Google has been quietly feeding rank trackers search results that are not real. According to a study published October 10 by SERP monitoring firm Nozzle, the company has since mid-September been serving complete, normal-looking search results pages whose underlying sites have been swapped out — no error, no captcha, no marker of any kind — to automated traffic it suspects of scraping. About 8% of the 17.8 million judged SERPs collected since September 13 came back poisoned, against a baseline of 0.17%. On the worst days, September 27 through 29, roughly a quarter of all pages served to these tools contained fabricated results. Google has not commented on the findings, and the company did not respond to press inquiries about the study.
The discovery matters well beyond the rank-tracking vendors it directly hits. A refused request tells a scraper to try again; a poisoned page gets stored as if it were a real ranking. That inversion is what makes this different from every previous round of Google's anti-scraping measures, and it forces anyone who consumes third-party ranking data — agencies, in-house teams, and the AI companies themselves — to rethink how much trust that data deserves.
A three-pronged squeeze, with poison as the new weapon
The timeline Nozzle reconstructed runs like this: on September 13, requests to Google from its collection infrastructure began failing in large numbers. The pages that still got through looked normal until September 20, when the first poisoned SERPs appeared. Poisoning peaked at 24.9% to 27.0% of daily SERPs between September 27 and 29, then settled at roughly 5% to 11% per day through early October.
Nozzle's CEO Derek Perkins described a three-part approach when he spoke with Search Engine Roundtable: wide IP blocking aimed at major residential proxy networks, a significant increase in captcha challenges, and the injection of bad data into SERPs served to requests that look suspicious. The first two are old tools pushed harder. The third is new, and it is the one that concerns him most — because a tool that fails to detect it will quietly report wrong positions to customers while believing everything is fine.
The context is Google's escalating fight against AI scraping. Since mid-2025, Google has tested "goto" redirect links to fingerprint scrapers, ramped up captcha rates, and — as of September 2025 — removed results beyond page 10 entirely. The presumed targets are AI companies harvesting search results to ground their models, and commercial scraping services like SerpApi. Rank trackers collect SERPs the same way those services do, so they are caught in the same net. Barry Schwartz's read is that SEO tools are collateral damage, not the target; Google has not confirmed any motive.
What a poisoned SERP actually looks like
The defining feature of a poisoned page is that nothing on it looks wrong. Nozzle compared the same keyword, device, and target location collected days apart. In one documented example, a desktop query for "white label local seo" in the United States returned a normal page of white-label SEO vendors on October 4. On October 7, the same query returned 26 results in which 9 of 9 previous page-one sites — specialist vendors — had been replaced by a WordPress development tool, a steakhouse in Jackson Hole, a directory of county health departments, and USA.gov. The HTML file size barely changed: 720 KB to 728 KB. A page-size check catches nothing.
Across the dataset, organic results on poisoned pages kept only 20.7% of the sites from the keyword's previous SERP, versus 88.5% normally. The vacancy gets filled mostly by social platforms and forums: Reddit appeared in the top 10 of 28.8% of poisoned SERPs versus 20.8% of the same keywords' other results, Facebook rose from 28.3% to 33.8%, LinkedIn and Instagram gained similarly, while YouTube, Yelp, and MapQuest lost ground. Pagination collapsed — on desktop keywords that normally paginate to page eight or beyond, half of poisoned SERPs stopped at page four or earlier — and AI Overviews appeared on only 34.9% of poisoned pages versus 54.2% of normal ones.
The pattern is not evenly distributed. Legal-industry SERPs in the US were poisoned at a 15.9% rate for the period and hit 51.3% on September 27. Austria ran at 18.8% for the period while the UK sat lowest among large markets at 3.0%. "Near me" queries were poisoned at 11.4% versus 7.8% for everything else, and low-volume long-tail phrases caught more poison than head terms. One of Nozzle's collection routes saw 9% to 30% poisoned SERPs on days when another route saw 1% to 7% — meaning the odds of receiving a fabricated page depend partly on which proxy path a request takes.
