Poisoned SERPs: why your rank tracker shows drops that aren't real

Poisoned SERPs are Google results pages that look normal but have some or all of their results swapped out. Since September 2026 Google has been serving such pages to automated queries, which means to rank trackers and SEO tools like Semrush, Ahrefs or AccuRanker. The tools then record positions that a real searcher never sees.

So if a ranking crashes overnight and is back the next day, that isn’t necessarily a Google update. Below is what the vendors have confirmed, how to tell a bad reading from a real drop, and how to work with the data in the meantime.

What’s happening: Google answers tools with wrong pages

Google used to block suspicious queries, for example with a CAPTCHA. The tool then knew it had no data. Now it gets a complete results page with the usual layout, only the results are wrong. Nozzle puts it this way: Google is not refusing these requests, “it answers them with a complete page whose results are wrong.” For a tool that is worse than a block, because nothing in the HTML marks the page as wrong.

An example from Semrush: a Brazilian search for a baby stroller model returned a dental supplier’s equipment page, because both use the Portuguese word “carrinho”.

What is confirmed, and by whom

All the major vendors see the problem, but they measure it differently, so the numbers aren’t directly comparable.

• Nozzle (study of 10 October 2026, updated 1 November): from 13 September, requests to Google started failing in large numbers, and from 20 September the ones that got through came back poisoned. Since 13 September, page 1 was poisoned on 14.3 % of judged results pages, against 0.17 % before. Page 1 or page 2: 26.9 % against 0.28 %. The worst day so far was 31 October at 64.0 % (page 1 or 2). The 8 % you’ll see in earlier coverage comes from the first version, with data up to 9 October.
• AccuRanker (post of 19 October 2026, updated since): from late September to mid-October, the share of poisoned pages after its own checks was between 1.1 and 1.7 % in each of 22 markets. On 25 October it was 0.6 % on desktop and 1.0 % on mobile. AccuRanker sees smaller rounds of the same pattern in its data from late January 2026.
• Ahrefs (update of 22 October 2026): depending on the data provider, 5 to 20 % of search results are affected. In an addition on 6 November, Ahrefs says page one data is now largely recovered and stable.
• DataForSEO (post of 23 October 2026): its CTO puts poisoned SERPs at roughly 20 to 30 % of scraped results since 26 September, with noticeably fewer since 10 October. Even two retries with historical validation leave 2 to 3 % poisoned. According to DataForSEO, there is no solution yet that filters them out with 100 % reliability.
• Semrush (statement of 27 October 2026): confirms an industry-wide issue and excludes clearly invalid results from its database. Semrush gives no percentage.
• DemandSphere put the share at about 11 % of its daily results pages, as reported by PPC Land, without describing how it measured this.

As of 7 November 2026, Google has not commented publicly. The Search Status Dashboard lists only the September 2026 spam update for this period, which began on 24 September and ran for almost 14 days. It overlaps with the problem, which makes real volatility and bad data harder to separate.

Why it is happening

There is no official reason. The vendors broadly agree that Google wants to make bulk collection of search results harder, mainly for AI training and AI answer engines, and that rank trackers are caught in the middle. That is their interpretation, not a confirmed fact.

What can be sourced is the context around it:

• In January 2025 Google started requiring JavaScript for Search. A Google spokesperson told TechCrunch this helps protect its services and users “from bots and evolving forms of abuse and spam”.
• On 19 December 2025 Google sued SerpApi, a search data provider, for circumventing its security measures and reselling content from the results pages.
• On 2 November 2026 DataForSEO reported that Google is throttling the new “goto” redirect links in its results. The faster a tool resolves the links on a page, the longer the delays get, up to 120 seconds per URL.

How to recognise a poisoned SERP

The warning signs from the Nozzle, AccuRanker and DataForSEO posts:

• Most of page 1 is replaced at once. The usual domains are missing, and in their place you find forums, social sites (Facebook most of all, per Nozzle), sites from other countries, or results that only match one word of the query.
• One keyword drops while related keywords hold. A real problem with a page usually hits several terms.
• The drop doesn’t match the keyword’s history. Nozzle advises comparing with the usual sites over the previous weeks, not with yesterday, which may have been poisoned too.
• It doesn’t last. Per Nozzle, 46.9 % of poisoned SERPs had their usual sites back within three days and 90.4 % within two weeks. A real ranking change stays.
• Page 2 looks odd. Nozzle finds it poisoned one and a half times as often as page 1.
• Search Console disagrees. If your page has “disappeared” in the tool but Search Console still shows impressions and clicks, nothing happened to your page.

To check, search yourself: a private browser window, logged out, with the right country and ideally the right location. That doesn’t scale, but it’s enough for the three or four keywords where a drop would really hurt.

How to deal with it

1. Treat single-day drops as unconfirmed. DataForSEO recommends taking a drop seriously only once a second reading or Search Console confirms it.
2. Look at trends over several days and across keyword clusters. A cluster that declines over a week is a signal. A single outlier isn’t.
3. Use Search Console as the reference for your own site. It shows real impressions and clicks from Google Search, not scraped data. Know its limits: average position is the topmost position averaged across all queries where your page appeared, and it is only recorded when there was an impression. The most recent days are preliminary. And the data only goes back 16 months, so export it if you want to analyse this period later.
4. Ask your rank tracker how it filters. Does it re-fetch suspicious pages? Compare with history? Mark uncertain readings? The vendors above describe this openly. Others don’t.
5. Add a note to reports. When you report to a boss or client, state that ranking data has been unreliable industry-wide since mid-September 2026, with a link to one of the sources above. DataForSEO suggests annotating reports from 20 September.
6. Don’t change anything because of one bad data point. Don’t rewrite content or revert changes before Search Console shows the decline.
7. Adjust automated alerts. DataForSEO recommends triggering alerts only when two consecutive readings show the same change.

For a quick look there is my free Ranking Check. It queries your keywords for Berlin, Frankfurt and Munich, compares the cities’ results pages with each other and fetches suspicious pages a second time. If a page still looks wrong after that, it shows “Result uncertain” instead of presenting a position as fact. The check uses DataForSEO data, so it is affected just like any other tool. It’s a snapshot, not a verdict.

What happens if you don’t know about it

The risk is less the data itself than the decisions built on it:

• You read a measurement error as a ranking loss and blame a Google update.
• You revert a good change or rewrite a page that was fine.
• A client report shows a drop that never happened, and a budget decision follows from it.
• Automated alerts report crashes, and the team spends a day looking for a cause that doesn’t exist.
• A competitor analysis compares you with domains that only show up on swapped pages.
• AI visibility tools built on the same Google data show the same errors. How to measure visibility in AI search and AI Overviews with first-party data is covered in my post on measuring AI search visibility with Search Console and GA4.

Conclusion: Search Console first

Rank trackers are still useful for trends and competitors. Right now they don’t work as the only source for “we lost rankings”. When a ranking moves, check Search Console first, then the results page itself in a private window, and act only when both show the decline.

For a quick look across three cities, use the free Ranking Check. If you want to know whether there’s something real behind a decline, such as a technical problem, the €290 diagnostic analyses 16 months of Search Console data against Google’s update timeline. It works from your real data, not from scraped results pages.