Google just quietly published its first proper guide to optimising content for AI search. Most of it isn’t about what to do. It’s a list of things the AI SEO industry has been selling to business owners that Google says don’t matter. If you’ve been approached recently with a pitch about LLMS.txt files, content chunking, or “AI-ready” markup, this one is worth a read.
A few days ago Google added a new page to Search Central called the “AI Optimization Guide.” It’s the first official guidance they’ve put out on how to think about getting your content surfaced in AI Overviews, AI Mode, and the various generative features now baked into Google Search.
I read through it expecting the usual mix of best-practice nudges. What I didn’t expect was that a large chunk of the document is essentially Google saying “please stop doing the things people are selling you.”
That’s the part I want to talk about, because a few of our clients have been pitched on these “GEO” or “AI optimisation” services over the last couple of months, and the advice is often confidently wrong.
What Google actually says
The bit that surprised me is a whole section they’ve titled “Mythbusting generative AI search: what you don’t need to do.” Google rarely writes pages like this. When they do, it’s usually because enough rubbish is circulating in the SEO community that they’ve decided to step in.
Here are the six things Google explicitly tells you not to worry about.
LLMS.txt files and AI-specific markup
There’s been a push over the last six months to add an LLMS.txt file to your site, a special text file that supposedly tells AI crawlers what to read and what to skip. I’ve seen consultants charging to set them up.
Google’s words: “You don’t need to create new machine readable files, AI text files, markup, or Markdown.”
That doesn’t mean LLMS.txt files cause harm. It just means they don’t help with Google’s AI features, which is the dominant slice of AI search traffic for most NZ businesses.
Content chunking
Another popular pitch: rewrite your pages into tiny “chunks” so AI can grab them more easily. Often this involves breaking long articles into a series of small, single-idea pages.
Google: “There’s no requirement to break your content into tiny pieces for AI. We understand the nuance of multiple topics on a page.”
If you’ve got a long, well-structured article that covers several related ideas, leave it alone. Splitting it into ten thin pages will probably make your site worse, not better.
Writing in a special way for machines
Some agencies are now offering “AI-optimised copywriting,” content written with particular sentence structures, keyword densities, or formatting that’s supposed to be more legible to large language models.
Google: “You don’t need to write in a specific way just for generative AI search.”
The same systems Google uses to evaluate your content for regular search results are doing the work behind AI Overviews. There’s no separate AI scoring layer that rewards machine-friendly prose. If anything, writing for humans first is more important than ever.
Chasing mentions across the web
This is the one I see most often. “We’ll get your brand mentioned across hundreds of sites so the AI sees you everywhere.” Sometimes it’s paid placements, sometimes it’s AI-generated commentary posted on directories and forums.
Google warns that this kind of inauthentic mention-seeking falls under their scaled content abuse policies and trips spam detection. It’s not just unhelpful. It can actively hurt you.
Pages for every possible query variation
The idea here is that because AI search uses “query fan-out” (it generates lots of related questions in the background), you need a dedicated page for every variation of every question your customers might ask.
Google: “High quantity of pages doesn’t make a website higher quality.”
It violates their scaled content policy and, in their words, is “an ineffective long-term strategy.” We’ve seen accounts where this approach has been taken: dozens of near-identical service pages, one per suburb, all written by AI. It rarely ends well.
Special schema markup for AI
There’s a persistent belief that you need particular structured data, some special AI schema, to be eligible for AI Overviews.
Google: “Structured data isn’t required for generative AI search.”
Schema is still worth using for rich results in normal search. But there’s no AI-specific markup that gives you an edge in AI features. Anyone telling you otherwise is selling something Google doesn’t recognise.
So what does work?
Reading the rest of the guide, the answer is unsurprising and a bit anti-climactic.
Google’s position is that generative AI features are built on top of the same core ranking systems that drive normal search. Pages need to be crawlable, indexed, and eligible for snippets. Content needs to be helpful, written for people, and bring a genuine point of view rather than being a restatement of what’s already out there.
The things that have always worked in good SEO are what keep working: clear page structure, technical soundness, unique perspective, real expertise. There’s no special key that unlocks AI visibility separately.
What’s genuinely new is small:
- AI features can pull in images and video alongside text, so well-tagged media has a slightly bigger role
- Systems are better at understanding synonyms and intent, which reduces the need to write content stuffed with keyword variations
- “Query fan-out” means a single page can satisfy a wider range of related queries than before, which is the opposite of needing more pages
None of this requires new files, new markup, or a different writing style.
Why I think this guide matters
Two reasons it’s worth paying attention to.
The first is that Google rarely writes “stop doing this” content. When they do, it’s a signal that a lot of money is being spent on the wrong things. A business owner reading a confident pitch from a GEO agency can now point to Google’s own guidance and ask the agency to reconcile their pitch with it. That’s a useful test.
The second is that it clarifies what’s actually worth investing in. Time and budget spent on LLMS.txt files, content chunking projects, and AI-specific schema is time and budget not spent on the things that genuinely move the needle: better content, better page experience, real authority signals.
I’m not certain Google’s guidance is the last word on every AI platform. ChatGPT, Perplexity, and Claude all draw on different sources and behave differently. The advice in Google’s guide is specifically about Google’s AI features. But Google’s AI features are where the majority of NZ search traffic touches AI right now, and following its guidance hasn’t really steered anyone wrong in the past.
If you’re being pitched on AI optimisation services, a fair question to ask is: which of these recommendations conflict with Google’s official guide? If any of them do, that’s worth a longer conversation before you sign anything.
If you’d like a second opinion on an AI SEO or GEO proposal you’ve received, the form below is the easiest way to get in touch.
