If your search report showed a cliff in late September or early October, do not act on it yet. There is now a documented reason the number may be fiction.
What did the research find?
Nozzle, an enterprise rank-tracking firm, published an analysis called “Poisoned Google SERPs” on October 10, 2026, covering 27.5 million Google results pages across 49 countries collected between August 20 and October 9.
Its central finding: from September 13 onward, 8.0 percent of the result pages its systems retrieved were not the pages a real user would see. The baseline before that was 0.17 percent. Between September 27 and 29, it reached roughly a quarter of all pages checked. Since October 1, the daily rate has run between 5 and 11 percent.
These pages do not look broken. They look like ordinary search results containing the wrong websites. On a second check, a normal page repeated the same sites 88.5 percent of the time. A poisoned page kept 20.7 percent. On poisoned pages, the tracked client site was missing from the top ten 58.6 percent of the time, against 2.7 percent normally.
The sequence Nozzle describes is that requests to Google started failing on September 13, and from September 20 the requests that did get through came back with wrong results instead of an error. Most of it is temporary: 77.9 percent of poisoned pages had their usual sites back within three days.
Roger Montti covered the analysis at Search Engine Journal on October 14, 2026, adding detail from the same dataset. AI Overviews showed on 34.9 percent of poisoned pages against 54.2 percent at baseline. Reddit, Facebook, Instagram, LinkedIn, and Quora appeared more often. MapQuest, Yelp, and YouTube appeared less. Paid and shopping results stayed stable. Google has made no public statement about any of it.
Is this one vendor’s problem or everyone’s?
It is one vendor’s measurement, and the mechanism would affect every tool that works the same way.
That is the honest framing. Nozzle sells rank tracking, which gives it both the data to see this and an interest in explaining a bad month. But the explanation is mechanically plausible: these tools do not ask Google for your ranking; they load a search page the way a browser does and read it. Anything that degrades automated page loads degrades every tool built that way, which is most of them.
What should you do about a sudden drop?
Check it against Google Search Console before you change anything.
Search Console reports impressions, clicks, and average position from Google’s own logs for your own property. It is not scraped. If a scraped report shows you falling out of the top ten in late September and Search Console shows impressions and clicks flat, the ranking did not move. If both show the decline, something real happened, and it is worth the diagnosis.
Then slow down the reaction cycle. This episode punishes anyone running weekly keyword positions as a management metric, because five to eleven percent of those readings have been wrong for a month and the errors clear within three days. A thirty-day trend survives that noise. A Monday-morning position check does not.
And look at what the drop costs you in the only currency that matters. Positions are a proxy. Inquiries, calls, and orders are the thing. If your inquiry volume is steady, you don’t have an emergency, regardless of what the dashboard says. Sorting out which of those numbers deserves a decision is the first thing worth fixing in a search program — and the reason Incognito Worldwide’s search work runs on a ninety-day minimum rather than a weekly report card.
The wider point
Every measurement system eventually gets measured by something other than the thing it claims to measure. A rank tracker is a model of search, and in September that model quietly stopped matching search for about one page in twelve.
Knowing which of your numbers are observations and which are estimates is most of the discipline in marketing measurement. The Art of Marketing spends a chapter on it, and this month is a good argument for why.