#poisonedSERP: How Google Search Results Changed and Why Rank Trackers Started Reporting False Drops
A sudden ranking drop in a report doesn't necessarily mean a site has lost its position in Google. Sometimes the rankings haven't changed, but the results returned to the tracking service have.
In the fall of 2026, SEO platforms reported incomplete and unreliable results from automated data collection. In one example from Serpstat, page one matched the query, while subsequent pages contained unrelated results and duplicates, with no CAPTCHA or error message.
For an SEO specialist, that means risking an inaccurate client report. For a site owner, it can mean changing pages that didn't need fixing. For a developer, it can mean saving bad data even though the collection process completed successfully.
1. Background: Three Waves of Tightening Restrictions, from num=100 to /goto Links
The October 2026 anomalies followed two earlier changes that had already made automated Google data collection more difficult.
Wave 1 (September–October 2025): The End of
&num=100 and a Tenfold Cost IncreaseIn September 2025, Google stopped supporting the
&num=100 parameter ¹, which had made it possible to collect 100 search results in a single request. Until then, a scraper could retrieve the top 100 results in one quick operation.The change had several consequences:
- Collecting the top 100 now required 10 sequential paginated requests instead of one.
- According to DataForSEO ² and ScraperAPI, the cost of collecting rankings down to position 100 rose almost tenfold.
- As SpyFu noted ³, some major SEO platforms, including Ahrefs, stopped regularly tracking positions 21–100 and reduced their standard reports to the first page.
Wave 2 (August 2026): Destination URLs Hidden Behind
google.com/gotoEleven months later, another Google change made collection more difficult: direct website addresses were hidden behind
google.com/goto?url=[hash] redirects. Scrapers had to resolve destination URLs instead of simply extracting links from the HTML.Moz/STAT ⁴ saw this as an attempt to increase the cost of OpenAI's grounding searches; The Web Scraping Club ⁵ saw an attempt to exert more control over SERP resellers. In August 2025, reports said ⁶ that OpenAI obtained some Google results through SerpApi. But that doesn't prove OpenAI was the target of
/goto.Search data providers soon adapted to the new link format. Soon, though, the problem shifted from links to the quality of the results themselves.
Wave 3 (September–October 2026): Data Quality Deteriorates
In October, the problem grew more serious: data providers started receiving incomplete and distorted SERPs. Before, the challenges were the number of requests and hidden destination URLs. Now, even a page that loaded successfully could contain unrelated results and duplicates. A scraper could finish without errors and save junk instead of a complete SERP. For rank trackers, that meant false drops in reports; for other systems, it meant bad data passed on for further processing. A scraper needs to collect complete, reliable results, not just retrieve a page.
2. How Incomplete SERPs Turned into False Ranking Drops
The first warning signs appeared in September: according to Serpstat, providers were receiving incomplete SERPs more frequently through automated collection. By early October, the problem was visible in reports: many rankings had dropped, but that didn't necessarily mean the sites had lost visibility.
September 28: SpySERP Detects Discrepancies in Search Results
On September 28, the SpySERP team detected discrepancies in the results it was receiving and paused data collection. Users saw this notice:
The issue concerned Google results, although the notice said collection had stopped for all search engines. On October 2, the team found a solution and resumed normal collection.
On October 5, Serpstat ⁷ published aggregated data from its Rank Tracker:
- Data for September 24–30 was relatively stable and used as a baseline.
- On October 1–2, the median recorded ranking position roughly doubled.
- By October 2, the share of positions in the top 30 had fallen by more than half compared with the baseline.
- The share of rankings at position 80 or worse rose to 1.8–1.9 times the baseline.
According to Serpstat, incomplete data from search results providers could have contributed to some of these changes. The company also noted that genuine ranking changes could have occurred at the same time: Google's September spam update was still rolling out.
October 16: SE Ranking Acknowledges a Data Issue
On October 16, SE Ranking acknowledged the issue: Google was returning incomplete or unreliable results to rank trackers. The company also published an incident status page ⁸ for customers:
The company said it had corrected data for the affected dates and restored checks for the vast majority of keywords. The issue still affected roughly 2% of tracked results.
3. Late October 2026: A #poisonedSERP Example
By late October, Serpstat had described another problem: an automatically collected SERP could look normal but contain unrelated results and duplicates. There was no error message or CAPTCHA.
On October 27, the company published a detailed case study ⁹ and described the signs of a #poisonedSERP in its knowledge base ⁷.
Anatomy of a Poisoned SERP: “best dedicated server usa”
Serpstat shared an automatically collected Google US mobile SERP for the query
best dedicated server usa.Here's what the page looked like:
- No visible block: No error message or CAPTCHA.
- Ads: The same SERP displayed an ad for dedicated servers in the US.
- Page 1: The results and search suggestions appeared relevant to the query.
But when the scraper requested page 2, covering positions 11–20, the results no longer matched the full query:
Code:
Search query: "best dedicated server usa"
─────────────────────────────────────────────────────────────────
PAGE 1:
Hosting providers and relevant search suggestions
PAGE 2:
According to Serpstat, results mainly match the word "server":
Tableau Server, Bitwarden on GitHub, Couchbase
PAGES 3 AND 4:
The GitHub result for Bitwarden appears on both pages
"People also search for" SECTION:
"Server meaning in English"
─────────────────────────────────────────────────────────────────How a Poisoned SERP Can Distort a Ranking Report
At first glance, collection succeeded: the page loaded, links and snippets were present, and there was no error message. But from page two onward, the results no longer matched the full query, and the same URL appeared more than once.
