Announced 10 Oct 2026 · Sources checked
What the rank trackers found
Since mid-September, a growing share of the Google results pages collected by rank-tracking software have looked normal but been wrong. Nozzle, a rank-tracking company, published research on 10 October saying that 8.0 percent of the 27.5 million results pages it collected from 13 September onward came back “poisoned”, against 0.17 percent in the three weeks before. On the worst days, 27 to 29 September, it was about a quarter of all pages.
Keyword.com, a competitor, published its own check on 16 October using a definition adapted from Nozzle’s. It found that 13.1 percent of the pages in its study from 13 September to 14 October met that definition, compared with 0.3 percent in August, and that the share peaked at 34 percent on 30 September. The two companies use different customer datasets and different collection routes, yet describe the same pattern.
Nozzle attributes the change to Google’s response to AI scraping, the bulk collection of search results to feed AI answers, and says rank trackers are caught in it because they collect pages the same way. Google has not commented publicly on the findings, according to Search Engine Journal and other trade outlets that covered them. Keyword.com is careful to say it has no data of its own on why Google serves these pages.
What a poisoned results page looks like
A poisoned page is not an error or a CAPTCHA. It arrives complete, uses the normal template and stays on topic, but most of the sites that usually rank for the query are gone. Nozzle counts a page as poisoned when fewer than half of a keyword’s usual sites, the ones in its top 10 on at least 60 percent of collections from 20 August to 9 September, appear anywhere on it, and those sites then come back on a later check.
Nozzle’s report shows side-by-side examples. For the US desktop query “white label local seo”, the normal page on 4 October listed white-label SEO vendors. Three days later, all nine page-one sites had been replaced by a WordPress development tool, a Jackson Hole steakhouse, a directory of county health departments and USA.gov, with pagination stopping at page three. Nozzle reads that as Google interpreting “local” literally.
Across the dataset, the organic results on a poisoned page kept 20.7 percent of the previous page’s sites, against 88.5 percent normally. Other page features moved too: People also ask kept 46.9 percent of its questions instead of 87.4 percent, and the sources cited in AI Overviews kept 42.3 percent instead of 68.8 percent. One part of the page became more stable rather than less. Ads kept 83.7 percent of their advertisers, up from 70.6 percent.
Blocked first, then poisoned
Nozzle describes a sequence. On 13 September, its requests to Google started failing in large numbers, and for about two weeks fewer pages came through. From 20 September, the pages that did get through started coming back poisoned. Since 1 October the daily share has run at about 5 to 11 percent.
Derek Perkins of Nozzle told Search Engine Roundtable that Google appears to be using three tactics at once: very wide IP blocking aimed at large residential proxy networks, a sharp increase in CAPTCHAs, and injected bad data on pages it suspects are being scraped. The first two fail loudly, leaving gaps that a vendor can report. The third fails quietly.
That is why Nozzle calls it poisoning rather than blocking. A refused request tells a scraper to try again; a plausible but wrong page gets stored as real data. Nozzle also found that the route a request takes matters: between 30 September and 9 October, one of its collection routes saw 1 to 7 percent poisoned pages while another saw 9 to 30 percent.
Who sees it most
The effect is uneven. In Nozzle’s data, the United States ran at 8.5 percent and the United Kingdom lowest at 3.0 percent, with Germany, France, Spain and Italy all under 5 percent. Austria (18.8 percent), South Africa (18.4 percent), the UAE (14.4 percent) and New Zealand (13.8 percent) ran highest. Device type barely mattered: 7.7 percent on desktop, 8.3 percent on iPhone and 7.1 percent on Android.
Some searches are hit harder than others. “Near me” queries were poisoned 11.4 percent of the time against 7.8 percent for everything else, long-tail phrases more than popular ones, and brand searches least. Among US industries, legal queries ran highest at 15.9 percent since 13 September, peaking at 51.3 percent on 27 September, while pet queries were lowest at 4.3 percent.
Social sites and forums fill the vacated slots. Reddit appeared in the top 10 on 28.8 percent of poisoned pages against 20.8 percent of the same keywords’ other pages, with Facebook, LinkedIn, Instagram and Quora also gaining. YouTube, Yelp and MapQuest appeared less often. Keyword.com saw Reddit and YouTube about twice as often on poisoned pages, a difference it says proves nothing on its own.
Why it matters for AI search and SEO
Google’s results are raw material for much of the AI industry. Answer engines, research agents and grounding tools often lean on scraped search pages, which is why a growing number of companies now sell search and fetch APIs built for AI agents. If Google is deliberately serving degraded pages to suspected scrapers, any system built on unofficial Google scraping inherits plausible but wrong data, with nothing on the page to say so.
The fight over who may collect what from the web is playing out on several fronts. Meta rewrote its commercial terms after publisher backlash over crawling business websites, while Cloudflare and OpenAI reported results from a crawl-discovery pilot that helped ChatGPT search find new pages. Google, meanwhile, faces courts that hold it liable for what its AI answers say, as when an Aachen court barred AI Mode from repeating a false claim about a doctor.
For SEO teams, the practical risk is false alarms. Nozzle found a tracked site missing from the top 10 on 58.6 percent of poisoned pages, against 2.7 percent before. Keyword.com put the figure at 72 percent. A one-day crash in late September or October may be an artefact rather than a ranking change, and both firms recommend checking Google Search Console impressions and average position for the same query, country and device before acting.
Limitations
The evidence comes from two vendors that sell rank tracking, both of which have a commercial interest in explaining sudden drops their customers see. Neither can say what a real person searching on those days saw, and neither has a statement from Google. The motive, AI scraping, is Nozzle’s interpretation; Keyword.com explicitly declines to test it.
The definitions also lag. A page only counts as poisoned once its usual sites return, so the most recent days in both studies will rise as later checks arrive, and Keyword.com’s figures mostly describe daily-tracked keywords because weekly ones had too little baseline. Late September also overlapped a confirmed Google spam update, as Technical SEO News points out, which makes some genuine volatility hard to separate from injected pages.
What to watch: whether Google acknowledges the practice or changes its spam policies for automated queries, whether other trackers publish matching numbers, and whether AI search products that depend on scraped Google results show visible quality drops.
Common questions
What is a poisoned Google SERP?
A complete-looking Google results page, with no error or CAPTCHA, where most of the sites that usually rank for the query are missing and then return on a later check. Rank trackers record it as a ranking crash.
Does this affect what normal searchers see?
There is no evidence it does. Both studies measured pages collected by automated rank-tracking systems, and neither can show what a person searching on those days saw.
How can SEO teams tell a poisoned page from a real drop?
Check Google Search Console impressions and average position for the same query, country and device, and see whether familiar competitors vanished too. Keyword.com found 78% of poisoned pages recovered within three days.
Sources & further reading
How this story was made
Written by Kristian Kostov with AI assistance and checked against the linked sources. Company performance claims are attributed to the company. Analysis reflects AiLookout’s interpretation; we have not independently tested the products discussed. The cover places the companies’ own logos, taken from public-domain or openly licensed files credited in the caption, on a plain background. It is not a product photograph, a screenshot, or an endorsement.
Our editorial standards
