Rank trackers are getting 'poisoned' Google results, Nozzle data shows

Since September 2026, Google has been answering a share of rank-tracker requests with complete, normal-looking results pages whose listings have been swapped out, according to research published by rank-tracking vendor Nozzle on 10 October 2026. Across 27.5 million results pages in 49 countries, 8.0% came back "poisoned" from 13 September onward, against 0.17% in the weeks before, and on the worst days, 27 to 29 September, it was about a quarter of all pages collected. Google has not commented on the findings.

For anyone who reports rankings to a boss or a client, the practical point is simple: a keyword that crashed out of the top 10 for a day in late September or early October and then came back may never have dropped for real searchers at all. The rest of this article covers what Nozzle measured, how it defines a poisoned page, what is new compared with the earlier goto-redirect problem, and how to check your own numbers.

What Nozzle found

Nozzle's study, titled Poisoned Google SERPs, draws on a representative sample of the results pages the company collected for its customers between 20 August and 9 October 2026. The sample covers desktop, iPhone and Android, and location targeting from whole countries down to postal codes. Nozzle says more than 660,000 keywords had at least one poisoned page, and four in five of the datasets in the sample were affected at least once.

The sequence Nozzle describes has two stages. On 13 September 2026, requests to Google started failing in large numbers, and for about two weeks fewer pages came through. The pages that did get through were normal until 20 September, when poisoned pages first appeared. The share climbed to 11.1% on 25 September, 18.4% on 26 September, then 24.9%, 26.7% and 27.0% on 27, 28 and 29 September. Since 1 October it has run at roughly 5 to 11% a day, with 4.2% on 9 October, the last day in the data. Nozzle notes that the most recent days are still filling in, because a page only counts as poisoned once the keyword's usual sites return.

The key numbers from the study are below.

How a "poisoned" page is defined

Nozzle does not call a page poisoned just because rankings moved. It first builds a baseline for each keyword: the sites that appeared in that keyword's top 10 on at least 60% of its pages between 20 August and 9 September, for keywords with at least five pages in that window. A later page is counted as poisoned when fewer than half of those usual sites appear anywhere on it, on any results page, and at least 80% of them are back on a later page within 14 days (or most of the sites missing from the previous page are back within three days).

That return condition matters. A real ranking change tends to stick. A poisoned page is a one-off: Nozzle found that 80% of poisoning spells last a single page, 15% last two and 5% last three or more. The next page has the usual sites back 55.9% of the time, and the median spell is over in two days.

The study also explains why Nozzle uses the word "poisoning" rather than "blocking". A blocked request fails visibly, so a scraper knows to try again. A poisoned request succeeds. Google returns a full page, with no error and no captcha, built on the same template and the same front-end build as a normal page, and Nozzle says nothing in the HTML marks it. The tool stores that page as real data.

What replaces the usual results

On poisoned pages, social platforms and forums show up more often. Comparing each keyword's poisoned pages with its own other pages since 13 September, Nozzle found Reddit in the top 10 on 28.8% of poisoned pages against 20.8% of the others, Facebook on 33.8% against 28.3%, and LinkedIn on 9.3% against 6.2%. Instagram, Quora, Etsy, JustAnswer and TikTok also appeared more often. YouTube, Yelp and MapQuest appeared less often.

Nozzle's worked example is the US desktop query "white label local seo". On 4 October the first page listed white-label SEO vendors. On 7 October the same query returned a WordPress development tool, a steakhouse, a directory of county health departments and USA.gov. All nine first-page sites from 4 October were gone, and pagination had shrunk from 10 pages to 3. Nozzle's reading is that Google took the word "local" literally.

The changes go beyond the organic list. On poisoned pages, the "People also search for" box kept 30.7% of its searches from the previous page, against 87.9% normally, and AI Overview sources kept 42.3% against 68.8%. AI Overviews appeared on 34.9% of poisoned pages against 54.2% in the baseline, while image packs nearly doubled. The paid parts held steady: shopping results kept 92.2% against 92.8%, and ads kept 83.7% against 70.6%.

Where and on which searches it hit hardest

Nozzle broke the results out by market. Of the larger markets, Australia (11.5%), India (10.9%) and Canada (10.5%) ran highest, the United States sat at 8.5%, Germany, France and Spain ran at 3.4 to 4.3%, and the United Kingdom was lowest at 3.0%. Device made little difference: 7.7% on desktop, 8.3% on iPhone and 7.1% on Android.

The study also found that the route a request takes to Google matters. Between 30 September and 9 October, one of the routes Nozzle collects through returned 1 to 7% poisoned pages and another returned 9 to 30% on the same days.

By search type, "near me" searches were hit more (11.4% against 7.8% for everything else), long-tail phrases searched fewer than 100 times a month more than head terms (9.4% against 6.1 to 7.1%), and brand searches least (5.3% for the most navigational phrases). Local targeting on its own made no difference once Nozzle compared the same phrase tracked nationally and locally on the same device: 7.5% and 7.4% across 21,508 pairs.

In the US data, every industry peaked between 27 and 29 September. Legal searches had the highest share since 13 September at 15.9%, reaching 51.3% on 27 September. Home services followed at 12.7%. Pets were lowest at 4.3%.

What is new since the goto-redirect story

Relevant Audience reported earlier on Google's goto links, under the headline "Google's goto links push rank trackers to 500 to 1,000 requests per query". That change wrapped organic result links in google.com/goto redirects, and the problem it created for tools was cost: more requests to resolve each link to its real destination. The data itself was still correct once a tool paid that cost.

