01
Noise Watch
SKIP THE PANIC
A measurement problem, not a ranking problem. Check before you react.
Your rank tracker may have been lying to you since September
If your SEO report showed rankings falling off a cliff this fall, there’s a decent chance some of that never happened. From mid-September, Google started handing fake results to the automated rank trackers that check where your pages show up. Google hasn’t said a word about it, but more and more tool companies have.
The best evidence comes from Nozzle, a rank-tracking company, whose study looked at 27.5 million results pages across 49 countries. Before September 13, about 0.17% came back wrong. After that date, 8% did, and on the worst days in late September it was close to a quarter. Ahrefs put the range at 5% to 20% depending on the data supplier, and Semrush confirmed it on October 27, calling it industry-wide. The likeliest explanation is that Google is trying to stop AI companies scraping its results, and the SEO tools got caught in the same net.
Not every drop was fake, though. Google’s September spam update ran from September 24 to October 8, right through the same weeks, and it hit some sites for real. Then came four rounds of “something’s changing” chatter later in October, all tracked by Search Engine Roundtable and none confirmed by Google. As of month end, the spam update was still the last ranking change on Google’s own dashboard.
The fake results have a tell. Your site gets swapped out for a page heavy on Reddit, Facebook, and forums, and most of it snaps back within a few days. Meanwhile Search Console, which gets its numbers straight from Google instead of scraping, doesn’t show the drop at all.
my read
I’d be wary of any rank tracker chart from mid-September onward. If it says you fell, open Search Console before you open a new project. That’s Google’s own record of where you showed up and who clicked. And if someone’s already pitching you a recovery plan off one of those charts, ask them what Search Console says first.
In Plain Terms
Rank tracker
A tool (Semrush, Ahrefs, and plenty of others) that runs searches on your behalf on a set schedule and records where your pages land. It’s checking Google from the outside, which is exactly why it can be fooled.
Scraping
Using software to load web pages automatically and copy what’s on them. Rank trackers scrape Google’s results. So do a lot of AI companies, which is the part Google seems to be fighting.
Search Console
Google’s free reporting tool for your site. It shows the searches you appeared for and the clicks you got, straight from Google, so nothing in between can fake it.
Sources: Official: Google Search Status Dashboard. Data: Nozzle, Poisoned Google SERPs (a rank-tracking vendor’s own research) · Ahrefs, October 22 · Semrush, October 27. Press: Search Engine Land · Search Engine Roundtable.
02
What Actually Changed
SIGNAL
Not new in principle, but now in writing. A quick check of your process, not a rebuild.
Google wrote down how it wants you to use AI in your content
Last month the rumour was that Google had switched on an AI-content detector. It hadn’t. This month it did something more useful and updated two of its official guides to spell out what it actually expects.
On October 1, Google’s guide to using generative AI content was updated to say that it’s “critical to manually factcheck and review all AI-generated content” before it goes live. That covers more than the article itself. As Search Engine Journal pointed out, page titles, meta descriptions, structured data, and image alt text all fall under the same check, and that’s the stuff teams most often let a tool write unattended.
Google’s guide to helpful content picked up a new warning at the same time: inventing an author is deception. A made-up writer with an AI-generated photo or fake credentials makes a page untrustworthy, in Google’s words, to its automated quality systems as well as to readers. Google’s quality raters have been told this for a while. Now it’s in the guidance written for site owners.
None of this is a penalty or a new update, and Google still says AI-assisted content is fine. The point is that a person has to stand behind it. John Mueller said much the same in an interview at month end: the quick tricks people use to show up in AI answers get caught eventually, so don’t build on them.
my read
Using AI to help draft is fine, and lots of good teams do it. What gets you in trouble is publishing something nobody checked, under a name nobody owns. Google has now said that plainly. If your blog still posts as “Admin” or a made-up persona, fix that before anything else on your list.
In Plain Terms
Meta description
The short summary under your page’s title in Google’s results. Visitors never see it on your site, which is why it’s easy to let a tool fill it in and forget it.
Structured data
Behind-the-scenes code that labels what’s on a page (this is a price, this is an event date, this is the author) so search engines can read it reliably.
Alt text
A written description of an image, read aloud by screen readers and used by search engines to understand the picture.
Quality raters
People Google hires to grade search results against a long rulebook. They don’t change rankings directly. Their grades help Google judge whether its systems are working.
Sources: Official: Google Search Central documentation changelog (October 1 entry). Press: Search Engine Journal on AI fact-checking · Search Engine Journal on fake authors · Search Engine Roundtable on Mueller’s interview.