If this SERP is saved without further checks, a site's absence from the results can be mistaken for a ranking loss. A successful scrape can still lead to the wrong conclusion about a site's rankings.
An ad relevant to the query masks the problem: part of the page looks normal even though the organic results are already suspect.
4. Early November: More Reports of SERP Anomalies
By early November, providers were still reporting anomalies. Nozzle recorded ¹⁰ another spike. Three more reports appeared at the start of the month:
- November 3: SE Ranking revealed the scale of the problem ¹¹: in SE Ranking's sample of 10,000 Google US desktop keywords, the share affected by unrelated results jumped from 0.09% to 57.3% overnight in September. In 84% of affected cases, the unrelated results matched just one word from the query. For example, a search for
business bank line of creditreturned the LINE messenger app, The LINE Hotel, and the country band Florida Georgia Line. In 96.5% of cases, the unrelated results began at position 9 or lower. - November 4: SerpDino encountered repeated poisoned snapshots ¹²: collecting rankings again didn't solve the problem. Google returned the same poisoned snapshot repeatedly, filling report histories with false drops. The service had to change its collection algorithms and return credits to affected customers.
- November 5: Search Engine Journal reported /goto throttling ¹³: major providers, including DataForSEO, Serper, and SerpApi, faced significant delays resolving redirects. Because of Google's limits, some destination URLs remained unresolved, and scrapers saved
google.com/gotolinks instead of actual website URLs.
What A-Parser Collected for the Query in Serpstat's Case Study
On November 6, 2026, we checked the keyword
best dedicated server usa from Serpstat's case study using A-Parser's built-in SE::Google parser. We collected two pages each of Google US mobile and desktop results using residential proxies in New York City.In every check, the results were relevant to the query, with no
/goto links in the output. The results included hosting providers' websites, reviews, and discussions about dedicated servers in the US—not dictionary definitions or unrelated documentation. The output contained direct website URLs:| Example result | Destination URL |
|---|---|
| IONOS — dedicated servers | https://www.ionos.com/servers/dedicated-servers |
| OneProvider — servers in New York | https://oneprovider.com/en/dedicated-servers/new-york-usa |
| ServerMania — servers in the US | https://www.servermania.com/dedicated-servers-usa.htm |
| Atlantic.Net — dedicated servers | https://www.atlantic.net/dedicated-server-hosting/ |
| ScopeHosts — Dedicated Server USA | https://www.scopehosts.com/bare-metal-server/dedicated-server-usa/ |
| Kimsufi — servers | https://www.kimsufi.com/en/ |
5. The Era of DIY Scraping Is Officially Over
A simple Python script and a dozen datacenter proxies are no longer enough to consistently collect Google's top 100 results.
Data collection now requires you to:
- Recover website addresses when
google.com/gotoredirects are used; - Detect incomplete SERPs before updating a report;
- Identify suspicious results, duplicates, and discrepancies with manual checks.
Maintaining your own SERP scraping and validation process takes time: you need to follow Google's changes, update data extraction, and check result quality.
What A-Parser Handles
At A-Parser, we maintain the built-in
SE::Google parser so you don't have to maintain your own scraping code.The
SE::Google settings include the Bypass poisoned SERP option:
- Destination URLs: We added
/gotohandling to collect website URLs instead of Google's intermediate links. - Collection settings: Choose the number of pages, mobile or desktop results, country, language, and location.
- Proxies for collection: You can use A-Parser Premium residential proxies to scrape Google.
- Adapting to changes: We continuously monitor collection stability and update parsers in response to changes in search engines.
6. Will Poisoned SERPs Become the New Normal?
The spread of
/goto links shows how quickly a change in Google results can become commonplace. At first, industry participants noticed redirects in only a small share of results; then they began to spread widely. Google redirect links in place of direct URLs have now become a familiar part of collecting organic results.Perhaps #poisonedSERP is going through a similar testing phase. If Google is testing this as a way to counter automated collection and considers it effective, it could expand its use. What now looks like isolated corrupted snapshots could then become a recurring problem for far more queries and data collection systems.
Either way, keeping up with Google's changes requires ongoing maintenance: both link formats and how search results can be collected are changing. Writing a working script once isn't enough. The tool needs to be updated as Google changes.
Sources
[1] Search Engine Roundtable — Google drops the
num=100 parameter.[2] DataForSEO — Google Organic SERP API depth changes.
[3] SpyFu — The Top 100 Doesn't Have to Die.
[4] Moz/STAT — the spread of
/goto links.[5] The Web Scraping Club — Google Removed the URLs.
[6] Search Engine Roundtable — Google results via SerpApi for ChatGPT.
[7] Serpstat — Poisoned SERP, knowledge base.
[8] SE Ranking — service and incident status.
[9] Serpstat — Poisoned SERP: What It Is and How It Affects Rank Tracking.
[10] Nozzle — Poisoned Google SERPs.
[11] SE Ranking — study of unrelated Google results.
[12] SerpDino — Poisoned SERPs.
[13] Search Engine Journal — throttling of
/goto result links.
Last edited: