A rank tracker checked 27.5 million Google results pages. It says 8% came back swapped.

Take one search, “white label local seo,” on a desktop in the United States. Nozzle, a rank tracking company, collected Google’s first page for it and collected it again three days later. The second page had the same template, no error, and no captcha. All nine sites on it had been replaced. [1]

A rank report reads that as a crash. Nozzle reads it as a decoy, and on October 12 it published the data to argue the point: 27.5 million results pages, collected for its customers across 49 countries from August 20 to October 9, with the method written out and the daily numbers offered for download. [1][7] It is the most detailed published account we could find of something several vendors have been saying for three weeks, and it changes how you should read any ranking chart that covers late September.

What Nozzle counted

The definition matters, because it decides what the 8% can and cannot tell you. For each keyword, Nozzle took the sites that were in the top 10 on at least 60% of its pages between August 20 and September 9. A later page is poisoned when fewer than half of those usual sites appear anywhere on it and they come back on a later page. [1]

Two things follow. A page only counts once the usual sites return, so the measure misses pages that stay wrong. And it would catch a real ranking change that reversed within days. The baseline controls for that: before September 13, only 0.17% of pages met the definition. [1]

The sequence is the part Nozzle leans on. On September 13, requests to Google began failing in large numbers. Its daily count of collected pages fell from 1,098,064 on September 12 to 109,952 on September 17, about a tenth. The pages that did arrive were normal until September 20. Then the poisoned ones begin. [1]

Those are Nozzle’s daily shares for selected days: 1.9% on September 20, 11.1% on September 25, 27.0% on September 29. [1]

One line in the study deserves more attention. Over ten days from September 30, one of the network routes Nozzle collects through returned 1 to 7% poisoned pages and another returned 9 to 30%. [1] The rate depends on how the page was requested, not on Google’s index, so every vendor has its own number. DemandSphere, another tracker, puts its own at about 11% of daily pages and adds: “Don’t take that 11% number as the final number because things could change rapidly.” [2] The two figures are different measurements and should not be averaged.

Why a wrong page is worse than a blocked one

Nozzle’s sentence for it: “A refused request tells a scraper to try again, while a poisoned one gets stored as real.” [1] A blocked request leaves a gap you can see. A poisoned page leaves a number you believe.

Most of the usual sites are gone, and social platforms and forums take more of the top 10: Reddit rises from 20.8% of pages to 28.8% and Facebook from 28.3% to 33.8%. [1] The parts of the page move by very different amounts.

The last two rows are Nozzle’s sharpest observation: “Only the parts advertisers pay for hold steady.” [1]

What it does to a report

For keywords where a customer’s own site was one of the usual sites, that site was out of the top 10 on 58.6% of poisoned pages. In the baseline it was 2.7%. [1] On a chart that is a cliff, then a recovery nobody did anything to earn.

Most of it is brief. Four in five spells last a single collected page, and 77.9% of poisoned pages had the usual sites back within three days. [1] It is also uneven. Searches with “near me” were hit 11.4% of the time against 7.8% for the rest, and US legal keywords reached 51.3% on September 27. The United States ran at 8.5% overall, the United Kingdom at 3.0%. [1]

There is a second casualty. On poisoned pages an AI Overview appeared 34.9% of the time, against 54.2% in the baseline. [1] Charts of how often AI Overviews appear are built the same way, by loading results pages automatically. DemandSphere’s series for branded keywords climbs from 26.12% on September 1 to 82.06% on September 29, and on the way reads 57.43% on September 18 and 31.15% the next day. [8] The post does not say whether the collection problems touched those counts. Until a vendor says, treat September figures on AI Overviews as a trend inside one panel, as we argued for click data in our piece on AI Overviews traffic.

