Google says GEO is just SEO, and the mythbusting list is the real news
The optimization debate gets less interesting. The list of tactics Google explicitly tells you to stop doing gets more interesting.
Photo by mockupbee on Unsplash
Google put out a new documentation page called "Optimizing your website for generative AI features on Google Search," and per Search Engine Journal, the headline-grabbing line is that "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Fine. That part is a rhetorical move. The part worth reading carefully is the Mythbusting section, where Google names specific tactics by name and tells site owners to stop bothering.
the mythbusting list
txt files, chunking content into small pieces, rewriting copy in some "AI-friendly" way, and chasing inauthentic mentions across blogs and forums.
That list is unusually specific for Google. It's also a direct shot at a chunk of the GEO consulting market that has been selling llms.txt setup and content chunking as table-stakes work for the past year. Danny Sullivan apparently told Search Engine Journal he'd talked to Google engineers who recommended against chunking, in January 2026.
None of this means llms.txt is useless everywhere. Anthropic still publishes one. So does Cloudflare. The honest read is that llms.txt may matter for some agent and retrieval contexts that aren't Google Search, and Google is telling you it doesn't matter for Google Search specifically. I lean toward thinking the file is closer to dead than alive for citation purposes across the major assistants, but that's a guess based on which crawlers actually fetch it, not something the article proves.
what Google says works instead
The positive guidance is short and old. Index-eligible pages, snippet-eligible pages, semantic HTML, reduce duplicates. The one piece with teeth is the "non-commodity content" framing, where Google contrasts "7 Tips for First-Time Homebuyers" against "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line." The first is a thing anyone could write. The second is a thing only the person who lived it could write.
That distinction is the one that maps cleanly to how RAG systems pick what to quote. Generic listicles compete against ten thousand other generic listicles for retrieval. A page with a specific dollar figure, a specific timeline, and a specific outcome doesn't.
one thing to do this week
Pull the five pages in your content calendar for the next 30 days. For each one, ask whether the draft could have been written by someone who never worked at your company. If yes, kill the draft or rewrite it around a specific decision someone on your team actually made, with the numbers attached. That's the closest operational translation of Google's commodity vs. non-commodity line, and it does double duty for ChatGPT and Perplexity retrieval, which weight specificity for the same reason.
The interesting downstream question is what happens to the agencies whose deliverables were exactly the things on Google's mythbusting list. The Universal Commerce Protocol mention at the end of the guide hints at where the real new surface is, and it isn't a text file at the root of your domain.
