The scaled content abuse policy does not ban AI writing

Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users. This abusive practice is typically focused on creating large amounts of unoriginal content that provides little to no value to users, no matter how it’s created.
Spam policies for Google web search, Google Search Central

Two sentences. That is the whole definition, it is free to read, and it does not contain the word “AI”. That is worth sitting with, given how much of the commentary about it is about nothing else.

Read the last clause again. No matter how it’s created. Method is not the test.

That is my claim, and I think it is right. It is also the claim that draws the sharpest objection I have heard on the subject, which is that the distinction is a lawyer’s distinction with no cash value. This piece states the text, then states that objection as strongly as I can put it, then says where I think it fails.

I am quoting the version of the spam policies page that carried “Last updated 2026-05-15 UTC” in its footer on the day I read it. The same page lists five examples, and only the first mentions generative tools:

Using generative AI tools or other similar tools to generate many pages without adding value for users
Scraping feeds, search results, or other content to generate many pages (including through automated transformations like synonymizing, translating, or other obfuscation techniques), where little value is provided to users
Stitching or combining content from different web pages without adding value
Creating multiple sites with the intent of hiding the scaled nature of the content
Creating many pages where the content makes little or no sense to a reader but contains search keywords
Spam policies for Google web search, Google Search Central

Every one of the five turns on the same two conditions: many pages, and no added value. Neither condition mentions a model. A policy that bans writing with a language model and a policy that bans flooding an index are two different policies, and Google published the second one.

What changed in March 2024, and what did not

Google announced the policy on 5 March 2024, alongside the March 2024 core update, in a post by Chris Nelson. Three spam policies launched that day: expired domain abuse, scaled content abuse, and site reputation abuse. The post is explicit that scaled content abuse was a rewrite rather than a new prohibition:

This new policy builds on our previous spam policy about automatically-generated content, ensuring that we can take action on scaled content abuse as needed, no matter whether content is produced through automation, human efforts, or some combination of human and automated processes.
What web creators should know about our March 2024 core update and new spam policies, Google Search Central Blog

The post anticipates the misreading and answers it directly:

Our long-standing spam policy has been that use of automation, including generative AI, is spam if the primary purpose is manipulating ranking in Search results. The updated policy is in the same spirit of our previous policy and based on the same principle. It’s been expanded to account for more sophisticated scaled content creation methods where it isn’t always clear whether low quality content was created purely through automation.
What web creators should know about our March 2024 core update and new spam policies, Google Search Central Blog

So the widening in 2024 went in the opposite direction from the one people assumed. The old rule named automation and could be dodged by putting a human in the loop. The new rule drops the method entirely and keeps only the purpose and the scale, which means a room of underpaid writers producing the same output is covered too.

Google’s position on AI writing predates all of this by a year, and sits in the FAQ at the foot of the February 2023 post by Danny Sullivan and Chris Nelson rather than in its body. Worth saying, because the difference matters if you go looking. On “Is AI content against Google Search’s guidelines?”:

Appropriate use of AI or automation is not against our guidelines. This means that it is not used to generate content primarily to manipulate search rankings, which is against our spam policies.
Google Search’s guidance about AI-generated content, Google Search Central Blog

And on whether it will rank: “Using AI doesn’t give content any special gains. It’s just content. If it is useful, helpful, original, and satisfies aspects of E-E-A-T, it might do well in Search. If it doesn’t, it might not.”

The objection, at full strength

Here is the case against everything above, and it is better than it usually gets credit for.

Google cannot observe purpose. Purpose is a fact about a publisher’s head, and no crawler reaches it. What Google can observe is the artefact (the page, the site, the rate at which pages appeared), so “primary purpose” has to be inferred from surface features. And the surface features that read as generated at scale are, in 2026, overwhelmingly the features of machine-written text: uniform section structure, uniform length, hedged summary prose, citations that name a publisher and no document.

Then close the loop. Google’s November 2024 site reputation abuse post says the company does not “simply take a site’s claims about how the content was produced at face value.” So the publisher cannot rebut the inference by explaining the process, and a disclosure page will not do it either.