03
What’s New
KEEP AN EYE ON IT
Genuinely new, and not live until 2027. Understand it, don’t act on it.
A second AI switch is coming, and it does something different
On October 30, Google said that it will support a new signal that lets websites tell AI companies not to train on their content. Microsoft, OpenAI, and Anthropic made the same commitment. It lives in your site’s robots.txt file, and Google expects to start honouring it in early 2027. Nate Hake, who’s been part of the standards discussions, cautioned that it isn’t final yet and disagreements remain.
Here’s why it matters. Since August, Search Console has had a setting that takes your site out of Google’s AI Overviews and AI Mode. That one decides whether you appear in AI answers. The new one decides whether your content gets used to build the AI models in the first place. Two separate choices, and per Google the Search Console setting has no effect on training.
my read
Nothing to do yet. I’m including it because these two switches are going to get mixed up in a lot of meetings next year. If someone says “let’s block AI,” the first question is: block it from what? Blocking training can be a reasonable call. Blocking the AI answers usually isn’t, and the numbers below show why.
In Plain Terms
Training
The stage where an AI model learns by reading huge amounts of text, possibly including yours. Different from an AI answer looking up your page today to quote it.
robots.txt
A small text file every website can have that tells automated visitors what they’re allowed to do. Your developer can change it in a few minutes.
Sources: Official: Google’s common crawlers documentation, via Search Engine Roundtable, which also carries Nate Hake’s caveat · Microsoft’s matching announcement, same day.
04
The Numbers That Matter
In keeping with the theme: three numbers that sound worse than they are, and one that sounds boring and isn’t. Read the labels closely. That’s where the story is.
80 %+
of Google searches now show an AI Overview, up from about 50% in mid-2025
8 %
of first-page results sent to rank trackers were fake, Sept 13 to Oct 9
0.9 %
estimated click rate on AI Overview and AI Mode appearances, from one site’s test
+ 10.5 %
growth in Google’s search ad revenue last quarter, to $82 billion
The 0.9% looks tiny. But when that site left the AI answers, its regular clicks dropped too, and on its most AI-visible pages they came back once it opted back in. Being mentioned does work that your referral report can’t see. And Google’s search business is still growing, so people are still searching there.
A NOTE ON HONESTY
The 0.9% comes from one site, run by a respected SEO who called it a small case study, and new posts plus brand searches lifted its rankings during the test. This muddies things. I’d treat it as a strong hint, not a rule. The 8% is Nozzle’s original first-page count. Their updated study adds page two and runs to October 21, which pushes the average to 22.6%. And Nozzle sells rank tracking, so it has skin in the game. The revenue number says Google is doing fine. It doesn’t say your traffic is.
Sources: Data: AI Overview coverage data from Semrush, SE Ranking and seoClarity · Nozzle’s study · Cyrus Shepard’s opt-out test, with follow-up detail from Search Engine Roundtable and Search Engine Land · Alphabet’s Q3 2026 results, via Search Engine Roundtable.
05
What This Means For You
#1
Check Search Console before you react to a rank report.
For anything from mid-September onward, treat a rank tracker drop as a question, not an answer. If Search Console shows steady clicks and impressions for the same pages, the drop probably wasn’t real. If your tool vendor has cleaned up its past data, re-export before the next report goes out.
TELL YOUR BOSS
Some of the ranking drops in our reports since September came from bad tool data, not real changes. I’m checking everything against Google’s own numbers before we act on it.
#2
Put a real name and a real check on every piece.
Make sure whoever reviews AI-assisted drafts also reviews the titles, meta descriptions, and alt text, since those are now named in Google’s guidance. Then look at your bylines. Every author should be a real person with a short bio.
TELL YOUR BOSS
Google now says in writing that AI-assisted content needs a human check and a real author behind it. We’re confirming every piece has both.
#3
Leave the AI opt-out switched off.
It’s tempting when AI answers send so few clicks. But the one real test we have says opting out cost regular clicks as well. If you want to limit AI training, that’s a separate decision coming in 2027.
TELL YOUR BOSS
Opting out of Google’s AI answers looks appealing, but early evidence says it costs us regular clicks too. We’re staying in.
#4
Make your official pages obviously official.
On October 15, Google said it’s testing a spot at the top of some AI answers for the one or two official pages that complete a task. It’s only a test. But the pages that would win it are ones you already own: booking, pricing, support. Make sure each one is clearly yours and says plainly what it’s for.
TELL YOUR BOSS
Google is testing a slot in AI answers for official pages. We’re tightening up our key task pages so they’re the obvious choice if it rolls out.
