Semrush Confirms Inaccurate Search Results

Semrush confirmed that Google is returning inaccurate search results to some automated ranking tools. It joins other SEO platforms reporting so-called “poisoned” search results.

Catch up quick.

Semrush called the problem industry-wide in its statement. Ahrefs, AccuRanker, BrightLocal and other rank trackers have reported similar issues since September. Search Engine Land’s Barry Schwartz first reported on this issue Sept. 15 in Google Search rank and position tracking is a mess right now.

Why we care.

Sudden ranking changes in third-party tools may reflect inaccurate search results rather than actual changes in Google rankings. Check Google Search Console and historical rankings before acting on unexpected drops or gains.

Semrush confirms inaccurate results.

Semrush described two types of poisoned search engine results pages (SERPs):

  • Irrelevant results: Pages contain results that don’t match the search query, sometimes on page one.
  • Partially inaccurate results: Pages initially look normal but contain loosely related results further down.

What’s happening.

Semrush cited a search in Brazil for a specific baby stroller model that returned a dental supply store’s equipment page. Both products shared the Portuguese word “carrinho.”

  • Semrush said it adjusted its data sources and added checks to detect and exclude invalid results. It also now collects potentially inaccurate results again.
  • The company advised customers who notice unusual ranking changes, particularly for unrelated keywords, to compare their reports with Google Search Console and contact support.

Widespread SEO impact.

In an Oct. 22 update, Ahrefs said 5% to 20% of search results were affected, depending on the data provider. Ahrefs said it uses filters to reject abnormal results and requests replacement data. It is also cleaning up historical results that may have been affected.

Other providers reported similar problems:

  • DemandSphere: Flagged 6.7% of collected SERPs as suspected poisoning on Oct. 23.
  • SEOmonitor: Reported that about 8% of its daily results were inaccurate or incomplete. Incorrect data affected roughly 10% of tracked keywords from Oct. 9 to 13.
  • SISTRIX: Confirmed receiving results that differed from expected rankings. Additional quality checks have delayed some keyword updates. It disabled on-demand updates in September.
  • The figures come from different samples and detection methods, so they are not directly comparable.

Nozzle analysis of millions of SERPs.

A Nozzle analysis examined 56.1 million Google SERPs across 53 countries from Aug. 20 through Oct. 21. Nozzle compared current results with websites that regularly ranked for each keyword during its Aug. 20 to Sept. 9 baseline period. Since Sept. 13, Nozzle flagged 22.6% of SERPs checked on both Page 1 and 2 as poisoned. That was up from 0.28% during the baseline period.

Among its other findings:

  • 43.5%: The highest daily rate, recorded Sept. 30, among SERPs evaluated on both pages.
  • 17.9%: The share of evaluated SERPs with suspected poisoning on Page 2, compared with 12.2% on Page 1.
  • 67.9%: The share of affected SERPs that returned to their usual results within three days.

The study also found that Facebook, Instagram, YouTube and Quora appeared more often in suspected poisoned results than in other results for the same keywords.

First-page rankings also affected.

BrightLocal said the problem primarily affected rankings below position 10. But Semrush, AccuRanker and Nozzle also documented inaccurate results on Page 1.

  • AccuRanker reported that 40% of its initial desktop requests on Oct. 24 returned suspected poisoned first-page results. After its checks, 3.2% of sampled desktop rankings still failed its test. That figure fell below 1% the next day after AccuRanker adjusted its collection methods.
  • AccuRanker compares results with historical rankings. When most of the usual websites disappear, it requests another collection. DemandSphere and SEOmonitor have added similar checks.
  • DemandSphere acknowledged that algorithm updates, Google tests, and normal ranking changes can produce similar patterns.

Google has not commented.

Google has not confirmed that it deliberately serves altered results to automated ranking tools.

  • Semrush and other providers have suggested Google’s efforts to restrict large-scale automated data collection, including collection for AI systems, may explain the changes. Google has not confirmed that explanation.
  • AccuRanker reported further changes on Oct. 20 and 23, affecting desktop and mobile results. After Oct. 23, between 4% and 6.5% of initial requests continued to return suspected poisoned results. That was up from 1.6% to 3.3% beforehand.
  • AccuRanker said its additional checks reduced the number of inaccurate results reaching customer reports. But they had not eliminated the problem.