Poisoned Google results: can you trust your rank tracker right now?

The answer in brief

• Not for sudden drops, unless Search Console backs them up.
• Since September, several rank-tracking vendors report that Google sometimes answers their automated searches with a complete, normal-looking results page that lists the wrong sites.
• Nozzle, the enterprise rank tracker whose study calls these pages “poisoned”, classed 8.0% of the pages it could check between 13 September and 9 October 2026 as poisoned. Its revised study, with data to 21 October, puts the figure at 8.7%. Poisoning rose sharply again on 21 October, and AccuRanker still found 3.2% of desktop keywords swapped on 24 October.
• Semrush, Moz and Ahrefs have confirmed the problem, and AccuRanker, SISTRIX, SE Ranking and SEOmonitor report it in their own data. Google has not commented: AccuRanker’s latest update says Google has made no statement.
• A sudden fall in your tracker since September may come from a poisoned page, with no change in what real searchers saw. Confirm it for the same query and page in Search Console before it goes into a report.

What has changed since 19 October?

Several vendors have published new information since 19 October, and their data shows the poisoning has not stopped.

Ahrefs has responded. In a post on 22 October, Ahrefs said that between 5% and 20% of search results are affected, depending on the data provider. It says its filters reject abnormal results and request replacements, and that it is cleaning up past results that may have been affected. The post does not say which dates, reports or metrics the clean-up covers.

Nozzle has revised its study. Nozzle updated its research on 21 October, with data to that date, and changed its method. A page now counts as poisoned when fewer than half of a keyword’s usual sites appear anywhere on it. The sites no longer have to come back later, so each day’s figure is final when published. On page 1, 8.7% of pages were poisoned between 13 September and 21 October, against 0.17% in the baseline.

Nozzle now checks page 2 as well. Of the desktop results it could judge on both pages, 22.6% had page 1 or page 2 poisoned since 13 September, peaking at 43.5% on 30 September. Page 2 is poisoned one and a half times as often as page 1, and in 58.4% of those cases page 1 still looked normal.

The poisoning came back. On Nozzle’s page 1 or 2 measure, the rate fell to between 4% and 6% on 14 and 15 October. It then rose to 25.9% on 19 October and 35.7% on 21 October, its highest since 8 October. On page 1 alone, across all devices, the figure for 21 October was 15.6%. In the revised data, the UK has the lowest rate of the 45 countries Nozzle reports, at 4.1%, and South Africa the highest, at 22.4%.

SEOmonitor has published figures. According to its help centre, about 8% of the results Google returns to SEOmonitor each day are incomplete or incorrect, and are checked and fetched again before being recorded. It does not say what period that average covers. Separately, about 10% of the keywords it tracks had incorrect results recorded between 9 and 13 October, and it reprocessed those keywords on 13 October. After an unusually large drop, it may keep showing the previous rank for up to two daily updates while it checks.

AccuRanker’s figures need care. Its “sandwich check”, run on a fixed sample of one in every 200 tracked keywords, flagged 30.2% of desktop keyword-days on 25 September and 0.58% on 17 October. AccuRanker reports that Google tightened its detection again late on 20 October, hardest on desktop, and took a bigger step on both devices on 23 October. After AccuRanker’s own checks, 3.2% of desktop keywords in its sample still had a swapped page 1 on 24 October; on 25 October the figures were 0.6% on desktop and 1.0% on mobile. Separately, AccuRanker rebuilt its click-through-rate model on 21 October and expects Share of Voice to fall by about 35% to 60% for most domains. AccuRanker says rankings and real traffic are unchanged; only the estimate of clicks has moved.

What is a “poisoned” Google results page?

It is the term Nozzle’s study uses for a results page that looks normal but carries the wrong organic listings. In Nozzle’s words, Google answers some automated requests with “a complete page whose results are wrong”. The page shows no error or captcha and uses the usual template, so a tracker stores it as a genuine ranking.

Nozzle published its study on 10 October, with data to 9 October, and Search Engine Journal reported it on 14 October. The figures in this section are Nozzle’s own, from a cross-section of the tracking data it collects for customers in 49 countries.

Blocking came first. Requests to Google began failing in large numbers on 13 September, and from 20 September some of the pages that still came through were poisoned.

The share peaked at about a quarter of pages from 27 to 29 September. From 1 to 9 October, the last day in its published data, it ran at about 5% to 11% a day, though the last three of those days are still provisional. Over the whole period from 13 September, Nozzle classed 8.0% of the 17.8 million pages it could judge as poisoned. In its baseline, from 20 August to 9 September, the rate was 0.17%.

How does Nozzle decide a page is poisoned?

It first works out each keyword’s usual sites: those in its top 10 on at least 60% of checks between 20 August and 9 September. A later page counts as poisoned when fewer than half of those sites appear anywhere on it and they then come back: at least 80% of them on a later check within 14 days, or most of the sites missing since the previous check within three days.

Because the sites have to come back, the most recent days in any count are provisional. A change that lasts more than 14 days is not counted. The method measures the pattern but cannot show why Google serves these pages.

