---
title: "Poisoned SERPs: Google is corrupting rank tracker data"
description: "Google is feeding rank trackers swapped search results. I look at data from Nozzle, AccuRanker and Ahrefs and show how to spot a fake ranking drop."
date: 2026-10-24
category: SEO
tags: ["Rank tracking", "Google", "Search Console", "SEO tools", "SERP"]
url: https://uper.pl/en/blog/google-poisoned-serps-rank-trackers/
---

# Google feeds rank trackers swapped results How to tell a fake ranking drop from a real one

**Since the second half of September, Google has been answering some automated queries with a complete, normal-looking results page on which most of the organic results have been swapped out.** Rank trackers store that page as real, so reports show drops that no ordinary Google user ever saw. The industry calls this "poisoned SERPs".

Google has not confirmed any of it. Every figure I quote comes from companies that sell rank tracking or SERP data, and none of them has been independently verified. Below I put together what these companies measured, where their numbers disagree, what it means for SEO reporting and how I check whether a drop in a tool is real. At the end I show data from my own site, catalog.polo.blue: Ahrefs, Semrush, Google Search Console and Bing Webmaster Tools for the same period, with screenshots.

## What a poisoned SERP is

A poisoned SERP is a Google results page that looks perfectly ordinary: the same template, ads, sometimes an AI Overview. There is no error message and no CAPTCHA. The only difference is that most of the organic results are different pages from the ones that usually rank for that query.

Why is that worse than a plain block? A refused request leaves a gap in the data that the tool can see and retry. A poisoned page looks like a valid answer, so it goes into the database as a real result. That is how Nozzle, the company that first put numbers on the problem, explains it.

An example from the Nozzle study: the query "white label local seo" (desktop, US). On October 4 the first page listed nine white-label SEO providers. Three days later none of them appeared anywhere in the results. Their places went to, among others, a WordPress tool, a restaurant, a directory of local health departments and USA.gov, pages that only matched the word "local". Pagination shrank from 10 pages to 3.

The vendors describe three variants:

1. **Swapped results**: the page answers the query, but the usual domains are almost gone (Nozzle, AccuRanker).
2. **An answer to one word of the query**: Google ignores the rest of the phrase, as in the example above (Nozzle; according to Nozzle, SerpApi has tracked this fault since May).
3. **Truncated pages and the wrong location**: results stop after one or two pages, and local results list businesses from another city even though the page footer shows the correct location (AccuRanker).

## What Nozzle measured

Nozzle's study "Poisoned Google SERPs" (by Derek Perkins) was published on October 10 and has been updated several times. I use the version of October 21, with data up to that day.

**Method.** For each keyword, Nozzle defined its "usual sites": those in the top 10 in at least 60% of measurements between August 20 and September 9. A result counts as poisoned when fewer than half of them appear on the page. The sample is 29.9 million judged SERPs since September 13, from 53 countries, on desktop, iPhone and Android. These are keywords tracked by Nozzle's customers, not a random sample of the whole search engine; more than half of the measurements are from the US.

**Timeline according to Nozzle:**

- September 13: requests to Google start failing in large numbers,
- September 20: the requests that get through start coming back with swapped results,
- September 27–30: the peak,
- October 14–15: it eases to 4–6% of measurements (counting page 1 or 2),
- from October 17 it grows again; on October 21, 35.7% of page 1–2 measurements were poisoned.

**Key figures:**

| Metric | Before Sep 13 | Since Sep 13 |
|---|---|---|
| Page 1 poisoned | 0.17% | 8.7% |
| Page 1 or 2 poisoned | 0.28% | 22.6% (peak 43.5% on Sep 30) |
| Sites kept from the previous measurement | 88.1% | 22.3% on poisoned pages |
| Customer's tracked site out of the top 10 | 2.7% | 60.8% on poisoned pages |

Page 2 is poisoned one and a half times as often as page 1, and in 58.4% of those cases page 1 looks normal. That is particularly bad news if you track positions 11–20.

### Why articles quote 8.0%, 8.4% or 22.6%

These figures come from successive versions of the same study. Search Engine Roundtable and PPC Land quoted the version with data to October 9 (8.0%), Coywolf the version to October 18 (8.4%). On October 21 Nozzle added page 2 and changed how some metrics are weighted. That is why I always give the version date and check the study page before quoting it.

