages with AI-generated passages,
- an affiliate site whose categories and product pages were built mainly from Amazon data and linked straight back to the store,
- a large site where many pages offered a photo and a set of facts available anywhere else,
- a site scaled to a quarter of a million URLs, redirecting users to even riskier destinations.
It makes for fascinating reading, and these are very different cases: from mass content production to misleading functionality and redirects. The common denominator is the question we already have:
Does the page actually give the user what they came for?
In September, Gabe again reported drops on sites scaled with AI and built programmatically. Lily Ray pointed early to sites generating a separate page for every phone number or area code. Barry Schwartz collected it all at Search Engine Roundtable. These are first observations during an ongoing rollout, not an analysis of the full update.
So my hypothesis is narrower than the dates suggest: the new passage describes the questions worth asking of pages hit by the recent updates. It does not tell us which signals fired, or why a particular domain dropped. Documenting a criterion on 1 October does not mean Google started applying it that day.
Four attributes of main content: effort, originality, skill, accuracy
The new section lists four aspects of main content that quality raters assess: effort, originality, talent or skill, and accuracy. Effort can mean original analysis, but equally the work that went into building a tool. Originality means information or a perspective that is not already everywhere else. Skill shows in whether the text is clear, the video well made, the tool actually functional. Accuracy carries particular weight on YMYL topics.
Definition for non-SEOs
YMYL (Your Money or Your Life) is Google’s term for topics that could significantly affect people’s health, financial stability or safety, or the welfare of society. Google’s systems give more weight to E-E-A-T signals on these topics, and main content is expected to be highly accurate and consistent with established expert consensus.
For originality, Google has its own term, defined in the guide to optimising for generative AI features from May 2026. I analysed that guide separately; what matters here is the contrast it draws between commodity and non-commodity content.
The term is borrowed from economics: a commodity is sugar, copper, wheat – a product where it makes no difference who produced it, because every batch is the same and only the price matters. Commodity content works the same way: it rests on common knowledge, anyone could have written it, and the reader loses nothing by taking a different result instead. Google’s own examples: a generic 7 Tips for First-Time Homebuyers against a specific Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line.
Which is the question sitting in my head whenever I start writing: could what I just wrote be assembled from five other results without reaching for mine? If yes, the conclusion is unpleasant – my content is easy to replace and carries none of my own perspective. Back to the draft.
These four attributes are not a list of new ranking factors. Google is describing the work of quality raters, and states in the same guide that their ratings do not directly set positions. They help evaluate whether the ranking systems are working.
Marie Haynes reads the guide more broadly, as a description of what Google wants to reward in results. That is an interpretation, and one well worth reading – it prompted a good deal of my own thinking here.
What about sources, in the age of AI-assisted research?
If you work with AI, you know how much faster source research has become. Google takes a clear position in the updated guidance: citing sources does not replace original effort. You can add fifteen footnotes and still have no observation of your own in the text. Good sourcing and originality are simply two different things. A footnote shows where the information came from, and it genuinely raises the value of a piece – but it does not yet show the work that went into it.
Attribution or giving credit to other sources doesn’t replace the need for original effort.
Google Search Central, Creating helpful, reliable, people-first content, “Main content”, “Effort”
My take. My latest book rests on a similar distinction: a claim is a message sent in one direction, while proof exists independently of what a brand says about itself. Content works the same way. Fifteen footnotes show what I read. They do not yet show what I did with it.
Google also names examples of content with little or no effort: pages generated automatically from feeds, and large amounts of text produced by AI “without manual oversight or curation”.
Which brings us back to AI slop – or at least to how I understand it. A model can produce text that is fluent and full of references. With a little work it will pick credible sources and avoid the phrasings people recognise as “AI voice”. That is still not enough if nobody checked the facts along the way, nobody did the work of taking a position, and nobody asked whether the piece helps one specific reader.
Book – Polish only
A claim is a message sent in one direction. Proof exists independently of you. Google now assesses effort and originality, and language models assemble your image from what other people say about you. Marka osobista w czasach AI i generatywnego wyszukiwania is about building a trace that holds up without you in the room. Currently available in Polish only.
More about the book (PL) →The author who exists only in the byline
The second new paragraph concerns fabricated creator profiles: made-up names, AI-generated headshots, false credentials. I have run my projects under my own name for as long as I can remember – it has caused me a few problems along the way, but I still think it is the right call. (Yes, in affiliate projects this is harder, and no, it does not surprise me that Google is pushing back.) So I smiled a little, because Google calls this what it is. In fairly strong words.
Fabricating creator profiles […] is a form of deception.
Google Search Central, Creating helpful, reliable, people-first content, “Who (created the content)”
My take. I do not read this as a requirement that a recognisable expert stands behind every piece. I read it more simply: if you are putting a name on content, that name should belong to someone who exists and had something to do with the text. A generated face attached to an invented author adds no credibility – it simulates human expertise, and that is what Google reads as a low-quality signal.
It is aimed at a very specific shortcut: Google asks about the author, so let’s produce one. A photo, a few lines of bio, Person schema, done. Except a profile does not create experience for a person who does not exist. On what you can show instead of a claim, I wrote earlier in a piece on E-E-A-T in practice.
