Publishing more articles stopped working. Here is what replaced it

A client asked us in July to write forty articles. The reasoning was sound on its face: competitors had more pages, more pages meant more keywords, more keywords meant more traffic. It is the arithmetic the industry ran on for years.

We said no, and suggested six. This is the argument we made.

What actually changed

Google introduced a spam policy it calls scaled content abuse in March 2024. The definition is blunt: producing many pages primarily to manipulate rankings rather than to help anyone. The policy applies regardless of who or what produced the pages. Human-written filler and machine-written filler are treated identically.

For two years that policy sat there without much visible enforcement. Then the March 2026 core update enforced it properly, and the results were not subtle. Sites that had accumulated rankings through high-volume publishing without editorial oversight reportedly lost somewhere between half and four-fifths of their organic traffic inside two weeks. The pattern was consistent enough that you could identify affected sites from the shape of the graph alone.

The important detail, and the one most people get backwards: none of this is about whether a tool was involved in the writing. Google's own guidance says generative AI is useful for researching a topic and adding structure, and warns specifically about generating many pages without adding value. The production method was never the trigger. The trigger is volume without originality.

A human writing forty thin pages and a machine writing forty thin pages produce the same result in search. The forty is the problem.

Which is why the "is it detectable" question is the wrong one

We get asked regularly whether content can be made undetectable as machine-assisted. It is an understandable worry and it is aimed at the wrong target.

Google's spam systems do not attempt to prove authorship. They look at behavioural and structural patterns that correlate with mass-produced low-value content — how fast pages appeared, how similar they are to one another, whether anything on them could only have come from someone who has actually done the work. A site that went from four pages to four hundred in six weeks is legible from the outside no matter how the sentences are arranged.

So the question worth asking is not whether a page reads as machine-written. It is whether the page contains anything a machine could not have produced on its own. Those are very different standards, and only one of them survives an update.

What that means in practice

Originality in this context is unglamorous and specific. It is the number from your own campaign. The thing that went wrong on a project and what it cost to fix. The screenshot of the actual dashboard. The opinion you hold that your competitors would dispute. A position you can defend in a meeting.

None of that can be researched into existence. It has to be extracted from the people who did the work, which is slower than writing and is the entire reason most content programmes avoid it.

The corollary is that six articles containing real material will outperform forty that do not, and will keep outperforming them after the next update. We have watched this happen on client sites for two years and it has stopped being a debate internally.

A test before you publish anything

Read the draft and ask what is on the page that could not have been assembled from the first ten results for the same query. If the answer is nothing, the page has no reason to exist, and publishing it makes the rest of the site slightly worse rather than slightly better.

That last part surprises people. Thin pages are not neutral. They dilute the crawl budget, they compete with your own stronger pages, and at sufficient volume they change how the whole domain is assessed. Pruning weak content is a recognised recovery tactic precisely because the pages were doing harm while they sat there.

The rhythm we actually recommend

One substantial piece a fortnight, sourced from a real project, reviewed by someone who knows the subject, and linked properly into the rest of the site. Twenty-six pieces a year of that kind is a serious body of work. Two hundred of the other kind is a liability.

This is also why we write our own Insights at the pace we do rather than filling a calendar. If you want the same approach applied to your site, that is most of what our SEO work consists of — and the companion piece on structured data and AI search covers the technical half of the same problem.