### What takes the usual sites' place

Social platforms and forums show up more often on poisoned pages. Nozzle compared top 10 shares for the same keywords on poisoned and other pages:

| Site | Other pages | Poisoned pages |
|---|---|---|
| facebook.com | 21.9% | 29.5% |
| instagram.com | 6.1% | 10.4% |
| youtube.com | 12.2% | 15.4% |
| quora.com | 2.4% | 5.0% |
| reddit.com | 20.7% | 23.0% |
| yelp.com | 11.5% | 10.8% |

On poisoned pages for US keywords, 12.4% have a foreign country-code domain (such as .ca or .co.uk) in the top 10, against 6.2% on the rest. Ads and shopping results stay stable: ads repeat from the previous measurement 83.9% of the time (against 70.2% at baseline), Shopping 93.1%. The list of AI Overview sources, on the other hand, loses a lot (37.5% kept against 68.1%), which matters to anyone monitoring visibility in AI answers.

### Which queries and countries are hit more

- **"Near me" queries**: 13.6% poisoned against 8.6% for everything else.
- **Long tail** (under 100 searches a month): 10.3%, against 6.6–7.6% for more popular keywords.
- **Brand queries**: the least, 5.3%.
- **US industries**: legal is hit hardest (17.4%, peaking at 54.6% on September 27), followed by home services (15.0%).
- **Device** barely matters: 8.8% desktop, 8.8% iPhone, 7.7% Android.

**By country**, the US sits at 8.8%, Germany at 6.4% and the UK at 4.1%, while Poland, where I work, is at 10.2% (81,000 judged measurements). I treat country figures with caution: they say more about how Google treats Nozzle's infrastructure in a given market than about that market's search results.

## Not just Nozzle: what other tools say

| Company | What it says | Scale |
|---|---|---|
| **AccuRanker** (Oct 19) | Large wave from late on Sep 24; Google tightened detection again on Oct 15, mostly on Android | 30.2% desktop and 25.7% mobile on Sep 25, 1.47% on Oct 1–9, under 1% from mid-October |
| **Ahrefs** (Oct 22) | Filters reject abnormal results and request replacements; historical data being cleaned up | 5–20% of results, depending on the data provider |
| **Semrush** (statement to Search Engine Roundtable) | Seen since late summer, extra checks on results | Impact "limited", individual cases mostly beyond page 1 |
| **SE Ranking** (Oct 16) | False drops and keywords shown as not ranking; data corrected retroactively | About 2% of tracked results still affected |
| **SISTRIX** (Oct 7) | Publishes only results that pass several checks; on-demand updates disabled; analysis focused on the top 10 | No figure; top 10 and Visibility Index "remain reliable" |
| **DemandSphere** (status page) | Mass refusals from Sep 10, "spurious results" from Oct 1; eventually a problem that is "not technical but cost" | About 11% of daily SERPs (post of Oct 7) |

The key observation from AccuRanker is that Google decides on poisoning **per browser session, not per keyword**. Two searches for the same thing at the same moment can get different pages. Hence their conclusion that the poisoned share depends on how a vendor collects data, not on what a customer tracks. Their distribution fits that: in each of 22 markets the share was between 1.1% and 1.7%. According to AccuRanker, more than 9 in 10 runs of poisoned days last a single day, and fetching the page again with another session returns a normal result about 3 times in 4. AccuRanker also found smaller rounds of the same pattern in its data as early as late January 2026.

According to PPC Land, on September 29 Nozzle measured about 27% and AccuRanker 1.4% (PPC Land used an earlier version of Nozzle's data). That gap does not mean one of them is wrong. They use different tests: Nozzle compares a result with a three-week baseline, while AccuRanker checks whether a day disagrees with both the day before and the day after. They also collect data on different infrastructure, so the numbers are not directly comparable. The practical takeaway for a tool user: the scale of the problem depends on the vendor, so ask your vendor for its own figures.

## Why is Google doing this? Facts and guesses

There are two facts. First: Google's spam policies, in the section on machine-generated traffic, explicitly list **scraping results for rank-checking purposes** as a violation of the policies and of Google's Terms of Service. Second: since 2025 Google has steadily made automated access to its results harder:

- **January 2025**: Search requires JavaScript (the SearchGuard system),
- **September 14, 2025**: the `num=100` parameter ends; collecting 100 results takes ten requests instead of one,
- **December 19, 2025**: Google sues SerpApi for circumventing SearchGuard (dismissed on July 20, 2026 with leave to amend; an amended complaint was filed in August and the case is ongoing),
- **summer 2026**: result links go through internal `/goto` redirects that have to be resolved separately,
- **July 2026**: Google and the FBI take down the NetNut proxy network, which some data collectors relied on,
- **September 2026**: mass refusals, followed by poisoned results.

