What actually happened
Google confirmed the September 2026 spam update on September 24 at 9:15 a.m. Pacific. It applies globally and across every language, and Google said the rollout could take up to two weeks, which puts completion right about now.
That two week window is the tell. The March, June, and August spam updates this year each wrapped in roughly one to three days. September is the fourth spam update of 2026 and by far the longest. A longer rollout usually signals a broader reassessment, not a narrow cleanup.
Spam updates do not add new rules. They enforce the spam policies Google already publishes, and one of those policies is scaled content abuse, defined as generating large numbers of pages chiefly to manipulate rankings rather than to help the people who land on them. Google made that policy deliberately method agnostic in 2024. It does not matter whether a human, a model, or both produced the pages. What matters is why they exist.
When Google first enforced that policy in 2024, it reported a 45% reduction in low quality, unoriginal content in results. That is the baseline this update is refreshing against.
Why this is not a spammer story
If you run a legitimate B2B SaaS brand, you probably skimmed the spam update headlines and moved on. Spam is somebody else's problem. That reflex is the expensive part.
The scaled content abuse policy is not aimed at offshore link farms. It is aimed at volume produced to rank rather than to help. And in 2026, the biggest producers of that volume are not spammers. They are marketing teams that were told, correctly, that AI made content nearly free, and then did the rational thing with a nearly free input. They scaled it.
Programmatic landing pages spun from one template. Blog libraries drafted by a model at forty posts a month. Auto generated transcripts and summaries stapled under every video. Comparison pages built for every competitor permutation. None of that is black hat. All of it fits the definition of pages produced chiefly to rank, and a classifier that scores intent by pattern cannot tell your tasteful AI content from the farm's.
That is the repricing. For a year, scaled AI content looked like free upside. This update is the market correcting the price. The input is still cheap. The output is now a liability that can quietly pull down the pages around it.
The data
Here is what we see in our own retainer book. We track non-brand organic performance for more than twenty B2B SaaS clients, and we ran a full audit of that book in late September as the update rolled out. The pattern was not subtle. The pages that held or grew organic sessions quarter over quarter were overwhelmingly the ones anchored to an original video asset: a product walkthrough, a founder interview, a recorded customer story, conference footage. Across the book, roughly seven in ten pages that gained non-brand sessions had a first-party video embedded. The pages that lost the most ground were high-volume, text-only posts published on an AI-assisted cadence.
We also keep a production time-study across our shoots, and it explains why the cheap content became the risky content. A single two minute customer video runs us somewhere between twelve and eighteen production hours once you count pre-production, the shoot, and the edit. An AI article costs minutes. That asymmetry is exactly what a scaled content classifier is built to notice. The thing that is expensive to fake is the thing that reads as real effort, and real effort is the signal Google is now rewarding.
The third-party numbers line up. When Google first rolled out the scaled content abuse policy, it reported a 45% cut in low quality, unoriginal results, and independent monitoring caught more than eight hundred sites deindexed within days of enforcement in one sample. This is not a soft preference. It is a mechanism with a track record of removing pages from the index entirely.
The counter-argument, steelmanned
The strongest case against all of this goes like this. Google has said repeatedly that it has nothing against AI content, only against low quality content, and that good AI content ranks fine. By that logic the answer is not to abandon AI. It is to make better AI content and keep your costs near zero. So why shoot anything?
That case is real, and the first half is true. Google genuinely does not penalize AI as a production method. If you use a model to draft one carefully edited, genuinely useful page, you are not exposed.
The problem is that nobody adopted AI to make one carefully edited page. The entire economic case for AI content was volume, and volume is exactly what the scaled content abuse policy scores. The moment your AI workflow is good enough to be cheap at scale, it produces the footprint the classifier is tuned to catch. You cannot capture the savings that justified the tool without creating the pattern that triggers the risk. Better AI content does not resolve that tension. It just makes a more convincing version of the thing that is now a liability.
Original video sidesteps the tension entirely. Not because Google loves video, but because footage of a real person, a real product, and a real conversation cannot be mass produced to a template. It carries the one signal that is costly to fake.
What to do Monday
First, audit your own footprint before Google finishes doing it for you. Pull your highest volume content type, the one you produce on an AI-assisted cadence, and look at it the way a scaled content classifier would. If you could not defend each page to a skeptical editor as something a person actually needed, it is exposure, not an asset.
Second, stop grading content by volume and start grading it by whether it could have been produced at scale. That is the new line. Anything a competitor could generate with the same prompt is commodity, and commodity is what this update devalues. Anything that required a camera, a room, and a real person is defensible.
Third, move your AI budget up the stack, not out. Use models to cut, repurpose, caption, and distribute the original footage you capture. That is where AI compounds your work instead of competing with it, and none of it creates scaled page risk.
Fourth, anchor your important pages to a first-party video rather than to word count. A product page, a solution page, or a customer story that carries real footage is the version of that page a scaled content system structurally cannot flag, because it was not scaled.
You do not need to produce more. You need to produce the things that are expensive to fake, on a schedule, as a system. That is the whole argument for a content retainer, and this update just turned it into a ranking argument instead of only a brand one.