An industry has grown up around the idea that generative engine optimization is a new discipline requiring new techniques. Google's own documentation disagrees, in unusually plain language:
"From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
That single sentence, from Google's official guide to generative AI features, contradicts a great deal of what is currently being sold as GEO. It is worth reading what the document actually says before buying anything labelled with the acronym.
It is also worth noting where Google's guidance stops applying — because Google is one system among several, and the answer is genuinely different for the others.
What Google says the machinery is Google's AI features are not a separate index or a separate ranking system. Two mechanisms do the work.
Retrieval-augmented generation, which Google also calls grounding, uses the core Search ranking systems to retrieve relevant pages from the existing Search index. The model then reads those pages to compose a response, with links back to the sources.
Query fan-out generates a set of related queries alongside the original one. Google's example: a search for "how to fix a lawn that's full of weeds" also fires queries like "best herbicides for lawns" and "how to prevent weeds in lawn," then pulls results for those too.
The consequence follows directly. If AI answers are assembled from pages retrieved by the normal ranking system, then being retrievable by the normal ranking system is the requirement. There is no separate door.
The four tactics Google says to ignore The guide contains a section titled "Mythbusting generative AI search: what you don't need to do." Four items in it are worth quoting, because each is currently sold as a service.
llms.txt files. Google's wording is unambiguous: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." The guide adds that maintaining one is fine and "will neither harm nor help" your visibility in Google Search.
Chunking content into small pieces. "There's no requirement to break your content into tiny pieces for AI to better understand it." And directly: "There's no ideal page length."
Rewriting content specifically for AI. "You don't need to write in a specific way just for generative AI search." Google's reasoning is that its systems understand synonyms and intent, so capturing every phrasing variation is unnecessary.
Chasing mentions. Google notes that AI features do surface what's said about you across the web, but that "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem."
If a proposal for your site consists mainly of these four things, it is proposing work that Google says has no effect on Google.
What Google says does matter The guide is equally direct about the thing that does:
"Creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide."
It then draws a distinction worth internalising, using its own examples:
The first is common knowledge that could have come from anyone. The second contains something only the person who lived it could write. Google's test for the difference: whether the content "could easily be produced by a generative AI model." If it could, it is unlikely to be the thing an AI model chooses to cite.
This is the uncomfortable part of the guidance, because it can't be outsourced to a checklist. There is no technical implementation of having something to say.
Where Google's guidance stops applying Here is the qualification most coverage of this document misses. Google is describing Google.
ChatGPT, Claude and Perplexity are separate systems. They do not use Google's index, they do not run Google's ranking, and they have published nothing equivalent. Several differences are load-bearing:
They fetch with different crawlers. GPTBot and OAI-SearchBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity. Each needs permission in your robots.txt independently. Allowing Googlebot does nothing for any of them.
Most of them do not execute JavaScript. Google renders JavaScript on a delayed second pass, so a client-rendered site is eventually indexed. Other AI crawlers generally take the initial HTML response and move on. A site can therefore be fully indexed by Google and functionally empty to ChatGPT — not because of anything in the content, but because of how the page is delivered.
ChatGPT search draws on Bing's index. Which makes Bing Webmaster Tools materially more useful for ChatGPT visibility than most people assume, and makes it the single highest-leverage registration outside Google.
Perplexity leans heavily on third-party sources. Roundups, comparisons and forum discussions carry more weight there than a vendor's own page. Being included in someone else's list often matters more than anything on your own site.
So the honest summary is this: for Google, GEO is SEO. For everything else, the difference is access and delivery — which crawlers can reach you and what they receive when they do. It is not a writing style.
What actually differs, in practice Strip out the marketing and there are four things worth doing that classic SEO advice doesn't emphasise:
- Check what non-Google crawlers receive. Request your own page while identifying as an AI crawler and count the words that come back:
curl -A "GPTBot" yoursite.com | wc -w If the result is a handful of words on a site with real content, your pages are being assembled by JavaScript that the crawler never runs. Prerendering or server-side rendering is the fix, and it's an engineering task rather than a content one.
-
Decide about each crawler deliberately. Open your robots.txt and look for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and Applebot-Extended. Google-Extended is the one that catches people: it governs AI Overviews and Gemini but has no effect on rankings, so blocking it breaks nothing visible. Most sites have never named any of them, which means the decision was made by a default.
-
Register with Bing, not only Google. Because ChatGPT search reads Bing's index, Bing Webmaster Tools does double duty. It takes about twenty minutes and is skipped by almost everyone.
-
Measure it. Google added generative AI performance reporting to Search Console in 2026, so AI Overview visibility is now observable rather than guessed at. There is also a Search Console control for whether your site appears in and grounds generative AI responses — worth confirming which way yours is set, since opting out removes both traffic and impressions from those features.
For the non-Google engines there is still no equivalent reporting. The practical substitute is a manual log: ask each engine the questions your customers would ask, monthly, and record whether you appear and which source it cited. The cited source is usually more actionable than your own absence — it tells you which page to get included in.
The short version GEO as a distinct discipline, with its own file formats and writing rules, is not supported by what Google has published. The parts of it that survive scrutiny are unglamorous: make sure crawlers can reach you, make sure they receive actual content, register with Bing, and write things a language model couldn't have written on its own. That last one is the whole game, and it's also the only part that can't be bought as a service.
Questions
Is GEO a real thing or not? The problem it names is real: AI systems are answering questions that used to be answered by a page of links. The claim that it requires separate techniques is what Google's guidance contradicts — at least for Google. For other engines, the differences are technical rather than editorial.
Should I still create an llms.txt file? Google explicitly ignores it. If you want one for other systems that read it, Google says it will neither help nor harm. Treat it as optional rather than foundational.
Does content length matter? Google states there is no ideal page length, and that both shorter and longer pages can work depending on the subject and audience.
What's the highest-value thing to do first? Check what an AI crawler actually receives from your site. If the answer is very little, no amount of content work will change your visibility in AI answers until that's fixed.
Sources: Google Search Central, "Optimizing your website for generative AI features on Google Search" and "Creating Helpful, Reliable, People-First Content."