We get asked this a lot.
“Can Google tell if my content is AI-written?”
Short answer: no, not the way you’d imagine. There’s no AI-detector sitting inside Google’s ranking system, flagging pages as “written by ChatGPT” and quietly burying them. Google’s line has been the same for years; it doesn’t matter how content gets made, only whether it’s actually helpful. What Google is very good at catching is the pattern that scaled AI production tends to leave behind. And that’s a different problem entirely.
Here’s what’s going on.
Google isn’t detecting AI. It’s detecting thin.
AI was never the target. The target is what happens when someone uses AI to publish 400 near-identical pages instead of one good one. Thin content, near-duplicate structure, pages that technically answer a query but say nothing that the ten pages linked to it didn’t already say, that’s what is going to get punished. The AI is just the tool that made doing this at scale so easy and quick.
John Mueller said as much last November, and said it bluntly. Site owners were rewriting their AI content by hand after a traffic drop, hoping that would fix things. His response: rewriting it doesn’t make it authentic. His actual advice was tougher than that – treat it like starting over with nothing, and rebuild around real value. Not a find-and-replace pass to sound more human.
The case studies are out, and they’re not subtle
The best indication we’ve had of this landed with Google’s August 2026 spam update. Glenn Gabe (an authority in the new SEO/GEO game) pulled apart the damage and gave the pattern a name I quite like: “Mt. AI.” , ie, sites that scaled AI content, often mixed in with programmatic pages until they were basically running a landfill of near-duplicate URLs.
To back up his claims, Gabe highlight to instances where Mt. AI got spanked:
- One YMYL (Your Money or Your Life) site lost rankings across 200,000+ queries.
- A thin affiliate site running almost entirely on programmatic, AI-filled pages lost 14,000 queries.
In both cases the content scored 97%+ as AI-generated on detection tools. But it wasn’t the AI origin that did it; it was the scale and the thinness.
If you’re publishing one or two genuinely useful AI-assisted articles a month, none of this applies to you. If you’re running a content farm pumping out hundreds of thin AI pages with no editorial eye on them, this is exactly what’s coming for you.
The GEO gossip worth knowing about
This is the bit I find particularly interesting, because a whole industry sprang up around it. “GEO” (Generative Engine Optimisation) tactics aimed purely at getting cited inside ChatGPT and AI Overviews.
Lily Ray spent early 2026 pulling this apart and it’s worth saying again: a lot of GEO advice is quietly wrecking the SEO foundation it depends on.
Her logic, once you see it, is obvious. Most AI answer engines are still retrieving from Google’s index behind the scenes. If you’re not ranking organically, you can’t get pulled into the model’s answer in the first place. So when a brand claims their shiny new GEO campaign is “working,” Ray’s take is that the causality usually runs backwards, ie, years of strong backlinks and organic authority (real SEO) got them cited, and the GEO campaign is just along for the ride.
A few tactics she flagged as clever-looking traps:
- Building dozens of self-promotional “best X” comparison pages (one site built 51 of them, then watched both organic traffic and AI citations drop at the same time),
- refreshing timestamps with no real content change; and my favourite bit of gossip from this year…
- Microsoft flagging “summarise with AI” buttons that quietly inject brand-favourable instructions as a security risk, not a growth hack.
Google’s John Mueller weighed in on the SEO vs. GEO debate too, and I appreciated how deflationary he was about it: “What you call it doesn’t matter, but AI is not going away.” His advice wasn’t to pick a side. It was to actually look at your own traffic numbers, work out how much is coming from AI tools versus classic search, and allocate your effort accordingly, instead of chasing whichever acronym is trending this month.
Where Google actually can see AI – and it’s not your writing
Here’s the twist most people miss. Google can detect AI now – just not in your text. At I/O 2026 they rolled out the ability to ask “Is this made with AI?” straight inside Search, Lens and Circle to Search, powered by SynthID (an invisible watermark baked into images, video and audio) and C2PA credentials (signed metadata documenting where a file came from). Over 100 billion images and videos have been watermarked this way already.
Worth saying clearly so nobody walks away with the wrong idea: this is media provenance, not a text penalty. It flags images, video and audio – not your written copy – and it isn’t a ranking signal on its own. But the “nobody can prove this photo is AI” era is closing fast and that matters if your content leans on stock-style AI imagery.
So what do you actually do with this?
Boiled right down: Google doesn’t detect AI, it detects thin, unhelpful, or manipulative content – at whatever scale it shows up. Genuinely useful AI-assisted content, edited by someone who actually knows the subject, is fine (as far the guidance Google’s given so far). Content built purely to scale a pattern – hundreds of near-identical pages, self-serving comparison posts, GEO tricks with no organic foundation under them – is exactly what the last two spam updates went after. And now, on the image side, watermarking makes your visuals traceable too.
Sources referenced: