Some of the rankings in your tracker right now may describe results that no real searcher ever saw. That is the claim from Nozzle, a rank-tracking vendor, and two of the largest SEO data providers have since described a similar pattern in their own data. Google has not commented. You do not need to settle the cause to act sensibly: verify a movement in Search Console before you report it.
What was measured
Nozzle published its study on 10 October 2026 and updated it on 21 October, with data through that date. It judged 56.1 million Google results pages collected since 20 August, across 53 countries on desktop, iPhone and Android. Its headline rates come from two different slices of that data, so the label on each figure matters.
Its definition matters. A results page counts as “poisoned” when fewer than half of the sites that normally appear for that keyword show up anywhere on it. “Normally” means sites that were in the top ten on at least 60% of the keyword’s baseline pages, collected between 20 August and 9 September.
On that definition, Nozzle reports:
- Page one, all devices: 0.17% of results altered at baseline, 8.7% since 13 September (about 29.9 million results judged since that date). The all-device peak was 28.6% on 29 September.
- Desktop only, both pages of the same results judged (about 4.7 million): 0.28% at baseline, 22.6% on average since 13 September. Page one alone was 0.04% at baseline and 12.2% since.
- Desktop peak and swings: 43.5% on 30 September, 5.7% and 3.8% on 14 and 15 October, then 25.9% on 19 October and 35.7% on 21 October.
Its timeline is that requests to Google began failing in large numbers on 13 September, and that from 20 September the requests which got through began returning altered pages. When pages were altered, Nozzle found social platforms and forums rising to fill the space: Facebook appeared in the top ten on 29.5% of altered pages, against 21.9% on the same keywords’ other results pages since 13 September.
Who else says it, and who has not
Ahrefs wrote on 22 October that Google is increasingly showing different results to suspected SERP data providers than it shows regular users. It calls this industry-wide, puts the share affected at 5% to 20% depending on the provider, and says it has added filters that reject abnormal results and request replacements.
Semrush published a statement dated 27 October. It says some automated requests return pages that look normal but differ from what a typical user would see, and suggests the driver may be an effort to limit large-scale data collection for AI systems. It adds that Google has not commented publicly.
Search Engine Journal relayed Nozzle’s findings. We have found no independent confirmation of Nozzle’s figures.
That is the whole picture, and the limits are worth stating plainly. Nozzle sells rank tracking. Google has made no statement. The idea that this is Google’s response to AI scraping is Nozzle’s interpretation, and Semrush is also reporting a suggestion, not a confirmation. What is well supported is the narrower point: several providers see tracker data that looks unlike what users see.
Why you cannot see it by eye
An altered page is not an error. It looks like a normal results page. Nozzle’s argument is that a refusal prompts a scraper to try again, whereas a plausible wrong page is simply recorded as data.
The spells are also short. Search Engine Journal reports that 80% of altered spells lasted one reading, 15% lasted two and 5% lasted three or more. A keyword that drops hard on Tuesday and returns on Wednesday fits that pattern, and a monthly average will smooth it into something that looks like ordinary volatility. Nozzle also found page two altered about one and a half times as often as page one, so the “deep” positions are the first to doubt.
This is a separate disruption from the redirect change we covered in Google put a redirect between the click and your site. Both make third-party position data less trustworthy, for different reasons.
What we saw on our own domain
On 2 November we looked up doublebarril.com in three places. Semrush’s South Africa database listed 7 organic keywords. Ahrefs listed 0. Google Search Console showed 110 distinct queries for 4 to 31 October 2026.
We do not claim to know why they differ. It could be method, thresholds, how each tool treats low-volume terms, or altered results. The data cannot separate those explanations, and we would be inventing a story if we picked one.
We also want to be careful in the other direction. Our impressions fell from August, and that decline began before the 13 September start date Nozzle reports, so we are not attributing it to altered results. What we take from the comparison is narrower: when tools disagree by an order of magnitude on the same site, the one that records Google’s own count of impressions and clicks for real searches is the one to report from.
The South Africa figure
Nozzle’s country table, which covers 45 countries, puts South Africa at the top, at 22.4% of judged results altered since 13 September, against 8.8% for the United States and 4.1% for the United Kingdom. Two cautions. The South African figure rests on 71,148 judged results, about 258 times fewer than the 18,383,115 behind the US figure. And the table counts a page as altered on the any-page definition above; it does not state which device it covers.
It is a reason to check South African reports, not a measured loss. If a rank report moved sharply in September or October, compare it with Search Console before it reaches a client.
What to do this week
- Verify before you report. Use our 45-minute method: compare against the keyword’s usual history, check page two, wait one more reading, then match the movement in Search Console.
- Ask your provider in writing how it detects altered results, what changed in collection after 13 September, and whether history was recalculated.
- Re-baseline and annotate your charts at 13 September 2026, when Nozzle says requests began failing, and 20 September, when altered pages first came back. Comparisons that cross either date may mix two collection regimes.
- Add a line to client reports stating that third-party positions are indicative for the period and that Search Console is the reference.
- Do not act on a one-day drop. Rewriting content or moving budget because of a single reading is the expensive way to be wrong.