If your SEO dashboard showed a cliff in late September, a strange recovery a day later, and a competitor you have never heard of sitting in your top three for a week, you were probably not hallucinating. Your rank tracker was.
On October 22 Ahrefs told customers that Google has been “poisoning” search results shown to suspected SERP data providers, and that between 5% and 20% of results are affected depending on the provider. Ahrefs is the biggest name so far to say it out loud, but it is late to the party. Nozzle, AccuRanker, Semrush, SE Ranking and DemandSphere have all described the same thing over the past two weeks. Google has said nothing publicly.
This is not a tooling story for the SEO team to sort out on its own. Every quarterly business review, every agency scorecard and every “we’re losing share of voice to them” slide built from third-party ranking data since mid-September now rests on evidence that is partly made up. Not by your vendor. By Google, apparently on purpose, to make scraping its results worthless.
What Google is actually doing
The most detailed account comes from Nozzle’s study, first published October 10 and updated October 19 with data through October 18. Nozzle looked at 27 million results pages collected since September 13 across 52 countries. On average, 8.4% came back poisoned, against a baseline of 0.17% before. The worst day was September 29, at 28.6%. Since October 14 it has run at 3% to 7% a day, and it was climbing again on October 17 and 18.
The important word is poisoned, not blocked. In Nozzle’s words: “Google is not refusing these requests; it answers them with a complete page whose results are wrong, and nothing on the page says so.” A blocked request tells a scraper to try again. A poisoned one gets stored as real. The page has the same template, the same layout and no error message. The only clue is that most of the usual sites are gone. On a poisoned page the organic results kept 22% of the previous day’s sites, against 88% normally.
For you, the number that matters is this one: when a project’s own site was one of a keyword’s usual results, it fell out of the top 10 on 60.2% of poisoned pages, against 2.7% in the baseline. That is a fake ranking crash, delivered at scale, and it looks exactly like a real one.
AccuRanker’s write-up, published October 19, tells a similar story with smaller numbers. On September 25, more than a quarter of its sampled keywords came back with a swapped page one (30.2% on desktop). Within two days it had that under 2%, and by October 17 under 1%. AccuRanker also found that Google seems to decide per browser session, not per keyword, so the damage follows how a vendor collects data rather than what you track. It found smaller rounds of the same pattern in its stored data going back to late January 2026, and says Google tightened again on October 15, mostly on mobile.
Semrush told Search Engine Roundtable it has seen inaccurate results “since late summer” and that the impact on tracked rankings has been “limited”, mostly beyond page one. SE Ranking said on October 16 that the issue still affected about 2% of tracked results and that it had corrected data for affected dates. Ahrefs says it is “cleaning up any past results that may have been affected.”
Why would Google do this? It hasn’t said. The working theory from Nozzle, Search Engine Roundtable and several SEO consultants is that the real targets are AI companies and SERP resellers bulk-collecting Google results to feed their own answers, with rank trackers caught in the blast. Google doesn’t owe anyone sympathy here. Its spam policies already say that “scraping results for rank-checking purposes” without express permission violates its rules. Every rank tracker you have ever paid for was operating on Google’s tolerance. That tolerance now has a price, and you pay it in bad data.
Why the numbers disagree, and why that’s the point
You will see people argue about whether the real figure is 1%, 8% or 20%. That argument misses what changed. Search Engine Watch laid the vendor figures side by side on October 19, and the spread is the story. Nozzle measured 8%, AccuRanker about a fifth of that under Nozzle’s own rule, SE Ranking 2%, and Ahrefs now says 5% to 20% depending on the provider. AccuRanker’s per-session finding explains a lot: two vendors tracking the same keyword can get different pages because they collect through different routes.
So the question for a marketing leader is not “what’s the true rate?” It’s “what’s the rate in the data my team and my agency actually used?” Most of you can’t answer that today. Your vendor probably hasn’t told you, and if it is quietly rewriting stored history, the chart your agency exported on October 5 may no longer match the same chart today.
Timing makes it worse. Google’s September spam update ran from September 24 to October 8, according to PPC Land, the same weeks poisoning peaked. Real demotions and fake ones landed on the same days. If someone on your team declared a site “hit by the update” from a tracker chart alone, that call deserves a second look. So does any victory lap about a competitor’s collapse.
Where it hits hardest
Nozzle’s breakdown is worth reading even if you never touch a rank tracker, because the damage isn’t spread evenly.
“Near me” searches were poisoned 12.8% of the time against 8.2% for everything else. Long-tail phrases searched fewer than 100 times a month ran at 10%. Brand searches were hit least, at 5.2%. If you run local, multi-location or long-tail content programs, which is most retail, home services, healthcare and franchise marketing, your data took more damage than a brand-led site’s.