The new phase is different in kind. The question is no longer how expensive it is to collect a page, but whether the collected page is true. Search Engine Roundtable reported on 12 October 2026 that Nozzle's Derek Perkins described a three-part approach from Google: very wide IP blocking aimed at major residential proxy networks, a large increase in captchas, and injecting bad data into suspect results pages. Search Engine Roundtable's own view was that the goto redirects are the easy part for tools to handle, and that the third tactic is the most worrying because tools may not notice it.

Nozzle is not the only vendor reporting it. PPC Land reported on 13 October 2026 that DemandSphere, another search data company, says Google has started returning spurious results when it detects suspected bot traffic, affecting about 11% of its daily results pages on current measurements. DemandSphere is adding reporting on error rates and sampling rates next to its ranking data. Search Engine Roundtable named Sistrix, DataForSEO, SE Ranking, Advanced Web Ranking and Morningscore among the tools that commented or posted notices about data issues, a list PPC Land repeated in its own summary.

The explanations on offer, and what Google has not said

Google has not commented publicly on the September changes or on either vendor's figures. Every explanation in circulation comes from the vendors and the trade press. Nozzle states that this is Google's response to AI scraping and grounding, meaning the bulk collection of its results to feed AI answers, and that rank trackers are caught in it because they collect pages the same way. DemandSphere's founder, quoted by PPC Land, made a similar argument. Search Engine Roundtable described it as an assumption that the target is AI bots and the scraping service SerpApi rather than SEO tools. None of these is a statement from Google, and this article does not treat it as one.

PPC Land also pointed out an overlap that makes this period harder to read. Google's September 2026 spam update ran from 24 September to 8 October, dates that match Google's Search Status Dashboard. Genuine ranking volatility from that update and noise from poisoned collection would look alike on a dashboard for those two weeks.

How to tell poisoned data from a real ranking drop

Nozzle lists five signs, and they translate into a short checklist for anyone reviewing September and October numbers.

• Compare against the keyword's history, not just yesterday. Yesterday's page may have been poisoned too. Nozzle recommends comparing against the keyword's usual first-page sites over the previous weeks.
• Wait before reacting. Most poisoning spells are a single page and 77.9% resolve within three days. A real ranking change stays.
• Look at the shape of the page. Poisoned pages are thinner on average: fewer AI Overviews and "People also search for" boxes, more image packs and social sites, shallower pagination. Nozzle warns that none of these signs is reliable on its own.
• Check Search Console. If a site "disappears" in a tracker but Search Console still shows impressions and clicks for the same query and dates, Nozzle's verdict is that nothing happened to the site.
• Ask your vendor what they are doing about it. DemandSphere has said it is publishing error and sampling rates. Most vendors do not yet publish comparable figures, according to PPC Land.

For reporting, that means a client or management report for September and October 2026 that shows a sudden top-10 loss should carry a note on data quality, or be cross-checked against Search Console, before anyone acts on it. A technical SEO audit that relies on third-party position data for this window should do the same.

What this means for Thai marketers

Thailand is in Nozzle's country table. Since 13 September, 11.8% of the Thai results pages Nozzle could judge were poisoned, across 15,097 judged pages. That is above the 8.0% overall figure and above the US rate of 8.5%, though the Thai sample is small compared with the 10.9 million judged US pages. Nearby markets in the same table include Singapore at 11.3%, Malaysia at 8.2%, Indonesia at 7.2% and Vietnam at 5.9%. Nozzle does not break these markets down further, and it does not say whether Thai-language queries behave differently from English ones.

The practical steps are the same as anywhere. Treat one-day drops between 20 September and early October with suspicion, confirm them in Search Console before changing pages or reporting a loss, and ask whichever rank tracker you use how it is detecting swapped pages. For teams running SEO in Thailand on top of third-party trackers, it is worth keeping Search Console exports for the same period, so that any later questions about September and October performance can be answered from Google's own data.

FAQ

What is a poisoned Google results page?

A poisoned results page is a complete, normal-looking Google page whose listings have been swapped out, as defined in Nozzle's 10 October 2026 study. Nozzle counts a page as poisoned when fewer than half of a keyword's usual top-10 sites appear anywhere on it and those sites return on a later page.

How common is it?

Nozzle measured 8.0% of results pages as poisoned between 13 September and 9 October 2026, against 0.17% before. The peak was 27.0% on 29 September, and the daily rate has run at about 5 to 11% since 1 October. DemandSphere separately reports about 11% of its daily pages affected, using its own undisclosed method.

Has Google confirmed it is doing this?

No, Google has not commented publicly. The view that Google is targeting AI scraping comes from Nozzle, DemandSphere and Search Engine Roundtable, not from Google.

Does it affect what real searchers see?

The sources do not say that real searchers see swapped results. Nozzle's study is about pages served to automated collection, and its advice is that if Search Console still shows impressions and clicks, nothing happened to the site.

Is Thailand affected?

Yes, Nozzle's data shows 11.8% of judged Thai results pages poisoned since 13 September 2026, across 15,097 pages. That sample is much smaller than the US one, so the Thai figure is less precise.

The bottom line

Rank tracking has moved from a cost problem to a data-quality problem. Since late September 2026, some of the numbers coming out of third-party trackers describe pages whose usual results, by Nozzle's measure, had been swapped out. Until vendors publish error rates as a matter of course, the safest habit is to treat Search Console as the reference for any ranking change and to read tracker data as a sample with known noise. If you want a second pair of eyes on how your rankings and reporting hold up through this period, the Relevant Audience SEO team can help.