Scaled Content Abuse: Where AI-Assisted Publishing Crosses the Line

By Muntazir · Published 12 October 2026

In the weeks after Google’s March 2024 core update finished rolling out, a study by Originality.ai tracked roughly 79,000 content-focused sites monetizing through three ad networks and found 1,446 of them deindexed — about 1.9%. Three of those sites had each been pulling in over a million visitors a month. All three fell to zero.

None of them were caught for using a particular writing tool. They were caught under scaled content abuse, a policy Google introduced that same month to cover content produced mainly to move a ranking rather than to help whoever searched for it.

What does the policy actually define as abuse?

Google’s own spam policies page states it plainly: scaled content abuse is pages “generated for the primary purpose of manipulating search rankings and not helping users,” typically a large volume of unoriginal material that provides little to no value “no matter how it’s created.” That last clause is doing real work. It puts a hand-staffed content farm and an AI-run one under the same rule, because the policy was written around the output and the intent behind it, not the tool that produced it. Scaled content abuse is itself a March 2024 rename and broadening of an older, narrower rule that only covered content clearly produced through automation.

Does writing with AI assistance break this by itself?

Not according to Google’s own description of the change. The March 2024 announcement is specific that the behavior being targeted is “producing content at scale to boost search ranking,” and that this can happen whether automation, humans, or some combination of the two is doing the writing. The post goes further, conceding that “whether content is created purely through automation isn’t always as clear” as it used to be — an admission that the old test, is a script writing this, had already stopped being a useful filter before the policy was rewritten around it.

That matters for a different reason than it might first appear. Writing for how a model reads a page has become a real discipline, and most of the advice in it has nothing to do with authorship. A page can be entirely AI-drafted and still clear this policy by a wide margin, the same way a page written entirely by a person can still fail it.

What did enforcement actually look like?

The aggregate numbers back that framing up. Google projected the March update, combined with earlier spam work, would cut “low-quality, unoriginal content” in search results by 40%; once the rollout finished on 19 April 2024, it reported the actual figure at 45%. Underneath that aggregate, Originality.ai’s sample is the sharper picture: sites monetizing through MediaVine, Raptive and Ezoic, 1.9% deindexed, and a cumulative estimated loss above 20 million visitors a month. Three of those domains, named in the study — zacjohnson.com, beingselfish.in and equityatlas.org — were among the ones that fell to zero after each had previously drawn more than a million visitors a month.

Where’s the actual line, for a site that publishes often?

This is the part Google has never spelled out. No page count, no publishing cadence, no ratio of reviewed-to-unreviewed content is named anywhere in the policy as the threshold that tips a site from fast into scaled. That’s presumably deliberate — a published number would just become the new target to stay under — but it also means nobody outside Google can point to a bright line, only a direction of travel.

This post is itself a small test case of that direction. It was drafted with AI assistance and reviewed by a person before it went live, which is closer to the actual distinction the policy is drawing than “AI or not” ever was: not whether a tool helped write something, but whether anyone can account for a specific page afterward — why it exists, what it’s claiming, and whether the claim holds up. A publishing operation that can’t answer that question for most of what it ships is the pattern Google is describing, independent of which words appear in its masthead.

Where does this sit next to the other manipulation tactics?

Scaled content abuse is one of three related policies Google tightened in the same stretch. Black-hat tactics retargeted at an AI reader — cloaking, hidden instructions, fabricated reviews — deceive the reader directly rather than relying on volume. Parasite SEO borrows someone else’s domain authority instead of producing enough of anything to need a volume policy at all. All three assume the same shift: a model composing one answer from a handful of sources raises the payoff for anything that nudges a page into that handful, whether by lying to the reader, borrowing credibility, or simply outproducing everyone else at the keyword level.

The three sites that fell to zero in 2024 weren’t shut down over which software opened first. They were shut down for publishing at a volume nobody on staff could actually stand behind, built to own as many rankings as possible rather than to answer anyone’s real question. Swapping which tool drafts the next batch of pages doesn’t change that math — the volume itself was usually the business model, not an input to it.