Since September 13, 2026, SEO tool vendors have described a Google that blocks their automated rank checks, then serves them normal-looking results pages whose results have been swapped. Out of 27.5 million pages collected, Nozzle counts 8.0% as poisoned since September 13, against 0.17% before, with a peak of a quarter from September 27 to 29; France is among the least affected countries, at 3.9%. DemandSphere measures around 11% of its daily checks in early October. These are vendor figures, which Google has not commented on publicly. For a small business, a September or October rankings curve should be checked in Search Console, which counts real impressions and clicks, and third-party tools are only useful for reading a trend over several weeks.
Contents
Since September 2026, rank tracking tools have been producing curves that several practitioners consider unreliable: on October 12, Search Engine Roundtable gathered the findings of the vendor Nozzle and messages from consultants whose rank checks no longer match the real figures of the sites they follow.1 These tools record a site's ranking by automatically loading Google's results pages, millions of times a day, and on October 13 PPC Land reconstructed the timeline of the obstacles Google has put in their way since the summer.2 This article goes over what the vendors measured, explains why a rankings curve from this fall may describe a page no searcher ever saw, and how to read your figures in the meantime.
01Blocked, then deceived: what the vendors measured
The available figures come from the tool vendors, who sell these rank checks and depend on them: PPC Land points this out about DemandSphere, whose every figure comes from its own systems.2 According to the same outlet, the two vendors that published measurements agree on the mechanism and, roughly, on the scale.2 The most detailed study is Nozzle's, published on October 10 with data up to October 9.3
Rank checks harder and harder since the summer
Since August 26, Google has been routing the links in its results through google.com/goto addresses. According to Derek Perkins, who runs Nozzle, finding the destination of every link in a five-page ranking now takes between 500 and 1,000 requests.2
Derek Perkins described three methods used by Google to Search Engine Roundtable: very broad IP blocking aimed at the large residential proxy networks, a much more frequent use of CAPTCHAs, and the injection of bad data into results pages deemed suspicious.1 At DemandSphere, the most aggressive blocking ever observed began around September 11, and on September 24 its founder Ray Grieselhuber described it as the new reality of the trade.2 Nozzle dates the two phases more precisely: requests sent to Google started failing en masse on September 13, and from September 20, those that got through came back poisoned.3 For two weeks, fewer pages could be collected.3
Complete pages, swapped results
Nozzle sums up the phenomenon this way: since September 2026, Google has been serving tracking tools complete, normal-looking pages whose results have been swapped, and nothing on the page gives it away.1 In its study, a page counts as poisoned when fewer than half of the keyword's usual sites appear on it and they come back on a later check.3 No error or CAPTCHA comes with it: the layout stays the same, but the organic results keep only 20.7% of the sites from the previous check, against 88.5% normally.3
| Nozzle measurement | Before (August 20 to September 9) | Since September 13 |
|---|---|---|
| Poisoned results pages | 0.17% | 8.0% (from 0 to 27% depending on the day) |
| Worst period | - | 24.9%, 26.7% and 27.0% from September 27 to 29 |
| Since October 1 | - | around 5 to 11% per day |
| Tracked site missing from the top 10 | 2.7% of checks | 58.6% of poisoned pages |
| Usual sites back within three days | - | 77.9% of poisoned pages |
The study covers 27.5 million results pages collected in 49 countries, on desktop, iPhone and Android, and more than 660,000 keywords had at least one poisoned page.3 On these pages, Reddit, Facebook, LinkedIn, Instagram and Quora appear more often in the top 10, YouTube, Yelp and MapQuest less often.3 DemandSphere, for its part, measures around 11% of erroneous daily results pages in its October 7 post, without describing its method, and warns that its figure can change quickly.2
France among the least affected countries
Nozzle publishes the share of poisoned pages country by country since September 13. Western Europe comes out lowest: 3.9% for France, 3.0% for the United Kingdom, 3.4% for Spain, 4.3% for Germany. Its neighbors are hit much harder, with 13.2% for Belgium, 12.8% for Switzerland and 18.8% for Austria, against 8.5% for the United States.3
The type of search matters too. “Near me” searches are poisoned at 11.4%, against 7.8% for the others, and phrases searched fewer than 100 times a month at 9.4%, against 6.1 to 7.1% above that.3 In the United States, legal pages reached 51.3% on September 27 and home services pages 48.4% on September 29.3 Finally, the share varies with the path by which the request reaches Google: on the same days, one of Nozzle's collection routes returned 1 to 7% poisoned pages, another 9 to 30%.3
What Google says about it
According to PPC Land, Google has not publicly commented on either figure, and none of the outside reports cites a statement from Google about the September changes.2 Its spam policies list machine-generated traffic among prohibited practices, meaning sending automated queries to Google, including scraping results to check rankings without express permission.4