By industry, legal results averaged 16.8% poisoned and hit 54.6% on September 27. Home services averaged 14.3%, weddings 12.4%, travel and hospitality 10.4%, finance 9.4%. Retail sat at 7%. By country, the US averaged 8.6%, Canada 11%, Australia 12.4% and the UK 3.6%.
The pages also change shape. Poisoned results showed AI Overviews less often (45.9% against 53.6%) and image packs and forums more often. Facebook showed up in the top 10 of 29.9% of poisoned pages against 21.6% of the same keywords’ normal ones. If your team reported a sudden surge of Reddit and Facebook into your category in early October, check whether it was a trend or a decoy.
Here is the detail I can’t stop thinking about. On poisoned pages, the parts advertisers pay for held steady. Nozzle found ads kept 83.7% of their previous advertisers, against 70.5% in the baseline, and shopping units kept 93.3%, essentially unchanged. The organic results got scrambled. The paid ones didn’t. Whatever Google’s motive, the effect is that the only parts of the page an outsider can still trust are the parts that are for sale.
What this breaks in marketing, not just SEO
Third-party rank data was never only an SEO tool. It’s how most companies see their competitors in search at all. Search Console shows your own site, and only your own site. Share of voice, competitor visibility, category landscape reports, content gap analysis and “who’s winning the AI Overview” tracking all come from tools that collect Google’s pages the same way rank trackers do. As PPC Land noted, Ahrefs’ SERP data also feeds keyword difficulty, traffic estimates and AI Overview tracking, and the company hasn’t said which of those products draw on the affected data.
That matters for decisions being made right now:
- Budget planning for 2027 is happening this month in a lot of companies. If the organic channel’s case rests on visibility charts from September and October, it rests on a contested window.
- Agency reviews. Plenty of SEO retainers are scored on ranking movement. An agency could be blamed for a crash that never happened, or credited for a recovery that was just Google switching the decoy off.
- Content and AI visibility programs. The new crop of AI search tracking tools also depends on collecting results at scale. If Google is willing to poison classic results to stop bulk collection, assume nothing about the stability of anything built on top of them.
- Competitive intelligence. A rival “disappearing” from your category for a week in late September is far more likely to be bad data than good news.
And Google’s own data doesn’t fill the hole. Search Engine Roundtable points out that Search Console gives no click or query data for generative results like AI Overviews and AI Mode. So the first-party view is incomplete, and the third-party view is contaminated. You are flying with one instrument that’s missing a needle and another that occasionally lies.
What to do this week
None of this needs a new tool. It needs a few decisions and a couple of awkward emails.
- Mark the window. Put an annotation on every SEO dashboard, client report and board deck covering September 13 to today: “third-party ranking data contested.” Anyone reading a chart from this period should see that flag before they see the line.
- Move your own-site reporting to Search Console. For your own rankings, use Search Console position, impressions and clicks by query as the record. Keep the third-party tracker for daily alerts, but treat any one-day drop that reverses within three days as suspect until Search Console agrees. Nozzle found 68.9% of poisoned results were back to normal within three days.
- Freeze big calls made on tracker drops. No budget cuts, agency firings, content rewrites or “we got hit by the spam update” post-mortems based on September or October tracker data alone. Recheck each one against Search Console and actual organic conversions first.
- Save what you have. Export the current state of the charts your team relies on and date them. If your vendor is cleaning stored history, you want to know what changed, and you can’t compare against a version you didn’t keep.
- Send your vendor five questions in writing. What share of results in my account was poisoned, by week, since September 13? How do you detect a poisoned page, and do you re-fetch it before storing it? Are you revising stored history, and will revised data points be flagged? Which of your other products (share of voice, keyword difficulty, traffic estimates, AI Overview tracking) use the same collection? What will you tell me the next time Google changes the rules?
- Report competitor and share-of-voice numbers as ranges. Until vendors publish their error rates, a single competitor visibility figure is false precision. Say “roughly 18% to 22%” rather than “20.4%,” and say why.
- Rewrite rank-based KPIs. If an agency or in-house target is tied to tracked positions, move it to Search Console metrics and organic revenue, starting with Q4. It’s fairer to everyone, including the agency that’s about to be blamed for Google’s anti-scraping war.
Search marketing has always run on a borrowed view of Google. For twenty years the borrowing was quiet and cheap. As of this fall, Google is deciding which outsiders get to see the real page, and so far the answer is: not the people measuring it. The teams that come through this well won’t be the ones with the best tracker. They’ll be the ones that knew which of their numbers were evidence and which were estimates, and said so out loud before someone made a budget decision on the wrong one.
