What the scaled content update actually punished

When the scaled content abuse guidance landed, half of marketing Twitter concluded that Google had banned AI writing. That reading is convenient if you sell writing services and wrong if you read the documentation. The word the guidance keeps returning to is not authorship. It is oversight.

We spent about three weeks going through sites that visibly lost traffic in the weeks after, mostly through public case write-ups and a set of forty domains we had already been tracking for other reasons. The pattern that showed up was boring and consistent.

The losers had the same four fingerprints

• Publishing cadence that jumped without explanation. One site went from four posts a month to two hundred and thirty in six weeks. Nothing else about the business changed.
• Near-identical article skeletons. Same headings, same order, same closing paragraph structure, topic after topic.
• No entity depth. The articles named the topic but never named a product, a price, a customer, a place, or anything else that would require knowing the business.
• No visible human. No author, no editing history, no updates after publication, no corrections.

None of those four are detectable by asking whether a model wrote the text. All four are detectable by looking at the site.

Sites that used AI and were fine

The counter-examples matter more than the casualties. We found publishers who were open about using models heavily and who came through the update flat or up. What they had in common was an editor who actually rejected things. One B2B site we looked at publishes about thirty pieces a month, all model-drafted, and their editor kills roughly a third before publication. Their traffic did not move.

That is not a moral point about human craft. It is a mechanical one. A rejection rate above zero means something is being filtered, and whatever survives the filter tends to carry information the model could not have invented on its own.

What we changed because of this

Two things, both unglamorous. First, the review step is not optional and never will be, including on the highest plan. Second, we cap how much a single site can publish in its first sixty days regardless of what plan it is on. A brand new site that suddenly produces eighty articles looks exactly like the pattern above, even when every article is good.

Customers occasionally ask us to lift that cap. We say no, and this post is the long version of why.

The part nobody wants to hear

If your plan is to publish faster than a competitor and win on volume, this category has stopped rewarding that. The sites that recovered did it by cutting pages, not adding them. One removed about 60 percent of its library and got most of its traffic back within two months, which tells you something about what the remaining 40 percent was worth.

The question is not whether a machine wrote it. The question is whether anyone would notice if it disappeared.

That is a harsh test and most content fails it. Ours does too, sometimes. The difference we care about is whether you find out before you publish or six months after.