Ahrefs Estimates 5-20% of Google Results Were Poisoned

Ahrefs estimates that 5% to 20% of search results have been affected by what it calls the "poisoning" of Google results over the past few weeks, depending on the data provider. The SEO tool maker said Thursday it is "cleaning up any past results that may have been affected". Its post does not specify which dates, reports or metrics the cleanup covers.

The problem, as rank-tracking company Nozzle framed it in an Oct. 10 article updated Oct. 21, is that Google has been serving results pages missing many or all of the sites that normally rank for a given term. Nozzle counts a first page as poisoned when fewer than 50 percent of its normal sites appear anywhere in the collected results. By that test, 8.7 percent of Nozzle-determined first pages were poisoned between Sept. 13 and Oct. 21, against 0.17 percent from Aug. 20 to Sept. 9. Nozzle states this trend "is Google's response to AI scraping, and rank trackers are caught in it." Roger Montti covered the original article on Oct. 14.

What do the vendors check before storing a result?

AccuRanker eliminates shortened Google results pages, then fetches the page again in another browser session. It compares the new page 1 against the page 1 sites from at least two of the previous three days. When fewer than half of those sites appear, it pulls the results again. If the second pull shows the same new sites, AccuRanker records the page as an actual change. According to AccuRanker, roughly one in every 20 desktop pages and one in every 12 mobile pages lack the history data needed to run this check.

SEOmonitor estimates that around 8 percent of results Google returns each day are incomplete or incorrect and get pulled again before recording. On Oct. 13 it switched to checking each result's content against the keyword's own recent history as results arrive. SEOmonitor says in its help center that between Oct. 9 and 13 "some of the incorrect results were recorded in the platform"; affected keywords were reprocessed on Oct. 13.

Ahrefs says its filters detect and reject abnormal results and request replacements. The post does not say whether its 5-20% share is measured before or after filtering.

BrightLocal is engaging its current data suppliers and adding more, so a failure at one provider does not leave it without data. CEO Myles Anderson wrote in an Oct. 16 blog post that there is "no single fix that solves it completely" for BrightLocal or anyone else in the industry.

What does a customer see while a result is unconfirmed?

The answer depends on each tool's validation and display rules. SEOmonitor, which has observed rank volatility since around May, "may keep showing the previously recorded rank for up to two daily updates" after unusually large drops, such as Rank #4 to Rank #14, or Top 50 to Rank above 99, until it confirms the lower rank persists. Its help article links to a recrawling option and a Chrome extension for comparing rank values.

AccuRanker runs a secondary fetch before a suspect page reaches the customer's dashboard, though it does not state what appears during the re-check. Users can request a keyword refresh in the application, which fetches a new page in a new session.

What happened to rankings already recorded?

SEOmonitor reprocessed all keywords affected by incorrect September results for Sept. 25 through 28 and says it blocked erroneous data from entering its product until Oct. 8. From around Oct. 9, many erroneous results no longer carried the signs its quality control depended on. Approximately 10% of monitored keywords had incorrect-result issues between Oct. 9 and 13, though SEOmonitor reports most movements in its Visibility metric were minimal.

AccuRanker applies a "sandwich check" to stored data, flagging any day when fewer than 50% of page 1 sites match both the previous and next day. Using a fixed sample of one in every 200 keywords, the flag rate was 30.2% for desktop keyword-days on Sept. 25 and 0.58% on Oct. 17. The company does not state whether flagged days will be revised in stored data.

SerpApi wrote in a public issue on Oct. 14 that it had deployed an update that "should reduce the number of instances in which poisoned results are returned" and that work continues. A customer asked in the same issue on Oct. 13 for a field flagging searches without relevant results, noting downstream services currently review each response manually. AccuRanker's API, AccuSERP, returns the page Google sends for a single search and cannot run history checks; it draws first on browser sessions that passed the rank tracker's check and have stayed clean since.

What should practitioners do now?

Teams reviewing late-September or October drops should first check how their vendor reported during that window. If an agency resends September or October data, it should ask the vendor whether stored data for those dates was updated, and compare older exports before sending them to a client. Google's Search Console can serve as additional evidence in a drop dispute, though average position varies by device, location, time and user history. As of Oct. 22, no vendor statement includes input from Google — AccuRanker writes "Google has made no statement" — and none cites independent testing of how well its checks work. SEOmonitor will update its incident page as Google's changes develop; Ahrefs will post updates in its dashboard. Check those updates before using figures from an affected period in a client or management report.