Everything beyond that is interpretation. Nozzle sees poisoning as Google's response to bulk scraping by AI companies, with rank trackers as collateral damage. SISTRIX writes that Google wants to make it harder for competitors such as ChatGPT to access its index and to stop its results being used to train open-weight models. Barry Schwartz of Search Engine Roundtable assumes the targets are AI bots and SerpApi rather than SEO tools. None of these views has been confirmed by Google.

For me this debate is secondary. Whether an SEO tool is the target or collateral damage, the effect on a report is the same.

## The overlap with the spam update

Google's September 2026 spam update ran from September 24 to October 8, right at the peak of the poisoning. Lily Ray wrote at the time that she wanted to share her findings, but rank trackers were too unreliable. Glenn Gabe described a site that genuinely got hit on September 29 during the update (confirmed by the owner), and the tools showed it very differently: Ahrefs immediately, Semrush late, SISTRIX not at all.

Any drop between September 20 and mid-October can therefore have three causes: a real effect of the update, a gap in data collection or a poisoned results page. Lumping them together is the quickest route to a wrong diagnosis. How I check whether volatility affects a specific site at all is covered in my article on [Google ranking volatility without a confirmed update](/en/blog/google-ranking-volatility-unconfirmed-update/).

## How to spot a fake drop in your tool

This is the checklist I go through before I accept a drop as real. Points 1–6 are based on guidance from Nozzle and AccuRanker, points 7–9 on my own practice.

1. **A one-day drop that is back the next day.** According to AccuRanker, before the wave of poisoning such a complete page change happened about once per 1,000 daily checks. Today it is the most common symptom of poisoning.
2. **Most of your usual competitors disappear at the same time.** A real algorithm change usually shifts a few positions; it does not replace the whole first page at once.
3. **Facebook, Instagram, Quora or foreign domains suddenly appear in the top 10.** That is the pattern in Nozzle's data.
4. **Pagination stops after 2–4 pages**, or the results match only one word of the query.
5. **Local results show businesses from another city**, even though the location is set correctly.
6. **Compare with several weeks of history, not with yesterday.** Yesterday's measurement may have been poisoned too.
7. **Check Search Console**: same query, same day, country and device. If impressions and average position hold, the problem is in the tool's data.
8. **Refresh the keyword in the tool** if you can. AccuRanker and SE Ranking allow it; SISTRIX has disabled on-demand updates.
9. **Ask your vendor** three things: how it detects poisoned pages, what share of its data is affected and whether it re-fetches a page before storing it.

## Search Console as the reference point, and its limits

[Google Search Console](/en/blog/google-search-console/) reports data from real impressions in Google, so poisoned results pages do not affect it. It does not replace a tracker, though:

- it only covers your own verified properties, with no competitor positions and no share of voice,
- **average position** is an average across all impressions (countries, devices, result variants), not the result of a single check, so it cannot be compared one-to-one with a tracker position,
- for AI Overviews and AI Mode it shows impressions but no clicks or queries (more in my article on [AI Search reports in Search Console](/en/blog/search-console-ai-search-reports/)),
- it has its own jumps caused by changes on Google's side: after `num=100` was switched off in September 2025, many sites saw fewer impressions and a better average position even though their visibility had not changed (SISTRIX).

Search Console checks your site; a tracker shows the market. You need both, but when they disagree, Search Console decides.

## My data: Ahrefs, Semrush, Search Console and Bing side by side

I checked one of my own sites: [catalog.polo.blue](https://catalog.polo.blue/), a Volkswagen Polo parts catalogue whose traffic comes mostly from part numbers and equipment codes. I use the free version of Ahrefs (Webmaster Tools), so I have no Rank Tracker. I compare the Ahrefs index (Site Explorer: keywords Ahrefs sees the site ranking for, and estimated traffic) with Search Console, with Bing Webmaster Tools as a control. Screenshots from all four tools, taken on October 23, 2026, are below the tables. A daily Search Console chart for this site with the spam update marked, and a comparison with a second site, polo.blue, are in my [article on Google ranking volatility](/en/blog/google-ranking-volatility-unconfirmed-update/#example-an-auto-parts-catalogue-in-september-and-october-2026).