I would not conclude that every expert now has to be a public figure or maintain a presence beyond their own site. Information about the author simply needs to be true and proportionate to their part in making the material. Looking further ahead, though, I do think personal branding matters for search.
Not every piece of content needs a specialist
In the section on main content quality, Google adds a caveat that seems to get lost: not all content requires specialised expertise. The documentation’s own example is someone sharing a new way they found to clear snow off their driveway. Let me give you a different one.
I run a keto food blog, ms-fox.pl, and travel blogs about Bornholm and Madeira. I work in SEO – I am neither a dietitian nor a tour guide. But I developed and cooked those recipes, and wrote three books from them. I explored both islands by car and took the photographs myself. I have first-hand experience of travelling with a dog and my own perspective on living in each of those places. Experience is the qualification here, and it does not need a title to prop it up. It does need to be shown – through your own photos, the date you visited, what the ticket cost at the time, the note that a recipe took three attempts and that one ingredient works better than another.
YMYL topics work differently, because there Google expects consistency with established expert consensus. Lived experience still has value, but it is not a basis for advice that contradicts that consensus. On my keto blog, which sits inside YMYL, I draw the line like this: telling you what I experienced is one thing, telling you what to do is another. I guard that line carefully.
I write with AI. So where do I draw the line?
I write a lot. Sometimes with AI, for years without it – and I wrote about that tension in an essay on authorship. I have no neat rule along the lines of “the model may fix commas but must not touch sentence structure”. There are stages where the tool genuinely helps: research, planning structure, editing, a kind of brainstorming when I cannot find a precise way to say what I mean. Google says as much outright – generative AI can be particularly useful when researching a topic and adding structure to original content.
What I do know is when it starts getting convenient in the wrong way:
- when I accept a paragraph from the model because it sounds smart, though I could not defend it,
- when I ask for something “for completeness” and then cannot say why the reader should read it,
- when fluency and the argument I am making start to matter more than the facts.
That line is reasonably consistent with Google’s own documentation.
My take. The chapter on AI-generated content in my book (written between January and March 2026) rests on one claim: a human has to answer for published content, whatever tools were used along the way. A byline is a declaration of that responsibility. Both texts, mine and this guidance, draw on the quality rater guidelines, which have described scaled content abuse for years. I did not work anything out ahead of Google – I just read their document carefully.
The nuance that disappears in translation
Back to that last row in the table. The English guide asks about situations where automation is used to substantially generate content. The Polish translation asks about generating content na dużą skalę – at scale. Those are not the same question. A single article can be substantially model-generated without belonging to any mass production run.
This is an older passage, not an October addition. But it is easy to read the translated version as being about industrial-scale automation only, while the original can just as well cover one text.
| English | Polish |
|---|---|
| If automation is used to substantially generate content, here are some questions to ask yourself | Jeśli do generowania treści na dużą skalę używasz automatyzacji, odpowiedz sobie na te pytaniaif you use automation to generate content at scale, ask yourself these questions |
My take. This kind of slippage is not new, and it is not specific to Polish. Documentation is written in English; translations arrive later and are sometimes less precise than the original. Which is why most SEOs I know work from the source version. Build a process on a translation and you may end up answering a narrower question than the one Google is actually asking. These small differences occasionally matter a great deal.
Google does not require labelling every text that involved AI. It encourages explaining the process where a reader might reasonably wonder how something was made. The separate generative AI guidance, updated on 1 October, also stresses manual factchecking before publication – including titles, meta descriptions, structured data and alt text.
So what should you check on your own site?
I would not respond to this update with mass content pruning, deleting older articles, or bolting longer bios onto every author. Those things may be needed, but they need a strategy behind them. I would start with the pages that actually lost visibility, and three questions about each:
- What does the user get on this specific page? Does the main content deliver what the title promised, and does it hold value distinct from your other pages?
- How was this page made? Does it bring original analysis, experience or a useful function – or is it one more variant of content available elsewhere, on your own domain or a competitor’s?
- What does the data really show? Did specific groups of URLs and queries drop, or the whole domain? When exactly did it start, and what changed on the site around that time?
That last question matters before you connect a drop to a spam update. Matching dates give you a starting point for analysis, not a diagnosis. All the more so while the September rollout is still running.
On my own sites I see no effects from these updates – though I have seen visibility drops before, because that is what SEO work looks like. I read the new passages mainly as a reminder that every page should be able to answer one very ordinary question: why should the reader stay here? AI can fill the space under a headline in seconds, and filling space for its own sake is what I understand AI slop to be. Google does not use the term, but it describes the problems that usually add up to it: no original effort, no originality, main content that fails the purpose of the page. The answer to the question about value for the reader still has to be worked out by a person.
Quotations come from Google Search Central documentation (Creating helpful, reliable, people-first content, as of 1 Oct 2026; Google Search’s guidance on using generative AI content on your website, as of 1 Oct 2026; Optimizing your website for generative AI features on Google Search, published 15 May 2026), available under CC BY 4.0. Translations from Polish are mine. Spam update dates from the Google Search Status Dashboard. Accessed 4 October 2026.