Barry Schwartz, of Search Engine Roundtable, believes these measures target AI bots and the SerpApi service.1 Ray Grieselhuber thinks Google wants to stop AI labs from using its results pages to answer questions live, and that SEO tools are caught in the crossfire.2 Nozzle sees it as Google's response to the mass collection of its results to feed AI answers, which affects tracking tools because they collect pages the same way.3
02Why rankings curves are no longer reliable
A blocked tool and a deceived tool do not produce the same flaw. The first leaves a gap in its data, the second displays a plausible figure.2
A poisoned page goes unnoticed
PPC Land describes the difference: a blocked request produces a gap that the vendor sees and can flag, whereas a poisoned page looks normal and yields a confident figure.2 Nozzle puts the same idea this way: a refused request prompts the bot to retry, a poisoned request is recorded as real.3 A brand that seems to drop out of the top 10 on an important query, replaced by a forum thread, may be seeing a real ranking change, the effect of a Google update, or a page served only to a suspicious bot.2
Several vendors have published notices to their customers, including Sistrix, DataForSEO, SE Ranking, Advanced Web Ranking and Morningscore.1 Barry Schwartz fears that most tools will not quickly spot the injection of bad data and will believe they are fine while collecting poisoned pages.1
Gaps between tools, on sites whose figures are known
The consultant Glenn Gabe compared the tools on sites whose real figures he knows. On September 18, on a site that had fallen almost to zero and then fully recovered, Ahrefs showed the recovery, Semrush showed it late and Sistrix did not show it.1 On October 7, on a site hit on September 29 during the spam update, a drop confirmed by its owner, Ahrefs had seen the drop, Semrush very late, and Sistrix still did not show it.1
The consultant Lily Ray wrote on October 8 that tracking tools were, for now, too unreliable for her to share her observations.1
The signs of a poisoned page
Nozzle gives several clues for telling a poisoned page from a real ranking change. The result does not match the keyword's history: the useful comparison is with its usual sites from previous weeks, because the previous day's check may itself have been poisoned.3 The episode does not last: 80% affect only a single check, whereas a real ranking change stays.3
A last clue cited by Nozzle: if a site “disappears” from the tool while Search Console still shows impressions and clicks, nothing has happened to the site.3
A spam update in the same window
The September 2026 spam update started on September 24 and ended on October 8, after 13 days and 16 hours.2 During those two weeks, the volatility of a real update and the noise of blocked or poisoned collection look alike on a dashboard, PPC Land notes.2
03Reading your figures this fall: Search Console first
The Search Console performance report counts the times the site appeared in Google search results and the clicks searchers made on those results.5 Barry Schwartz reminds readers of this in his article: Google provides plenty of ranking data in Search Console, even if it does not cover every need SEOs have.1
Search Console as the reference
To judge traffic, the reference is the clicks and impressions curve of this report. Its average position is that of the site's highest-ranking result; in the table, it is calculated for the page or query of each row.5 When a third-party tool announces a drop, the first check is to open the same page in the report and look at its clicks over the same period: if clicks and impressions have not moved, the drop remains to be confirmed before acting.
The same rule applies to tracking rankings after a redesign: queries recorded before launch are compared page by page in Search Console. Checking in Google yourself settles nothing either, because results vary with the time, the place, the device and the recent history of the person searching.5
Third-party tools for the trend
Third-party tools keep two uses: following a trend over several weeks and watching competitor sites, which Search Console does not show. For its own report, Google recommends the weekly or monthly views, which smooth out daily fluctuations and bring out the overall rise or fall.5 A one-day movement in a third-party tool therefore justifies no decision: according to Nozzle, a tracked site that dropped out of the top 10 for a single day at the end of September, then came straight back, was probably hit by this phenomenon.3 For a site tracked on Google France, the share of poisoned pages measured by Nozzle is lower than elsewhere without being zero, and a site that also targets Belgium or Switzerland is more exposed.3
To measure the effect of a change on a few pages, comparing two periods of the same length, page by page, is done in Search Console, leaving out any window in which a Google update is rolling out.
Questions to ask your tool or your provider
DemandSphere has announced error rate and sampling rate indicators published alongside its rank checks, and acknowledges collecting some pages by sample instead of all of them every day.2 According to PPC Land, most vendors do not yet publish this kind of data about their collection.2
| Question | What it lets you check |
|---|---|
| Were the September and October rankings cross-checked with Search Console? | the report counts the site's real impressions and clicks |
| Does the tool publish an error or sampling rate? | the share of checks to treat with caution, day by day |
| Does the period compared overlap September 24 to October 8? | the spam update may explain a real movement |
| Does a drop in the tool show up in clicks? | without a drop in clicks or impressions, it remains to be confirmed |
If the drop shows up in Search Console clicks, it is real, and the investigation continues with the crawling and indexing checks of an audit.