**Ahrefs:**

| Date | All positions | Keywords in top 3 | Estimated traffic |
|---|---|---|---|
| Sep 7 | 229 | 101 | 328 |
| Sep 20 | 186 | 76 | 269 |
| Sep 30 | 159 | 64 | 279 |
| Oct 13 | 121 | 39 | 256 |
| Oct 21 | 115 | 35 | 245 |

**Search Console and Bing (daily averages):**

| Period | GSC: clicks | GSC: average position | Bing: clicks | Bing: impressions |
|---|---|---|---|---|
| Aug 20 – Sep 9 (Nozzle baseline) | 274 | 7.7 | 21 | 983 |
| Sep 20 – Oct 13 (peak poisoning) | 307 | 6.1 | 20 | 1,180 |
| Oct 14–21 (Bing: to Oct 19) | 343 | 6.5 | 25 | 1,224 |

The directions are opposite. Since September 7, Ahrefs sees half as many positions and 65% fewer keywords in the top 3. Over the same period, Search Console shows 12–25% more clicks than the baseline and a better average position, while Bing impressions have grown steadily since spring at a stable position. Two independent data sources from the search engines themselves show stable or growing visibility, so I can rule out a drop in demand or a problem with the site itself.

![Ahrefs Site Explorer for catalog.polo.blue: pages with traffic and estimated organic traffic over 6 months, with a clear decline from September](../../assets/images/blog/zatrute-serpy-ahrefs-catalog-polo-blue.webp)

*Ahrefs (Webmaster Tools), Site Explorer for catalog.polo.blue, 6 months. I left only "Organic pages" and "Organic traffic" on the chart. Both lines rise until August and fall from September.*

![Google Search Console for catalog.polo.blue: clicks, impressions and average position over 3 months, with no drop in September or October](../../assets/images/blog/zatrute-serpy-gsc-catalog-polo-blue.webp)

*Google Search Console, property https://catalog.polo.blue/, 3 months. Clicks (blue line) and impressions (purple) hold or grow; average position (orange) improves from late August.*

![Bing Webmaster Tools for catalog.polo.blue: clicks and impressions over 6 months, growing steadily](../../assets/images/blog/zatrute-serpy-bing-catalog-polo-blue.webp)

*Bing Webmaster Tools, catalog.polo.blue, 6 months. Impressions and clicks grow from May, with no dip in September.*

For comparison I also looked at the free version of Semrush. It only shows data for the whole polo.blue domain (including other subdomains, as the notice in the screenshot says), but it does show daily figures for the last month. In the Polish database the keyword count fell from 689 (September 22) to 636 (October 8), about 8%, then recovered to 707 (October 22). Keywords in the top 3 stayed between 20 and 26 throughout. That is a calm, smooth curve, with no daily dips and none of the halving Ahrefs shows. Two tools, the same period and a completely different picture, which illustrates AccuRanker's point that the scale of the problem depends on how data is collected.

![Semrush Organic Rankings for polo.blue in the Polish database: daily keyword count over the last month, a mild dip in early October and a recovery](../../assets/images/blog/zatrute-serpy-semrush-catalog-polo-blue.webp)

*Semrush (free account), Organic Rankings, Polish database, desktop, 1 month. The report header shows catalog.polo.blue, but on the free plan the data covers every page of the root domain polo.blue, including other subdomains.*

I also checked a few individual keywords. For part number G052150A2, Ahrefs shows position 9, while Search Console averages about 3. For 6R0412249 it is 5 and 2.2. For "rivazza rims" the two sources agree, and Search Console shows a real drop from position 1.4 to about 4.5 that began on September 10, before the wave of poisoning.

**What this data does not prove.** I do not know whether poisoned results pages caused the gap. The Ahrefs decline is a smooth curve over several weeks, not the one-day dips Nozzle and AccuRanker describe. There are several possible explanations:

- Ahrefs sees the site on 100–240 keywords, while Search Console counts thousands of part-number queries,
- Ahrefs filters may be rejecting suspicious results while re-fetches lag behind (62 of the 100 keywords in the report were last updated more than two weeks ago, although without data from before September I cannot tell whether that is the normal rhythm),
- Ahrefs checks one location, while Search Console averages all countries and devices.

