Google SERP anomalies what providers measured and how to verify a ranking drop

In short

• Nozzle reports that some automated checks received a complete Google page with a different set of sites instead of a captcha or error. This is a vendor sample, not a mechanism or intent confirmed by Google. The tracker sees no failure and stores the page as real.
• Published estimates differ because teams measure different samples and periods. In Nozzle’s updated study, 8.7% of page-1 SERPs were poisoned from September 13 to October 21, against 0.17% in the baseline period.
• In a separate desktop sample of 4.7M SERPs judged on both pages, an average 22.6% had page 1 or 2 poisoned from September 13 to October 21; the daily rate reached 35.7% on October 21. This cannot be compared directly with the page-1-only estimate.
• When a project’s site is one of a keyword’s usual sites, it was missing from the top 10 on 60.8% of poisoned SERPs, against 2.7% in the baseline. Nozzle counts a return when at least 80% of the usual sites are back: that threshold was met within three days in 67.9% of cases with a follow-up in that period, and within two weeks in 92.5% of cases.
• Four in five spells last one SERP. Don’t act on a single check; compare repeated measurements with Search Console.
• Google’s published timeline lists a spam update from September 24 to October 8. A late-September drop can reflect a measurement anomaly, the update, or both. Search Console helps tell them apart.
• Google hasn’t confirmed any of this. All figures below are third-party measurements with different methods.

Complete page, wrong sites

With a normal block, the tracker gets a captcha or an error, detects the missing data and retries. In the scenario described by Nozzle, the page arrived complete, in the normal template, with ads and SERP features, but with a different set of organic results.

What changes on such a page. Nozzle compared poisoned SERPs with normal ones for the same keywords:

In Nozzle’s sample, organic results turned over almost completely in one check, while ads were kept even more often than usual. No single sign proves anything. Together with a short spell and a return of at least 80% of the usual sites, they are consistent with the vendor’s reported anomaly, but do not establish its cause.

Social sites and forums show up more often, but they don’t take over. Facebook rises the most; YouTube also appears more often. Yelp is one of the few sites with a small decline.

21 months of Google making rank tracking harder

Nozzle’s observations came amid changes that made automated SERP collection harder from early 2025. Some changes affected rank trackers; vendors connect the latest anomaly with complete pages being saved instead of an explicit error.

Why estimates of poisoned SERPs differ

Published estimates differ because teams measure different samples, pages and periods. Check what each one counted before comparing the figures directly.

In Nozzle’s updated method, a page is poisoned when fewer than half of its usual page-specific sites appear anywhere in the results. Usual sites appeared in the top 10 on at least 60% of five or more baseline SERPs (August 20–September 9). Return is measured separately: at least 80% of the usual sites must appear on a later SERP, and a return is not required for classification. In the separate desktop sample of 4.7M SERPs judged on both pages from September 13 to October 21, 12.2% of page 1s and 17.9% of page 2s were poisoned; together, an average 22.6% versus 0.28% at baseline. The daily rate for this two-page sample was 35.7% on October 21; its peak was 43.5% on September 30. Page 2 was hit 1.5 times as often; in 58.4% of page-2 cases, page 1 still had its usual sites. These figures are not directly comparable with the 8.7% page-1 figure for the full sample. TrueRanker counts mismatches from page 2 onward, so it answers a different question.

One check proves nothing

Four in five poisoning spells lasted a single SERP. Nozzle counts usual sites as back when at least 80% of the keyword’s usual sites are present on a later SERP. That threshold was met within three days in 67.9% of cases with a follow-up in that period, and within two weeks in 92.5% of cases. One check alone cannot establish a lasting ranking change; repeat measurements and compare them with Search Console.

Here is one way poisoning can look on a ranking chart. This is a diagram, not real data: one keyword, 14 daily checks, two scenarios. With poisoning, the site vanishes from the top 100 for one or two checks and comes back to the same spot, while Search Console clicks don’t change. A real ranking loss may persist or worsen, and clicks may also fall; the speed of change alone does not establish the cause.

Where poisoning is higher

Poisoning varies by country, from 4.1% in the United Kingdom to 22.4% in South Africa. The US is at 8.8%; Austria, Switzerland and Belgium are notably higher.

The device shares are close: 8.8% on desktop, 8.8% on iPhone and 7.7% on Android. “Near me” queries were affected more often (13.6% vs 8.6% for other queries). The same phrase tracked nationally and locally was nearly equal: 8.2% and 8.4%.

The study covers 53 countries. The map shows the 45 with at least 10 datasets and 10,000 judged SERPs. The figures describe Nozzle’s sample only; the study has no comparable measurements for other search engines.

Poisoning or a real drop: a six-question check

Answer six questions about the keyword that “dropped”. The check tells you which explanation fits better and what to do next. These are clues, not proof of cause: a short spell followed by a return, unfamiliar sites in place of the usual ones, and unchanged Search Console data.

How to read rankings while poisoning continues

Seven rules we suggest for ranking reports while poisoning continues.

• Median, not a snapshot

  Report the median of the last three checks. Any single check may be poisoned.

• Repeat checks before a conclusion

  Compare repeated checks with Search Console. Nozzle counts usual sites as back when at least 80% return; the 67.9% three-day figure covers cases with a follow-up in that period, and 92.5% were back within two weeks. Four in five spells last one SERP.

• Search Console decides

  If rankings drop but the page’s clicks and impressions hold, trust Search Console.

• Check against the usual top 10

  Keep the sites that usually rank in the top 10 for a keyword over 4+ weeks. If fewer than half are present, flag the check and repeat it.

• Mark the observed period

  When comparing results, mark Nozzle’s sample from September 13 to October 21, 2026 and keep it separate from the August 20–September 9 baseline.

• Analyze the update separately

  Check drops between September 24 and October 8 against Search Console, because the spam update overlapped the tracking failures.

• No hasty fixes

  Don’t rewrite pages or remove links based on one SERP snapshot.

FAQ

Sources and method

This is a review, not our own SERP sample. We used published figures only and named the source for each. Where methods differ, we show the figures side by side and explain the difference rather than averaging them. Sources accessed October 24, 2026.

The SERP diagram and the “poisoning vs real drop” chart are illustrations with made-up data, and they are labeled. The ranking rules and the six-question check are our method, built on the signs from the sources.

Nozzle sells rank tracking, and the other sources are also market players or media. Google has neither confirmed nor denied their findings.

1. Nozzle — Poisoned Google SERPs
2. Search Engine Journal — Analysis: Google Is Poisoning Search Results For Rank Trackers
3. Search Engine Roundtable — Third-Party Google Rank Tracking Not Fully Working Now
4. PPC Land — Google returns wrong results on 11% of DemandSphere's daily rank checks
5. TrueRanker — Google is serving rank trackers pages that look real and aren't
6. Dalin de Graff — Your Rank Tracker Broke for Three Weeks
7. LoopLoop — Search & AI news, 8–12 Oct 2026
8. Relevant Audience — Google Search goto passthrough URLs confirmed
9. SerpApi — Google now requires JavaScript
10. Noble Studios — Google retires num=100
11. Search Engine World — Google's SerpApi scraping lawsuit tossed