Did Your Rank Tracker Show a Collapse in Late September? Google Likely Served It a Fake SERP

If your rank tracker showed a collapse in late September, hold off on rewriting the strategy. On October 10, Nozzle published "Poisoned Google SERPs": starting September 20, Google began serving rank-tracking services substituted results. No errors, no captcha, outwardly completely normal. Here's what's in the data, how to tell a substitution from a real drop, and why I personally stopped watching positions a long time ago.

What happened

Nozzle analyzed 27.5 million SERPs between August 20 and October 9, 2026 — more than 660,000 keywords across 49 countries. The timeline:

• September 13 — requests to Google started failing in large numbers; for about two weeks fewer SERPs reached the trackers than usual;
• September 20 — the first poisoned pages appeared;
• September 27–29 — the peak: 24.9%, 26.7%, and 27.0% of SERPs respectively.

For comparison: before September 13 the rate of such anomalies was 0.17%. After, it averaged 8.0%. Nearly a fiftyfold increase.

What the substitution looks like

It's an ordinary results page. Standard layout, normal HTML, a clean response. The research puts it plainly: "No error, no captcha." Only one thing differs — the results themselves.

In normal SERPs, 88.5% of the previous measurement's sites are still there on the next one. In poisoned SERPs, 20.7%. Roughly four-fifths of page one is replaced by other sources. Social networks and forums fill the gaps:

Detection works by comparing against a keyword's history, not by inspecting the page. Nozzle treats as "usual" the sites that appeared in that keyword's top 10 in at least 60% of measurements between August 20 and September 9. A SERP counts as poisoned when fewer than half of those sites are present and they return in later measurements.

What this does to reports

The headline figure: on poisoned pages the tracked site fell out of the top 10 in 58.6% of cases. In the baseline, 2.7%. On the same keywords' other measurements after September 13, 10.2%.

And here's why substitution is more dangerous than an outright block. When a scraper gets an error or a captcha, the tracker marks the collection as failed and retries. When it gets a valid HTML page, it faithfully writes into history: the site is not in the top 10. The client's chart shows a collapse that never happened, and that collapse stays in the data forever.

The good news is that the effect is brief. 80% of spells lasted exactly one measurement, 15% lasted two. In 77.9% of cases the usual set of results was back within three days, with a median of two.

A three-minute test

1. Compare against history, not against yesterday. Yesterday's measurement may have been poisoned too. Look at which sites held the top 10 for this keyword over several weeks.
2. Look at the neighbors. If every niche competitor vanished from the top 10 at once and Reddit, Facebook, Quora, and LinkedIn took their place, it's a substitution. A real update doesn't look like that.
3. Open Search Console. If impressions and clicks are still there, the site hasn't gone anywhere. This is the fastest and most reliable test.
4. Wait three days. A substitution resolves on its own. A real drop stays.

What the research doesn't say

The material is strong, but it deserves to be read exactly as written.

No cause is named. The theory that this is Google defending itself against AI scraping is circulating around the industry, but it isn't in the research. Nozzle describes observed behavior, not Google's motives. No 403 or 429 status codes are cited either — only that requests started failing.

This is tracker-level data, not user-level. The study describes what Google serves to rank-tracking services. It does not claim that a human in a browser saw the same results.

There are no recommendations in it. All Nozzle offers is a way to identify the substitution: a multi-week baseline, expecting recovery within days, and cross-checking Search Console.

My opinion, and it's broader than this story

What follows is a personal position, not a conclusion from the research. I believe rank tracking became obsolete the day search became personalized. Position long ago stopped being a property of a website: it depends on the user's history, location, device, and session. A tracker doesn't show you the SERP — it shows you what Google served the tracker. The poisoned SERP story is a vivid illustration of that fact, not a new problem.

Average position in Search Console has always been enough for me. It isn't perfect either — it's an average across all queries and all users — but it has one decisive advantage: Google itself calculates it from real impressions, rather than a scraper measuring from one machine in one data center. How to measure it properly, with non-overlapping windows and a baseline from the rest of the site, I covered in the contrast analysis piece: a store's own data there showed that without a baseline, any position figure means nothing.

The one thing I genuinely need SERP scraping for is competitive content analysis — seeing who holds the top for a query and what's on their pages. For that I use an XML feed. It differs slightly from what a human sees in a browser, but for comparing content it's good enough.

What to do now

• Review your reports for September 20 – October 9 before showing them to a client. If there's a sharp one- or two-day dip in that window, cross-check Search Console.
• Don't make decisions on one-day drops. Ever, really, but especially in this period.
• Add a check of the top-10 composition to your process. Social networks and forums where they've never been is a diagnosis.
• Move client reporting onto Search Console data and keep the tracker for competitor analysis.
• If your tracker recorded a collapse that didn't happen, correct the history manually or add an annotation. Otherwise you'll be explaining that dip on the chart all over again six months from now.

Nozzle's research is worth reading in full — it has a country breakdown and the full methodology. The conclusions are everyone's own: some will switch trackers, while I'm simply reminded once more that reporting belongs on the numbers Google measures itself.

Maxim Safianov

I work at the intersection of technical SEO and software engineering: helping sites stay visible for classic search and AI answers alike, and building the Python tooling that automates it.