Poisoned Google SERPs

A full page, with the wrong sites

Since September 2026, Google has been answering some automated Search requests with a normal-looking results page whose organic results have been swapped out. No error. No captcha. Nothing in the HTML marks the page. The keyword's own history is what gives it away.

Nozzle published the measurements on October 10, 2026: 27.5 million SERPs across 49 countries, collected through October 9. Derek Perkins calls them poisoned SERPs. Search Engine Journal, Search Engine Roundtable, and Mashable covered the report over the following days. Google has not said anything publicly.

A SERP counts as poisoned when fewer than half of the keyword's usual sites appear anywhere on it, and those sites come back on a later SERP. Usual sites are the ones in the top 10 on at least 60% of that keyword's SERPs from August 20 to September 9.

What the numbers say

Before September 13, that definition matched 0.17% of SERPs. From September 13 through October 9, it matched 8.0%. More than 660,000 keywords were affected. Nozzle only marks a SERP as poisoned once the usual sites come back, so the last days of that window are still filling in.

The order of events matters. Requests to Google started failing in large numbers on September 13. The pages that still came back stayed normal until September 20. Poisoned SERPs then climbed, and from September 27 to 29 they were about a quarter of what Nozzle collected: 24.9%, 26.7%, and 27.0%. Since October 1 they have run at about 5 to 11% a day.

On a normal repeat of the same keyword, organic results keep 88.5% of the previous SERP's sites. On a poisoned SERP they keep 20.7%. Where the tracked site is one of the usual sites, it is missing from the top 10 on 58.6% of poisoned SERPs, against 2.7% in the baseline.

Most of it is short. 80% of poisoned spells are a single SERP, 15% last two, and 5% last three or more. 77.9% have the usual sites back within three days. A one-day disappearance in late September is more likely this than a ranking change.

What shows up instead

The page is not empty. Social sites and forums take the slots. On the same keywords, Reddit is in the top 10 on 28.8% of poisoned SERPs, against 20.8% of the others. Facebook goes from 28.3% to 33.8%. LinkedIn, Instagram, and Quora move the same way. YouTube, Yelp, and MapQuest show up less often.

Nozzle's example is the US desktop query "white label local seo". On October 4 the page lists white-label SEO vendors: seohive.co, synup.com, brightlocal.com, and the rest. On October 7 the same query returns a WordPress tool, a Jackson Hole steakhouse, a directory of county health departments, and USA.gov. All 9 page-1 sites are gone. Pagination drops from 10 pages to 3.

The rest of the page thins out in the same direction. AI Overviews appear on 34.9% of poisoned SERPs, against 54.2% in the baseline. "People also search for" appears on 52.7%, against 87.2%. Image packs nearly double, 44.2% against 23.0%. Forum blocks rise from 20.5% to 27.0%. Shopping barely moves: a poisoned SERP keeps 92.2% of the previous page's shopping items, against 92.8% normally. Ads, when they are on the page, keep more of the previous SERP's advertisers than a normal page does (83.7% against 70.6%).

Device barely changes the rate: 7.7% on desktop, 8.3% on iPhone, 7.1% on Android. Country does. Austria 18.8%, South Africa 18.4%, the UAE 14.4%. The United States is at 8.5%. Germany 4.3%, France 3.9%, Spain 3.4%, the United Kingdom 3.0%. "Near me" queries run higher, 11.4% against 7.8%. The most navigational queries run lower, at 5.3%. In the US, legal was the hardest-hit industry: 15.9% of SERPs since September 13, and 51.3% on September 27.

The route into Google matters as much as the query. On the same days, September 30 to October 9, one of Nozzle's collection routes returned 1 to 7% poisoned SERPs and another returned 9 to 30%.

A bad page is worse than a block

Nozzle's reading, which Barry Schwartz reported after talking to Perkins again: Google is not only refusing suspect requests. Some of the requests that get through come back with the results swapped.

A refusal is obvious. The client retries, or it logs an error. A poisoned page gets stored as the SERP. The rank report then shows a crash that did not happen to the site.

Perkins described three things at once: wide IP blocks aimed at large residential proxy networks, more captchas, and bad data injected into suspect SERPs. Schwartz's view, shared by Mashable, is that the target is AI scrapers collecting Google results to ground answers, and the SERP APIs Google has been fighting in court. Rank trackers collect the same pages, so they get the same pages back. Search Engine Journal adds that the same swap would also muddy anyone trying to reverse-engineer rankings from scraped SERPs.

The /goto redirects from this summer are a separate change, and an easier one to handle. We wrote that up in google.com/goto: Google's anti-scraping update.

It follows the other changes from the past year: &num=100 removed in September 2025, the SearchGuard lawsuit in December 2025, /goto in August 2026. Poisoned pages are the first of these where the failure looks like a 200 with the wrong sites.

How to tell it from a ranking drop

The page is complete. Same template, no error, no captcha.
The result set does not match the keyword's usual page-1 sites over the previous weeks. Yesterday alone is a weak comparison, because yesterday may have been poisoned too.
It does not last. Four in five spells are a single collection. A real ranking change stays.
Search Console still shows impressions and clicks. If the site vanished in a tracker and Search Console did not move, the site did not vanish.

If you read rankings from an API

A 200 does not prove you received the SERP a person would see. Keep each keyword's usual sites, and flag a page that drops most of them. If the sites are back within three days and Search Console did not move, do not ship that day as a ranking change.

The Google Search API returns the results from the page Google served, with /goto links already resolved to the destination URL. If you diff rankings day to day, compare against the keyword's usual sites, not only against yesterday.