TL;DR: On 18 August Google started a “spam update” — a change to the automated systems that decide which websites Google trusts enough to show people. It runs worldwide, in every language, and Google said it would take a few days. As of today it’s still going. The update itself is routine. The rulebook behind it isn’t: back in May, Google rewrote its definition of spam so that it now covers attempts to influence the AI-written answers at the top of Search, not just attempts to rank higher. This is the second spam update to run under that wording.
First, what a spam update actually is
Google keeps a public list of things it treats as cheating. It calls this list its spam policies. Break them and your site can drop down the results, or disappear from Google altogether.
A spam update is when Google changes the software that enforces that list. Nobody is reviewing your site by hand. A system re-scores an enormous number of pages at once, and some sites move.
This one began at 9:27 a.m. Pacific on 18 August, logged on Google’s Search Status Dashboard — the public page where Google records changes and outages in Search. Google published no blog post to go with it and named no new policy, as Search Engine Land and Search Engine Journal both noted.
It’s Google’s third spam update this year. March finished in roughly 19 and a half hours. June took two days and one hour. The August entry on the dashboard still has no duration next to it, which is how you can tell it hasn’t finished.
The bit that changed in May
On 15 May, Google edited the first sentence of its spam policies page. It now defines spam as “techniques used to deceive users or manipulate our Search systems into featuring content prominently, such as attempting to manipulate Search systems into ranking content highly or attempting to manipulate generative AI responses in Google Search.”
That final clause is the new part. Before, spam meant trying to get a page ranked where it didn’t belong. Now it also means trying to steer what the AI answer says about you — or about a competitor.
Search Engine Land reported the change at the time and read it as extending spam enforcement to a whole category of newer tactics: the ones aimed at getting named inside an AI Overview rather than at position three. Seeded forum threads. “Best tool for X” listicles written to be scraped by a model rather than read by a person. Pages published purely so that an answer engine repeats a particular sentence. We wrote about the seeding version of this earlier in the month.
Nothing about the August update tells us Google is enforcing that clause hard. Google hasn’t said. But the clause is live, this is the second enforcement pass since it went live, and it applies in every language.
Why it will be hard to tell what hit you
Google never says which parts of the policy an update enforces. The dashboard entry says “spam update” and that’s it.
So if a client’s traffic falls this week, Google gives you nothing to attribute it to. Ranking spam and AI-answer spam are the same two words on the same status page.
It gets muddier. PPC Land points out that rank trackers picked up unexplained volatility on 1–3 August and again on 12–13 August, before any of this was confirmed. Movement in that window could have been early deployment, or something unrelated. There’s no way to separate them from the outside.
And the clock is unforgiving. Both Search Engine Journal and PPC Land land on the same figure from Google’s documentation: recovery from a spam action is measured in months, and isn’t guaranteed.
What to do this week
Four things, in order.
- Mark the date. In Search Console, compare 18 August onward against the two weeks before it, broken down by page and by query. Do this before the rollout finishes, while the boundary is still clean.
- Ask two separate questions, not one. Did rankings move? Did citations move? A page can keep its position and still stop being quoted in the AI answer. Those are different failures with different causes, and one report won’t show you both.
- Audit anything you paid for. Placements, sponsored listicles, seeded threads, syndicated “reviews”. If the point of it was to change what an AI answer says, it is now inside the definition of spam. That’s true whether or not this particular update targets it.
- Don’t panic-rewrite. Google’s standing advice on volatility hasn’t changed: fixing the fundamentals beats reacting to a three-day wobble.
Most of that list is checking, then checking again next week, across every page and every client. Preferium runs 47 automated checks over every page and scores each site out of 1000, then finds, fixes, deploys and verifies the fixes on its own — a real browser reloads the live page after each deploy to confirm the change actually landed. It measures 4 AI engines on every plan (ChatGPT, Claude, Perplexity and Gemini), with Google’s AI Overviews and AI Mode tracked separately, so “we lost rankings” and “we stopped getting cited” show up as two different numbers rather than one bad week. More on how the system works.
Key takeaways
- Google began its third spam update of 2026 at 9:27 a.m. Pacific on 18 August. It’s global, covers all languages, and was still rolling out as of today.
- Google published no blog post and announced no new policy alongside it.
- Since 15 May, Google’s definition of spam has explicitly included “attempting to manipulate generative AI responses in Google Search”. This is the second spam update since that wording went live.
- Tactics sold as GEO — seeded threads, listicles built for scraping, pages written for models — are inside that definition now, regardless of what this update happens to target.
- Google doesn’t say which policies an update enforces, so a traffic drop this week can’t be attributed from Google’s side. Track rankings and AI citations as separate metrics or you won’t know which one moved.
- Recovery from a spam action takes months, per Google’s own documentation. Auditing paid placements now is cheaper than appealing later.