Why this cuts deeper than a bad data week
Rank tracking is infrastructure. Agencies budget, report, and get judged on position data; enterprise teams tie SEO outcomes to dashboards fed by these tools; and the tools themselves have spent the past month issuing status-page notices — Sistrix, DataForSEO, SE Ranking, Advanced Web Rankings, and Morningscore among them — while many others stayed quiet or mistook this for the earlier redirect changes. Lily Ray, Glenn Gabe, and other practitioners reported tool disagreements and unreliable numbers throughout late September and early October. Dr. Pete Meyers of MozCast went as far as documenting and handing off the volatility project he is best known for.
The second-order effect is on AI data quality itself. AI Overviews remain prominent in Google's results, and the sources they cite are drawn from the same index that poisoning corrupts. Nozzle found AI Overview sources kept only 42.3% of their previous citations on poisoned pages. Any dataset built from scraped Google results — including training and grounding pipelines at AI companies — now carries a contamination risk that has no visible signature. In effect, Google has demonstrated that it can degrade the value of search data at the source, selectively, without detection by the recipient.
There is also a trust dimension. The SEO industry's measurement layer has always rested on an unspoken bargain: tools scrape, Google tolerates it within limits, and the data approximates reality. Google has now shown it is willing and able to falsify the data itself — not block access, but corrupt it — and to do so in a way indistinguishable from a genuine ranking change unless you compare against weeks of keyword history. That bargain is renegotiated, and the industry does not yet know the new terms.
What practitioners should actually do
First, treat any single-day ranking crash from late September onward as unverified until checked. Nozzle found 80% of poisoned spells lasted a single fetch, and 77.9% of affected keywords had their usual sites back within three days. A drop that reverses itself within days is more likely bad data than a ranking change.
Second, use Search Console as the tiebreaker. It is Google's own measured data on impressions and clicks, and it is the one source Google cannot poison without poisoning itself. If a tracker reports your site vanished but Search Console still shows impressions and clicks, nothing happened. For AI-surface visibility, the same logic applies to server-side logs, which record the visit regardless of how any tool classified it.
Third, compare against the keyword's typical page-one set over several weeks, not yesterday's snapshot — yesterday may have been poisoned too. A real ranking change persists; a poisoned page is thinner, with more social results, more image packs, shallower pagination, and fewer AI Overviews. No single marker is reliable on its own, but the pattern is detectable.
Fourth, ask your rank-tracking vendor specifically whether they detect and filter poisoned SERPs, and how. The goto redirects were the easy part; injected bad data is the part most tools have not publicly addressed. Where a vendor cannot answer, weight their data accordingly. And for client reporting, prefer trend direction over point-in-time positions, widen the date windows, and annotate the September 24 to October 8 window — which overlaps Google's officially confirmed September 2026 spam update — as a period of degraded third-party data.
Finally, treat the episode as a strategic signal. If Google is willing to serve fabricated results to suspected scrapers, the value of any scraping-derived metric — rank, visibility share, even competitor monitoring — carries a new class of risk. The vendors who survive this will be the ones who can prove which of their data points came back clean.
Google has not said whether poisoning will continue, scale up, or become standard practice. The industry now knows it is technically possible, already deployed at 8% scale, and nearly invisible when it happens. That is the part no one can un-see.
Sources
- Nozzle Research — Poisoned Google SERPs (October 10, 2026, data through October 9): https://nozzle.io/research/poisoned-google-serps/
- Search Engine Roundtable — Third-Party Google Rank Tracking Struggling Now (October 12, 2026), including Derek Perkins' three-pronged description: https://www.seroundtable.com/third-party-google-rank-tracking-broken-42272.html
- Mashable — Google poisoning search results for AI scrapers, rank trackers, new report claims (October 14, 2026): https://mashable.com/tech/google-poisoning-search-results-for-ai-scrapers-rank-trackers-report
- TrueRanker — Google is serving rank trackers pages that look real and aren't (October 13, updated October 16, 2026): https://trueranker.com/blog/google-poisoned-serps/
- Google Search Status Dashboard — September 2026 spam update (September 24, 2026, 13 days): https://status.search.google.com/