What Google has said, and what it has not

Google has not commented on altered results. The only statement on the record is from August, when it confirmed it was wrapping result links in google.com/goto redirects. As quoted by DemandSphere, a spokesperson said: “We have a long history of deploying technical measures against evolving forms of abuse, and we regularly take steps to protect our services and users.” [3]

The vendors agree on who the target is, and it is not them. DemandSphere’s Ray Grieselhuber writes that “Google does not want the AI labs to use their SERPs as part of their live retrieval process to answer questions.” [9] Search Engine Roundtable, which has reported the story since September 18, assumes the aim is AI bots and the scraping company SerpApi, not SEO tools. [7][10] These are readings of Google’s motive by people Google is blocking. They may be right. They are not confirmation.

What is on the record is the escalation, and a court order fills in the early part.

The order states Google’s grievance in Google’s terms: automated queries impose “a deadweight loss,” because it earns nothing from them. [11] Whether the ruling led to what followed, the sequence alone does not show.

How to check a drop before you report it

Nozzle’s own advice ends with the simplest test: if a site seems to disappear and Search Console still shows impressions and clicks, nothing happened to the site. [1] That works because of how Search Console counts. Its help page says: “A link must get an impression for its position to be recorded.” [4]

It is a check, not a replacement. Search Console averages the topmost position of your property across the searches where you appeared, and Google warns that for a specific search “your position might be different than the average because of many variables, such as your search history, location, and so on.” [4] Compare the direction of clicks and impressions. Do not try to match the number.

And the check has its own bad days. On October 11 the Performance report was more than 64 hours behind, according to Search Engine Roundtable, and normal again the next morning. [5] Google’s log of Search Console data anomalies had no entry for it when we read it on October 13. [12]

• Mark September 13 to October 9 on every ranking chart you publish, and say why.
• For any keyword that fell, look at the next three days of collection. A fall that reversed on its own belongs in a footnote.
• Compare a suspect page with the keyword’s usual page-one sites over several weeks, not with yesterday, because yesterday may have been poisoned too.
• Open Search Console for that query and page. If clicks and impressions held, the site did not move.
• If the fall stayed past October 8, treat it as possibly real and read it against the spam update, the same way we argued in our note on reacting to one.
• Ask your vendor for its poisoned or spurious rate and its sampling rate by day. DemandSphere says it is adding both to its product. [13]

The honest summary is narrow. Two vendors have published numbers from their own traffic, one of them with its method and data, and Google has said nothing. What you can act on is smaller than the headline and more useful: a ranking that fell for a day in late September and came back is more likely a page Google served to a machine than a judgment it made about your site.

Sources

1. Nozzle · Poisoned Google SERPsA rank tracking vendor’s study of its own collection, August 20 to October 9, 2026, with method and downloadable daily data. Not independently verified.
2. DemandSphere · Spurious data in Google SERPsVendor post, October 7, 2026. No method published for the 11% figure.
3. DemandSphere · An update from DemandSphere on Google’s goto redirectsVendor post, August 27, 2026. Source for the Google spokesperson’s statement as quoted.
4. Google Search Console Help · What are impressions, position, and clicks?Primary.
5. Search Engine Roundtable · Google Search Console Performance Report Extremely DelayedTrade press, October 11, 2026, updated October 12.
6. Google Search Status Dashboard · Summary of incidentsPrimary. Read October 13, 2026.
7. Search Engine Roundtable · Third-Party Google Rank Tracking Struggling NowTrade press, October 12, 2026.
8. DemandSphere · AI Overviews on branded queries tripled in SeptemberVendor post, October 1, 2026, with daily figures.
9. DemandSphere · Google’s Summer of LoveVendor post, September 24, 2026.
10. Search Engine Roundtable · Google May Be More Successful In Blocking Scrapers & Tracking ToolsTrade press, September 18, 2026.
11. US District Court, Northern District of California · Order granting motion to dismiss, Google LLC v. SerpApi, LLC, No. 4:25-cv-10826-YGRPrimary. Filed July 20, 2026; 21 pages. Via CourtListener. The background section recites Google’s allegations.
12. Google Search Console Help · Data anomalies in Search ConsolePrimary. Read October 13, 2026.
13. DemandSphere · Understanding Observability in the world of AI search dataVendor post, October 6, 2026.

MS

Editor-in-chief

Mikołaj Salecki

Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.

LinkedIn ↗Facebook ↗salecki.digital ↗

More articles →