Google rolled out its third spam update of 2026 between 18 and 21 August, finishing early on the Monday morning. It announced no new policy, named no new target, and said nothing about AI-generated content specifically. Officially, nothing changed.
The traffic data says otherwise, and the pattern in it is worth understanding before you publish anything else.
What lost
The consistent finding across published case studies is that high-volume, lift-and-shift content lost rankings, AI Overviews citations and traffic together. Not one of the three — all three at once, which is what makes this look like a judgement about the pages themselves rather than a ranking wobble.
The losses were not small. One site tracked in Glenn Gabe's analysis lost more than 14,000 queries, meaning those queries stopped returning the site at all rather than returning it lower down. Movement also started roughly a week before Google's stated start date, which is normal and worth remembering the next time you try to date a drop precisely.
What did not lose
Here is the part that gets left out of the panic coverage. AI-assisted pages that carried real imagery, charts and comparison tables held their positions or grew. One site with a domain rating of 17 rose 10% across the update while a site rated 72 fell 30%. Authority did not save the second one, and the lack of it did not sink the first.
So the dividing line is not human versus AI. It is whether a page contains something that took effort and exists nowhere else. A comparison table you actually built, a chart from data you actually pulled, a screenshot of the thing you are describing — those are cheap signals of real work, and they appear to be doing real work.
The rule this enforces
No new rule was introduced. The one being enforced harder is scaled content abuse: publishing many pages with little effort or originality, at volume, primarily to capture search traffic.
Read that definition carefully, because every clause matters. Volume alone is fine — large sites are not penalised for being large. Automation alone is fine; Google has been explicit that how content is produced is not the issue. What gets caught is the combination: many pages, little original value, made to catch traffic rather than to be read.
The honest test is uncomfortable and takes about ten seconds. Pick one of your pages at random and ask whether a person who landed on it would get something they could not get from the first result they already saw. If the answer is no on a page you published, it will be no across the template you published it from.
The separate change to site reputation
Unrelated to the spam update but landing in the same window: from 30 August, manual actions under Google's site reputation policy behave differently depending on where the searcher is. Outside the EEA, a manual action hits the affected section of the site and leaves the rest alone. Inside the EEA the manual action does not apply, but the affected section may be separated in Google's systems so that over time it ranks independently from the rest of the domain.
This targets the practice of renting a trusted domain's reputation to third-party content. If you have ever considered placing your content on a big publisher's site to borrow its authority, the borrowed authority is being systematically unbundled.
If you publish at volume
Four things are worth doing regardless of whether you were hit.
Find your thin template, not your thin page. Programmatic sections fail as a group because they were generated as a group. Sort your pages by impressions and look at the bottom of the list; the pattern will be a template, not a topic.
Add the thing that cannot be generated. Original data, real screenshots, a table you compiled, prices you actually checked. This is the single clearest difference between the pages that survived and the ones that did not.
Consolidate rather than expand. Ten pages that each answer a variation of the same question compete with each other and look like exactly what the policy describes. One good page beats them and is cheaper to maintain.
Be willing to delete. Pages that have never received a click in twelve months are not neutral. They are evidence about how the site is run.
How long recovery takes
Google's position, repeated consistently, is that recovery from a content-side spam demotion requires months of sustained compliance rather than a weekend of edits. That is not a threat, it is a scheduling fact: whatever you fix now, expect to be waiting through several update cycles before you see the benefit.
Which argues for fixing it before you need to.
The bottom line
The update did not punish AI writing. It punished pages that took no effort, and AI happens to be the cheapest way to produce those at scale. The sites that grew through it were using the same tools with something real attached — data, images, comparisons, first-hand checking.
If your content strategy is volume, the fix is not to stop publishing. It is to make each page carry something that would have cost someone time to produce, and to delete the ones that never did.