01
Noise Watch
SKIP IT
A research paper sold as a ranking change. Nothing for you to do.
No, Google didn’t just switch on an AI-content detector
If you were on LinkedIn the week of September 24, you probably saw some version of this: Google has a new system that catches AI-written content, and it’s behind the latest spam update. If your team uses AI anywhere in its writing process, that’s an unsettling thing to read, but most of it doesn’t hold up.
Google did start a spam update on September 24. That part’s real, and it’s still rolling out. It’s the fourth one this year, and it’s slower than the others: the last three each wrapped up in a couple of days, and John Mueller confirmed this one should take longer. Google’s own description, posted on LinkedIn, was “a normal spam update.” It didn’t name a target.
The AI-detector part arrived a day later, from a Google research paper describing a system called SAFE. Search Engine Journal wrote it up under a headline saying Google had deployed it, and suggested it might be part of the update. Read past the headline and it gets a lot thinner. The paper is mostly about video, specifically networks of mass-produced AI videos on YouTube. PPC Land’s read is that nothing in the document ties it to Search. Glenn Gabe pointed out that it’s dated late May, months before this update. And Google hasn’t connected the two anywhere.
To be clear, Google isn’t relaxed about AI spam. The rule it enforces is called scaled content abuse, and it’s been in Google’s spam policies for years. It’s about publishing piles of pages that don’t help anyone, regardless of who or what wrote them, so a team putting out two thoughtful posts a month with some AI help in the process is not the target.
Google’s John Mueller described where the line actually sits earlier this month, talking about a site built by generating pages from every combination in a spreadsheet. He said that kind of programmatic SEO often ends up flagged as spam, borderline spam, or just low quality, that Google’s systems can lose faith in the whole site because of it, and that earning that back takes time and real effort. That’s the risk to look for on your own site. Whether a person or an AI wrote the words doesn’t come into it.
my read
This is the second month in a row I’ve written some version of this, and I expect to keep writing it. Every spam update now arrives with a story about Google catching AI content, because that’s the version that gets shared. If someone on your team wants to start running every draft through an AI detector, I’d push back. Those tools are unreliable, and honestly they’re measuring the wrong thing. What gets sites into trouble is publishing pages at scale mainly to rank, and a detector can’t tell you whether you’re doing that.
In Plain Terms
Spam update
An upgrade to the automated systems that Google uses to catch pages breaking its spam rules. It only goes after rule-breaking, unlike a core update, which re-ranks everything (though you can still see some collateral wobble while it rolls out).
Scaled content abuse
Google’s name for churning out lots of pages mainly to rank rather than to help anyone. Google doesn’t care whether a person, a template, or an AI wrote them. It cares that there are hundreds of useless pages that are published in attempt to ‘game’ the system.
Programmatic SEO
Building lots of pages automatically from a template and a spreadsheet, like one page per city per service. Done well, each page contains information that’s genuinely specific to that location or service. Done badly, it’s the same page a thousand times with the city name swapped.
Sources: Official: Google Search Status Dashboard incident for the September 2026 spam update · Google Search Central’s LinkedIn announcement · Google’s spam policies · John Mueller on Bluesky, on the rollout length and separately on programmatic SEO. Press: Search Engine Roundtable’s coverage of the update and Mueller’s programmatic SEO comments · PPC Land on what the SAFE paper does and doesn’t say. Practitioner: Glenn Gabe on the paper’s date. Search Engine Journal’s original write-up is named but deliberately not linked, since it’s the claim being questioned.
02
What Actually Changed
SIGNAL
A real shift, and a good reason to revisit your FAQ content.
Google now answers almost every “People also ask” question itself
You know the People Also Ask box: the stack of related questions partway down a Google results page that open up when you click them. For years, each answer was a snippet lifted from somebody’s website, with a link back. Getting your page into “position zero” was an SEO goal, and plenty of FAQ pages were written specifically to win it.
That box has quietly changed hands. Fourteen months ago, about 12% of those answers came from Google’s AI. This month, Mark Williams-Cook checked a sample of 19.2 million English-language questions from his company’s tool, AlsoAsked, and found 97% of the answers in the first week of September were AI Overviews, up from 86% in August. One tool vendor rounded it up to 100%. It isn’t quite there, but it’s heading that way. The questions are still on the page; the answers are increasingly Google’s own, stitched together from a few sites at once.
A lot of FAQ content on business websites was built by copying the questions straight out of that box, or out of a keyword tool, and answering each one in a paragraph. That’s exactly the content Google can now summarize on its own page without anyone clicking. I made this argument earlier this year in SEO Isn’t Dead: the FAQs that hold up are built from what your prospects actually ask your team, in their own words. September’s data makes that a lot less theoretical.
my read
I wouldn’t delete anything this week. But if your FAQ page came out of a keyword tool, it’s due for a rewrite. Ask your sales team for the 10 questions they answered most last quarter. Those are the questions to answer, because the authoritative answer lives inside your business rather than in a keyword database.
In Plain Terms
People Also Ask
The box of related questions that shows up partway down many Google results pages. Click one and it expands into a short answer. Some SEOs call it “PAA,” if you hear that in a meeting.
AI Overview
The AI-written summary Google shows at the top of many results pages, and now inside the People Also Ask box, too. It pulls from several websites at once and only sometimes links to them.
