Google is deliberately serving fake results pages to suspected scrapers. It isn’t blocking them. It answers them, with complete, normal-looking pages whose organic results have been swapped out. Nothing on the page gives it away.
That’s the finding of Nozzle’s study, published 10 October 2026 and updated through 21 October. 29.9 million SERPs judged since 13 September. 8.7% of them came back poisoned on page 1, against 0.17% before. Count page 2 as well and it’s 22.6%, peaking at 43.5% on 30 September.
Savage, but not at all surprising. And it settles something: a live SERP you load yourself, plus your own Search Console data, are fast becoming the only sources of truth in SEO. Everything in between now needs checking.
What actually happened
Nozzle calls a page poisoned when fewer than half of the keyword’s usual sites appear anywhere on it (“usual” meaning in its top 10 on at least 60% of SERPs over a three-week baseline). On poisoned pages the organic results keep just 22.3% of the previous SERP’s sites, against 88.1% normally. Social platforms and forums fill the gap: Facebook up 7.6 points, Instagram 4.3, YouTube 3.2, Quora 2.6, Reddit 2.3.
| Date | What happened |
|---|---|
| 22 Oct 2026 | Ahrefs says 5–20% of results are affected, depending on the data provider (via Schwartz) |
| 21 Oct 2026 | Poisoning climbs back to 35.7% of SERPs judged on both pages (Nozzle) |
| 16 Oct 2026 | SE Ranking: fixes in, roughly 2% of tracked results still affected (via Schwartz) |
| 12 Oct 2026 | Barry Schwartz reports third-party rank tracking is struggling industry-wide (SER) |
| 10 Oct 2026 | Nozzle publishes Poisoned Google SERPs |
| 8 Oct 2026 | Lily Ray: rank trackers too unreliable right now to share findings from (via Schwartz) |
| 30 Sep 2026 | Peak: 43.5% of SERPs poisoned on page 1 or 2 (Nozzle) |
| 20 Sep 2026 | First poisoned SERPs appear (Nozzle) |
| 13 Sep 2026 | Tracker requests to Google start failing at scale (Nozzle) |
| Jul 2026 | Google tests goto redirects in results to identify scrapers (via Schwartz) |
| Sep 2025 | Google kills the num=100 parameter; bulk SERP collection gets 10× dearer (via Schwartz) |
Semrush told Schwartz it had seen the issue since late summer. Ahrefs is cleaning poisoned results out of its historical data. Google has said nothing.
Why poisoning beats blocking
Blocking a scraper gives the game away. A 429 or a CAPTCHA tells the operator they’ve been spotted, so they rotate IPs and go again. Eventually the scraper wins, and the cost of the whole arms race sits with Google.
A poisoned page does the opposite. The request succeeds, the page looks fine, and the fake gets stored as real, then sold to clients or piped into an LLM’s grounding data. Now the cost sits with the scraper, who has to verify everything. A block invites a retry. A poison gets archived as truth.
Google doesn’t even need to poison every page. At 8.7% fake, all of the data is suspect. Corrupt a slice and you devalue the lot.
Two details give the strategy away. Shopping results hold at 93% retention on poisoned pages, and ads actually keep more than baseline (70.2% up to 83.9%). The data advertisers pay to measure is untouched; only what other people resell gets corrupted. And the target isn’t SEOs. Nozzle reads it as Google’s response to AI scraping and grounding, and Schwartz agrees: this is aimed at AI bots and SerpApi, with rank trackers as collateral. Any LLM grounding its answers on scraped Google results is now grounding on pages Google may have salted.
The logic holds up. Google’s SERPs feed its competitors’ answers for free, blocking has failed for a decade, and poisoning makes the stolen asset worth less than the cost of stealing it. You don’t have to win the technical arms race if the prize is worthless.
What this breaks
Everything downstream of scraped SERPs:
- Trackers disagree with reality, and with each other. Where a project’s own site was tracked, it fell out of the top 10 on 60.8% of poisoned SERPs, against 2.7% at baseline (Nozzle). Glenn Gabe documented confirmed drops that Ahrefs caught, Semrush caught late and Sistrix missed entirely (via Schwartz, Sep–Oct 2026).
- Phantom crashes. A tracked site vanishes for a day then comes straight back: 67.9% of poisoned SERPs had the usual sites back within three days, and four in five poisoning spells lasted a single capture. If you don’t know to check, that reads as a penalty or an unconfirmed update.
- Contaminated history. The fakes sat in databases as real for weeks before anyone caught on. Ahrefs says it is cleaning affected past results, which means every dataset built on scraped SERPs since mid-September carries an unknown error rate.
- AI answers inherit it. Any pipeline grounding on scraped Google results has been ingesting swapped pages since 20 September.
One UK note: we ran lowest in Nozzle’s country table, 4.1% of SERPs poisoned against 8.8% in the US. Lower is not clean, though. Page 2 is hit one and a half times as often as page 1, usually under a page 1 that looks fine.
What’s left: live SERPs and Search Console
Two data sources can’t be poisoned, because neither is scraped.
A SERP you load yourself is, by definition, what Google shows a human. The poisoning only fires at suspected bots. The old limits still apply (personalisation, location, a single snapshot in time) but the result is real.
Search Console is Google’s own record of impressions and clicks on real searches, and it happens to be the cleanest poisoning test there is. Nozzle’s own advice: if a site “disappears” in your tracker while GSC still shows impressions and clicks, nothing happened to the site. GSC has gaps of its own. Positions are averaged, and as Schwartz points out, there’s still no click or query data for AI Overviews or AI Mode. Google’s countermeasures end exactly where its own reporting begins.
Which is the uncomfortable part. Kill third-party measurement and Google becomes the sole auditor of its own results. Motive or side effect, the outcome is the same: the referee now keeps the score as well.
What to do before you report a ranking drop
- Check GSC first. If impressions and clicks are steady on the dates in question, the drop didn’t happen.
- Compare against the keyword’s history, not yesterday. Yesterday’s capture may have been poisoned too.
- Wait a capture. Four in five poisoning spells are a single SERP. A real change stays.
- Load the SERP yourself in a clean browser from the right location, and screenshot it with the date on.
- Treat tracker data as directional: useful for trends across thousands of keywords, no longer proof of any single movement.
And date every manual check. If this episode teaches anything, it’s that undated ranking data can’t be audited.
The prompt to run before you panic
Copy this into ChatGPT, Perplexity, Claude or Grok before acting on any ranking drop:
My rank tracker shows [site] dropping for [keyword] on [date]. Before I treat it as real, help me rule out a poisoned SERP (Google serving fake results to scrapers, documented by Nozzle, Sept–Oct 2026).
Walk me through, one step at a time:
1. What my Google Search Console impressions and clicks for that page should look like on those dates if the drop is fake vs real.
2. How to compare today's top 10 against the keyword's usual top 10 over the past 3 weeks, not just yesterday.
3. What a one-capture anomaly vs a persistent change tells me (most poisoning spells last a single capture).
4. How to run and document a clean manual SERP check for my target location.
Then give me a one-line verdict format I can log: date, keyword, poisoned/real/unclear, evidence.Sources
- Poisoned Google SERPs, Derek Perkins, Nozzle, 10 October 2026, updated 21 October 2026. All poisoning figures above.
- Third-Party Google Rank Tracking Struggling Now, Barry Schwartz, Search Engine Roundtable, 12 October 2026, updated 22 October 2026. The vendor statements (Ahrefs, Semrush, SE Ranking), practitioner reports, and the num=100 and goto-redirect background.
All figures checked against the sources on 23 October 2026. Google had made no public comment in either source as of that date.