Google Just Told the AEO Industry Half Its Playbook Doesn't Work
Google quietly updated its official guide to optimizing for generative AI search on June 29, and it reads less like best-practice advice and more like a shot at the entire AEO cottage industry. In a section titled "what you don't need to do," Google names the exact tactics that agencies have been selling for the past year and says, plainly, that they do nothing for visibility in Google Search or its AI features.
What Google says to stop doing
The list is specific. LLMS.txt files and other "special" markup: ignored, because Google Search doesn't read them. "Chunking" content into tiny AI-friendly pieces: unnecessary, because the ranking systems already understand multiple topics on a page. Rewriting content in a special way just for AI: pointless, because the models understand synonyms and intent. Seeking inauthentic "mentions" across the web: not helpful, and partly caught by spam systems. Over-investing in structured data as an AEO trick: not required for AI answers, though still fine for rich results.
If you have been paying for a service built primarily on any of those four things, this is Google telling you where your money went.
The part worth reading twice
Strip out the mythbusting and Google's actual advice is almost boringly familiar: create non-commodity content with a genuine point of view, keep the site technically crawlable, and write for humans instead of for the model. Their own example is sharp. A generic "7 Tips for First-Time Homebuyers" post is commodity content anyone could produce. "Why We Waived the Inspection and Saved Money: A Look Inside the Sewer Line" is not, because it carries first-hand experience a model can't manufacture.
That is the whole game now. AI answer engines pull from a lot of sources at once, so the thing that gets you cited is the perspective nobody else on the page has.
Where this framing falls short
Here is where I'd push back on the coverage treating this as the final word. Google is describing Google Search. It is not describing ChatGPT, Perplexity, or Claude, which build answers differently and, in some cases, do read structured signals Google shrugs off. Telling publishers "llms.txt does nothing" is true for Google and not automatically true everywhere else. Google also has an obvious incentive to discourage tactics that route attention to rival answer engines. Read the guide as accurate about Google, and as one interested party's view of the wider landscape.
The honest takeaway is that most AEO "hacks" were always a repackaging of SEO fundamentals with a fresh coat of jargon. If you want to actually understand those fundamentals rather than rent them from a consultant, a solid, current SEO and AI-search reference ↗ will take you further than another checklist of tricks.
This is also the logic behind prismpublication.com: the win isn't gaming the answer, it's being genuinely worth surfacing at the moment someone asks.