Nozzle changed this method on 21 October. The figures in this section use the original method and data to 9 October. Under the revised method, a page counts as poisoned when fewer than half of the keyword’s usual sites appear on it, whether or not they return, so recent days are no longer provisional.

What changes on a poisoned page?

The organic results change most. Several search features change too, while ads and Shopping results hold steady.

Social platforms fill the gaps. Reddit, Facebook, LinkedIn, Instagram and Quora appear more often on poisoned pages, and YouTube, Yelp and MapQuest less often.

Most episodes are short. Four in five spells last a single check. Among poisoned pages whose keyword was checked again within three days, 77.9% had the usual sites back by then, with a median wait of two days. In Nozzle’s revised data to 21 October, the figure is 67.9%.

Exposure also depended on how a request reached Google. Between 30 September and 9 October, one of Nozzle’s collection routes returned 1% to 7% poisoned pages while another returned 9% to 30% on the same days. By country, rates ran from 3.0% in the UK to 18.8% in Austria. In the revised data to 21 October, the UK rate is 4.1%.

Nozzle says the pattern is Google’s response to bulk collection of its results to ground AI answers. Rank trackers, it says, are caught because they collect pages the same way. That is Nozzle’s reading of its data, and Google has not confirmed it.

Why does a poisoned page matter more than a blocked one?

Because it fails silently. A blocked request errors out, the tool retries and the data arrives late. A poisoned page parses cleanly and goes into the database as a ranking, so your chart can show a drop that never happened in real results.

One of Nozzle’s examples is the US desktop search “white label local seo”. On 4 October, page one listed nine sites, mostly white-label SEO vendors. By 7 October all nine had gone. Their replacements included a WordPress development tool, a Jackson Hole steakhouse, a directory of county health departments and USA.gov, and pagination stopped at page three instead of ten.

The timing makes diagnosis harder. Google’s Search Status Dashboard shows its September 2026 spam update running from 24 September to 8 October, the same weeks the poisoning peaked. A drop on 29 September could be the update, a poisoned page, or both.

Trackers can also miss real changes. SEO consultant Glenn Gabe described a site the spam update hit on 29 September, confirmed by its owner. Ahrefs showed the drop, Semrush showed it much later, and SISTRIX still had not shown it on 7 October. SISTRIX says it keeps showing the last verified data until new results pass its checks, so some of that lag may be deliberate.

Which rank trackers say they are affected?

Semrush, Moz and Ahrefs have confirmed the pattern. Several smaller vendors have published their own measurements.

Semrush told Search Engine Roundtable it had been “seeing the ‘poisoned SERPs’ issue with Google returning inaccurate results since late summer”. It asks users to report anything that looks wrong.

STAT’s notice, posted on 1 October, says its team spotted the mismatched pages at the end of the previous week and now re-collects the pages it flags. Moz’s Tom Capper went further on LinkedIn on 16 October, writing that to Moz’s knowledge “every scraper and tool is affected to some degree”. Moz’s approach, he says, is to detect scrambled pages and retry until it gets a real one, which has removed most of the problem for some time. He adds that positions 11 to 20 tend to be hit hardest, so a big unexplained loss there in the past two weeks is probably not the spam update.

Ahrefs published nothing until 22 October, when it estimated that 5% to 20% of results were affected, depending on the data provider, and said it was cleaning up affected past results.

All of these companies sell rank tracking or search data, and several use their findings to promote their own checks. Read each figure as a vendor’s account of its own systems.

Why are the vendors’ numbers so far apart?

Over roughly the same weeks from 13 September, Nozzle counted 8.0% and Keyword.com 13.1%. DemandSphere gave an early estimate of about 11%, and AccuRanker, after a spike above a quarter on 25 September, saw about 1% to 2% a day until stricter checks in mid-October took it below 1%. Each vendor tracks its own keywords and markets, and Keyword.com used its own adaptation of Nozzle’s definition, so the figures cannot be ranked against each other. Ahrefs’ 5% to 20% is a range across its data providers, and SEOmonitor’s 8% counts results it caught before recording them, so neither compares directly with Nozzle’s share of judged pages.

AccuRanker applied its reading of Nozzle’s rule to its own data and got an average of 1.5% a day for 1 to 9 October, about a fifth of Nozzle’s rate. It also reports that Google seems to choose per browser session whether to serve a poisoned page. Its conclusion is that a tool’s collection method sets its exposure far more than the market or keyword being tracked.

AccuRanker also finds smaller rounds of the same pattern in its stored rankings from late January 2026, which makes 24 September the big step in a longer story. It adds that Google tightened detection again late on 15 October, hitting automated Android sessions hardest, so drops from 16 October, especially on mobile, need the same checks.

The vendors also disagree on detail. Keyword.com saw AI Overviews slightly more often on poisoned pages (69% against 64%) and YouTube nearly twice as often, the reverse of Nozzle’s findings. No single feature marks a poisoned page.

Has Google said anything?

