Google told you to ignore most GEO advice.It didn't tell you about ChatGPT, Claude or Perplexity.

Google has just done something useful: it published its own developer guidance on AI search visibility, naming the tactics sold under the GEO and AEO label and stating plainly which ones do nothing for its systems. It's worth reading directly. It's also worth reading carefully, because every word of it describes Google's own AI features. It says nothing about the other engines a buyer might be using.

What Google says you can stop worrying about

The mythbusting section is the most useful part, mostly because it names things directly that most GEO advice tiptoes around. An llms.txt file isn't read by Google Search at all, so it does nothing for visibility there. Breaking a page into small, "AI-friendly" chunks isn't required either, since Google's systems already work out which part of a longer page answers a given question. Rewriting copy in some imagined AI dialect doesn't help, because the models handle synonyms and intent without it. Chasing mentions across forums and directories purely to manufacture buzz gets called out directly, with Google describing it as unhelpful and close to the kind of behaviour its spam systems exist to catch. Even structured data, which Surfacing recommends and most GEO consultants sell hard, is described as unnecessary for AI visibility specifically. It still earns rich results in ordinary search. It isn't the lever that gets a model to recommend you.

What Google says matters

This part is duller than the mythbusting, mainly because it's just SEO with newer vocabulary. Google's AI Overviews and AI Mode are grounded in the same index that powers ordinary search results, using retrieval-augmented generation: the system pulls from pages it has already ranked, then builds an answer from what it finds there. It also runs "query fan-out," quietly firing off several related searches behind a single question to gather more context before answering. In practice, that means writing content with a real point of view instead of a commodity summary anyone could produce, keeping the site technically clean and crawlable, and not expecting any of it to guarantee an outcome. Google says outright that meeting every requirement doesn't guarantee a page gets crawled, indexed, or shown. There are no guarantees here, whatever anyone selling AI visibility services tells you, Surfacing included.

The part this guidance can't cover

Google's AI Overviews, AI Mode and Gemini all draw on Google's own search index. ChatGPT, Claude, Perplexity and Microsoft Copilot don't. They're built on different infrastructure entirely, which means the advice that gets a page into a Google AI answer doesn't automatically get it into the others. That's not a footnote. It's the actual reason "do good SEO" isn't a complete answer to "will AI recommend my business."

Four engines, at least three different indexes

This is the part worth understanding properly, because it's where most of the wasted effort happens. A site can be doing everything right by Google's playbook and still be structurally invisible to half the engines a buyer might be asking.

Some of this is observed behaviour rather than something every company has confirmed outright, but it's corroborated across independent sources, including the platforms' own documentation where it exists. Treat the architecture as solid and any precise statistics attached to it as directional.

What to do about it

The honest version is less exciting than most GEO pitches, because the fix for each engine is mostly infrastructure rather than a clever content trick. For Google's own AI surfaces, ordinary technical SEO and useful, well-written content already covers most of it. For ChatGPT and Copilot, the unfamiliar step is Bing Webmaster Tools: verifying the site, submitting a sitemap there separately from Google's, and checking robots.txt isn't quietly blocking Bing's crawler. For Claude, it's confirming Brave's crawler can reach the site at all, a check almost nobody runs because almost nobody has heard of it as a requirement. For Perplexity, it's leading with the actual answer near the top of the page, keeping content current, and earning a place on the third-party sources Perplexity already trusts rather than relying only on owned content.

None of this replaces the harder distinction underneath it. Being crawlable is the starting condition, not the goal. Whether a model puts your name forward when someone asks a buying question is a separate problem, and it's the one that decides whether a deal happens before you ever hear about it.

Want to see where your own site stands? Run the free check, or work through the full GEO checklist for the longer version of this list.