Sources: Data: AlsoAsked‘s 19.2M-question sample, shared by Mark Williams-Cook on LinkedIn. Press: Search Engine Roundtable’s September write-up and its earlier coverage of the 12% figure. The “100%” claim came from a separate tool vendor and is mentioned only to show the headline ran slightly ahead of the data.
03
What’s New
KEEP AN EYE ON IT
Early tests pointing in two directions. Understand it, don’t act on it.
AI Overviews are linking out more. Some of those links don’t go to websites.
Two findings landed the same week and point in opposite directions. On September 23, Malte Landwehr shared data from his company, Peec AI, showing the share of AI Overviews that link to outside websites jumping from close to zero to over 25%. Barry Schwartz, who reported it, said it didn’t feel right to him and disclosed a small investment in Peec. So it’s one vendor’s chart, and even the reporter wasn’t sold on it.
Two days earlier, Gagan Ghotra spotted Google testing ordinary-looking links at the bottom of AI Overviews that don’t lead to a website at all. They open AI Mode with a follow-up question already typed in (Barry’s write-up has the screen recording). They aren’t necessarily contradictory. Google can add more citations to AI Overviews while also testing ways to keep the next search inside Google. You’ll probably hear one of these quoted in a meeting without the other.
The same question has a new lever attached. Since August 31, every site has a Search Console setting that takes it out of AI Overviews, AI Mode and AI Overviews in Discover, both as a cited link and as a source the answer is built from. It’s one switch for the whole site. Google says opting out won’t affect your regular rankings, but you’d get no traffic or impressions from those AI features. And there are still open questions about it, such as whether news results shown inside an AI Overview count, which Google’s help page doesn’t answer yet.
my read
I’m not viewing either of these as a trend yet. One’s a single dataset, and the AI Mode links are a test that Google could pull next week. What they do show is where the real decisions are happening now: inside the AI answer, before anyone gets near your site. That’s the case for being a source those answers lean on, which puts us right back at the basics.
In Plain Terms
AI Mode
Google’s chat-style search, where you have a back-and-forth conversation instead of scrolling a list of links. Each follow-up you type counts as another search, and you never leave Google to do it.
Sources: Official: Google’s announcement of the Search Console AI control. Data: Peec AI, via Malte Landwehr on X (single vendor; the reporter disclosed an investment). Practitioner: Gagan Ghotra on X. Press: Search Engine Roundtable’s coverage of the external links data and the AI Mode links test · Search Engine Journal on what the help page doesn’t yet cover.
04
The Numbers That Matter
The top two are the numbers making people anxious this month. The bottom two put them in proportion.
97 %
of “People Also Ask” answers are now written by Google’s AI
+ 101 %
growth in referral traffic from ChatGPT, January to August
~ 88 %
of referral traffic across the web still comes from Google
~ 1 %
is ChatGPT’s share of that same traffic, in its biggest month so far
ChatGPT’s growth is real, and it’s growth from a small base. Doubling about 1% of the web’s referrals is good news that’s still tiny next to Google. The bigger shift is the first number: the answers are moving onto Google’s own page.
A NOTE ON HONESTY
You’ll see BrightEdge’s headline that ChatGPT now sends 95% of AI referral traffic. That’s true, but it’s a share of AI referrals only, which is why it sounds so much bigger than the 1% above. And Cloudflare’s numbers describe traffic across its own network, so your split will be different. Check your own analytics before you quote either one to anyone.
Sources: Data: AlsoAsked, via Mark Williams-Cook · BrightEdge’s September 24 release · Cloudflare Radar referral data for July 2026, the most recent full month published at the time of writing. Official: Google’s generative AI performance report announcement, for the report referenced in action 4.
05
What This Means For You
#1
Mark September 24 in your analytics, then wait.
The spam update is still rolling out and could run into the second week of October. Add a note in GA4 for September 24 and don’t judge anything until Google’s status dashboard says it’s finished. If you weren’t publishing low-value content at scale, don’t assume the update is about you.
TELL YOUR BOSS
There’s a Google spam update finishing up. We’ve marked the date and we’ll review once it’s done. No changes before then.
#2
Rebuild your FAQs from real conversations.
Ask sales or customer service for the questions they heard most last quarter, in the customer’s own words. Swap out any keyword-tool questions for those.
TELL YOUR BOSS
Google answers the generic questions itself now. Our FAQ should answer the ones only our customers ask us.
#3
Look hard at any pages you made from a template.
City pages or service-plus-location pages built from a spreadsheet are the usual suspects, and they’re exactly what Mueller was describing in Section 01. If those pages don’t each say something different and useful, merge or improve them.
TELL YOUR BOSS
Some of our templated pages could drag down the whole site. I’d like time to consolidate the weak ones.
#4
Check your AI impressions before anyone touches the opt-out.
Someone will eventually ask whether to flip the switch described in Section 03. Start with the generative AI performance report, which sits under Performance in Search Console. It shows how often your pages appear in AI Overviews and AI Mode, and which pages. It doesn’t show clicks or the searches behind them, so it can tell you how visible you are there but not exactly what traffic you’d lose. For most teams reading this, the AI answers are where prospects are increasingly doing their research, so I’d leave the switch on unless you have a specific reason not to.
TELL YOUR BOSS
We’re staying visible in Google’s AI answers for now. If we ever consider opting out, we’ll look at how often we show up there first.