My conclusion is simple: for this period, Ahrefs data for catalog.polo.blue cannot tell me whether anything happened. If I reported to a client based on Ahrefs alone, I would show them a dramatic drop that does not exist in their traffic.

## What this means for reports and clients

- **Flag the period in reports.** From September 13 until further notice, I label rank tracker data as "uncertain" and add a note pointing to Search Console data.
- **Report weekly trends, not daily positions.** Most one-day dips recover within a few days (according to Nozzle, 67.9% within three days and 92.5% within two weeks).
- **Focus on the top 10.** SISTRIX has narrowed its analysis to page 1 because later positions can no longer be compared reliably over time, and Nozzle's data shows page 2 is hit more often.
- **Measure what the client feels**: clicks, traffic and conversions. I wrote about why positions alone describe visibility less and less well in my piece on [new SEO KPIs](/en/blog/new-seo-kpis-ai-citations/).
- **When a client sends a screenshot from a tool** saying "we've dropped off page one", I check Search Console for that query and day first, and only then look for causes on the site.

Data collection costs are rising (DemandSphere says outright that the problem is no longer technical but one of cost), so tools are moving to sampling and less frequent updates. I expect this to show up in pricing and in data coverage, although tool prices were rising before for other reasons, as with [Semrush's price increases](/en/blog/semrush-price-hike-2026/).

## What we still don't know

- **Whether Google will confirm anything**, and whether this is temporary. None of the sources knows; AccuRanker says plainly that Google has not said how long it will last.
- **When it really started.** Nozzle says September 20, Semrush "late summer", and AccuRanker sees smaller rounds from January 2026.
- **The real scale in your tool.** All figures are vendor claims, measured with different methods and with no independent audit. Only Nozzle and AccuRanker publish daily data.
- **Whether historical data has been corrected yet.** Ahrefs and SE Ranking say they are correcting it but give no details.
- **How badly volatility sensors** and AI Overview monitoring have been affected, since they rely on the same scraped results.

## Summary

A one-day drop in a rank tracker is no longer a reason to act. Since the second half of September, part of the data in these tools has been swapped results pages, and the scale depends on the vendor: from about 1% at AccuRanker to well into double digits in other measurements. Before you go looking for a penalty or a technical fault, check several weeks of keyword history and compare it with Search Console. Ask your vendor how it filters these results, and flag the period in your reports. On my own site, Search Console and Bing show growth, Semrush a small dip and recovery, and Ahrefs a fall by half, so for now Search Console is what I go by.

<FaqBlog
  questions={[
    {
      question: 'What is a poisoned SERP?',
      answer: 'A Google results page that looks normal (no error, no CAPTCHA) but has most of its organic results swapped out. Google serves such pages to some automated queries, including those sent by rank trackers, which then store them as real data.'
    },
    {
      question: 'Has Google confirmed that it deliberately serves swapped results to SEO tools?',
      answer: 'No. As of October 24, 2026, Google has not commented on the reports from Nozzle, AccuRanker, Ahrefs and others. The only known fact is that <a href="https://developers.google.com/search/docs/essentials/spam-policies" target="_blank" rel="noopener noreferrer">Google\'s spam policies</a> prohibit scraping results for rank-checking purposes without permission.'
    },
    {
      question: 'How can I tell whether a drop in Semrush or Ahrefs is real?',
      answer: 'Check whether the drop lasted one day and whether most of your usual competitors disappeared at the same time. Compare the keyword with several weeks of history, not just yesterday, and check the same query, day and country in Search Console. If impressions and average position hold, the problem is in the tool\'s data.'
    },
    {
      question: 'Can Google Search Console data be poisoned too?',
      answer: 'Not by poisoned results pages, because Search Console counts real impressions in Google. It has its own limits, though: it only covers your site, reports an average position across all impressions and shows no clicks or queries for AI Overviews and AI Mode. More in my <a href="/en/blog/google-search-console/">Search Console guide</a>.'
    },
    {
      question: 'Is a rank tracker still worth paying for?',
      answer: 'In my view, yes, because only a tracker shows competitors and the market, which Search Console does not. But you now need to ask your vendor how it detects poisoned results, read weekly trends rather than single days and verify every worrying drop in Search Console.'
    }
  ]}
  heading="Frequently Asked Questions"
  id="faq"
/>

## Sources