Not publicly. Search Engine Journal noted on 23 October that none of the vendor statements it reviewed included one from Google. Neither Nozzle’s study nor Search Engine Journal’s report carries a Google statement, and AccuRanker’s latest update, with data to 25 October, still says Google has made no statement. Christian Ott of SEO-Kreativ checked Google’s Search Central blog, its Search Status Dashboard and its search staff’s Bluesky accounts and found nothing up to 16 October.

The nearest Google has come was in August, when a spokesperson confirmed to Search Engine Roundtable that google.com/goto redirects on result links were rolling out. The spokesperson said Google has “a long history of deploying technical measures against evolving forms of abuse”. Google’s spam policies also say that “scraping results for rank-checking purposes” without express permission breaches its policies and Terms of Service. Given that rule, a public explanation seems unlikely.

Why would Google do this?

The explanations on offer come from vendors and journalists. Nozzle points to AI grounding. SISTRIX believes Google mainly wants to make it harder for new competitors such as ChatGPT to use its index, and to stop its results being used to train open-weight AI models. Barry Schwartz of Search Engine Roundtable thinks the main targets are AI bots and SerpApi, a scraping service Google has sued, with SEO tools hit along the way.

None of these is confirmed. Nothing published by 19 October 2026 shows Google singling out the large SEO platforms.

How do I check whether a ranking drop is real?

Start with Search Console, because its figures come from real searches with no scraper in between. Then use your tracker’s history to see whether the whole page changed. Search Engine Journal’s weekly round-up recommends the same two checks.

1. Check Search Console for the same query and page. In the Performance report, filter to the query and the page, match the tracker’s country and device, and look at the days either side of the drop. If impressions, clicks and average position held steady while the tracker shows a collapse, the tracker’s reading is most likely wrong, and a poisoned page is the likeliest cause. If Search Console shows a matching fall, treat the drop as probably real. Google’s help page notes that Search Console labels days in Pacific Time, usually shows data after two to three days and leaves rare queries out when you filter by query, so a missing row proves nothing on its own.
2. Compare the page with the sites that ranked on page one in late August. Open the stored results page for the day of the drop and set it beside two or three from late August. If most of the regular competitors vanished on the same day, suspect the data. Avoid comparing only with the previous day, which may have been poisoned too. Keyword.com found a smaller wave from 4 to 10 September, so early September is not a safe baseline either.
3. Wait a few days and look again. Most poisoned episodes clear within three days, while a real ranking change persists. If your tool offers a manual refresh, run it on the affected keyword. AccuRanker says a refresh in its own app, which fetches the page in a new session, will most likely fix a bad day, and that a second fetch comes back normal about three times in four.
4. Use the warning signs to decide what to check first. More social and forum sites in the top 10, local results from another town and pagination that stops early all show up on poisoned pages. None of them proves anything alone. Moz’s Tom Capper suggests checking positions 11 to 20, which tend to be most affected by the scrambling. Some gaps below position 10 are simply missing data: DemandSphere said on 15 October that it was still fetching page one only for most accounts, so its Top 20 figures are effectively Top 10 for now.

Search Console has limits. Its position figure is an average across impressions, so expect it to differ from a single tracker reading, and compare like-for-like date ranges. It covers only your own site, so you still need a tracker for competitors and search features. Neither tool shows whether AI answers mention your site; our guide to AI search visibility has a checklist for that.

What does this mean for ranking reports since September?

Rankings collected since mid-September are less reliable than usual, and how much less depends on the tool that collected them. Treat any tracked figure from those weeks as a vendor’s estimate, and check one that would change a decision against Search Console first.

The worst of it, on the evidence so far, ran from 20 September into October. Nozzle’s poisoned share peaked on 29 September, during the spam update, so those are the days when a real loss and a false one are hardest to tell apart. Several vendors have tightened their checks since: AccuRanker was below 1% by 17 October, and SE Ranking says it has corrected every affected date. AccuRanker also saw Google tighten its detection again late on 15 October, late on 20 October and on 23 October, so newer data needs the same care.

Nozzle’s data shows poisoning rising again from 19 October to 35.7% on 21 October, the last day in its revised study, and AccuRanker still found 3.2% of desktop keywords swapped on 24 October. Check drops on 21 and 24 October against Search Console, as you would one from late September. AccuRanker users should also expect lower Share of Voice and Traffic Value from 21 October because of its new click model. Treat that as a change in the estimate, and check rankings and visits before reporting it as a loss.

The errors run both ways. A poisoned page can show a fall that never happened, most often on page two by Moz’s and Semrush’s accounts. A real change can also show up late: in Glenn Gabe’s case, SISTRIX had still not shown a spam-update loss eight days on. The second source settles it. Search Console shows whether real searchers saw a change, and organic visits and enquiries show whether it mattered.

Analytics has had its own October problem. From 8 October 2026, many ChatGPT links lost their UTM tag and referrer, so GA4 often counted those visits as Direct. Our GA4 audit guide shows how to check that before those figures go into the same report.

Until Google or the vendors say more, treat a rank tracker as a sample of Google and Search Console as the record of your own site. A sudden drop since September needs a second source before anyone acts on it.

If a drop in your reports needs checking against Search Console, talk to AIWIZ.