Put those together and the objection lands: a rule that says no matter how it’s created, enforced by inference from features that machine writing produces, is a rule about machine writing in everything but wording. “Method is not the test” is then true on paper and worthless in the hand: a distinction you cannot cash, drawn by the only party who gets to score the exam.

I do not think that argument is silly. I think it is half right, and the half that is wrong is the half that matters.

Where it fails

It fails on the same document it appeals to. If the policy were secretly about detecting machine authorship, the rater instructions would tell raters to look for machine authorship. They tell them the opposite.

Section 4.6.5 of the quality rater guidelines, in the version dated 11 September 2025:

Pages and websites made up of content created at scale with no original content or added value for users, should be rated Lowest, no matter how they are created. Even if you are unsure of the method of creation, e.g. whether or not the page is created using generative AI tools, you should still use the Lowest rating when you strongly suspect scaled content abuse after looking at several pages on the website.
General Guidelines, 11 September 2025, section 4.6.5

Uncertainty about the method does not protect the page, and it is not supposed to resolve into a guess about the method either. The rater is told to judge the output and to stop asking.

Two neighbouring instructions show the same standard cutting against human work. Section 4.6.6 applies the Lowest rating to copied or paraphrased main content “even if the page assigns credit for the content to another source”, which catches hand-built aggregation sites with impeccable attribution and no model anywhere near them. And section 3.2 defines the yardstick as labour: “Effort: Consider the extent to which a human being actively worked to create satisfying content,” with the counter-example given as “the automatic creation of thousands of pages by running existing freely available content through existing translation software without any oversight, manual curation, etc.”

The operative words in that counter-example are oversight and manual curation. Both are human acts, both are performable on machine output, and neither is available at ten thousand pages a week. That is the axis the policy actually runs along, and it is not the same axis as authorship. The two come apart in both directions: a content farm staffed entirely by people fails, and a page a model drafted, a person checked against named sources, and a person put their name on, passes.

So the objection is right that intent is read off the artefact. It is right that you cannot talk your way out. What it gets wrong is the conclusion, because the artefact is being read for evidence of work, not for evidence of a model. Evidence of work is something you can put on the page on purpose. Google’s helpful-content page even tells you one of the things it wants to see there: “Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?” A disclosure will not launder a thin page. It is still the answer to a question the policy asks out loud.

What the policy prohibits, in plain terms

Three tests, all of which have to fail before you are in trouble:

Scale. One page is not scaled content abuse. The word “many” is in every version of the definition.
Purpose. The pages exist mainly to rank, not mainly to answer someone. The helpful-content page frames this as the “Why” question, and it is the only one of the three that is about intent rather than artefact.
Absence of added value. Section 3.2’s effort standard, above, is what the output gets measured against.

Nothing in there says do not use a model. It says do not ship volume you have not supervised.

One clarification before the end, because the third policy from March 2024 keeps getting folded into this one. Site reputation abuse was rewritten on 19 November 2024 and now reads:

Site reputation abuse is a tactic where third-party content is published on a host site mainly because of that host’s already-established ranking signals, which it has earned primarily from its first-party content. The goal of this tactic is for the content to rank better than it could otherwise on its own.
Spam policies for Google web search, Google Search Central

The November rewrite removed the escape hatch that first-party oversight used to provide; Google’s stated reasoning was that reviewing real cases showed “no amount of first-party involvement alters the fundamental third-party nature of the content.” Different policy, different failure, and a coupons subdirectory is not a scaled content problem.

The practical reading

Draft a page with a model, check it against sources you can name, put your name on the result, and this policy is not about you. Point the same model at a directory and fill it with eight hundred pages nobody asked for, and it was written about you specifically. Since March 2024 it has not mattered whether a person pressed the button.

What I cannot tell you is how any of this is enforced against a particular site. I have no manual action notice to show you, no message from Google, and no interest in inferring a verdict from a traffic chart. The documentation on generative AI content is careful about the same boundary from the other side: the rater guidelines “are not a guide to ranking first in Google,” and raters’ “ratings don’t directly influence ranking.” What the guidelines give you is the standard, written down, by the people who wrote it. Read as a description of what a sceptical stranger is told to look for, it is unusually clear. Read as a rule about robots, it is unreadable, which is why so much of the anxiety about it lands on people it was never drafted for.