Coalition has long suspected Google can fingerprint AI-generated text. And we anticipated that Google’s ability to identify AI generated text would soon equate to Google’s ability to devalue or penalize AI generated text.
We also expected a delay whenever a new model landed, followed by detection catching up. Google’s new research gives that suspicion a lot more weight. The paper describes a deployed system that finds synthetic content through text embeddings and repeated narrative patterns. It also groups related accounts using infrastructure signals.
“I’ve long argued that Google has a low tolerance for AI generated content, and will quietly work to remove its contribution to rankings when it can detect it. This research report may be a brief glimpse of the tools and methods the search giant has already at its disposal.”
Jordan Brannon, President of Coalition Technologies
A huge chunk of the SEO industry spent the last few years saying Google didn’t care how content was made. People took Google’s public position that generative AI can assist content creation and turned it into permission for mass publishing. For SEOs, who usually could care less about how well they write, AI seemed like a silver bullet. And since it did take a minute for Google to identify AI content, it seemed like a win for SEO agencies used to losing clients in 3 months anyways.
But Coalition disagreed. And we got downvoted for saying that AI content, improperly humanized and edited, would become a problem.
Google’s own guidance warns that generating lots of pages without adding value can violate its scaled content abuse policy. “Google doesn’t care” was always a ridiculous reading.
The language in the research is even harder to ignore. Its authors repeatedly call the material “AI slop” and “synthetic spam.” They describe an adversarial flood built to overwhelm quality filters and justify terminating whole clusters.
Google’s research page says the deployed system terminated 50,000 clusters covering 130,000 channels over six months. It can also adapt its detection system when attackers switch to newer generative models.
Search Engine Journal’s article focused on the text signals: Sentence-BERT, embeddings, salient terms and templated narratives. The paper covers abuse on a major video platform, so nobody should pretend it documents Google Search’s ranking stack. The idea that Google lacks the ability to detect AI text looks dead.
We’ve also learned not to treat every Google statement as a complete technical disclosure. Google can say it rewards useful content regardless of how it was produced, then broadly discount synthetic signals because the web is being buried in cheap crap.
Google has no obligation to review each AI page with care.
We expect Google to reduce how much AI-generated copy contributes to a URL’s ranking without erasing every other signal. Strong links and brand demand can still carry a page. User behavior can help too.
Large sites will point to AI pages that rank and declare victory. Smaller sites using the same tactic will struggle to hold visibility because the content itself adds little weight.
Human-led content still produces real gains. Coalition Technologies’ Big Time Rush Limited case study reports that fresh, optimized blog copy helped increase new organic users by 50.82% and organic traffic by 36.36% in one month.
Use AI for research and structure. Let it handle boring production work too. But publishing thousands of generated pages because Google supposedly can’t tell is stupid.
Our AI Content Detector gives teams another check before that gamble reaches the site. And if you’d like a more human touch- check out our work in GEO (AI SEO).
Would you build a long-term search strategy around Google’s inability to recognize the cheapest content ever produced?
