Google Wrote the GEO Rulebook. The Problem Is, Google Doesn't Own the Game Anymore.
For eighteen months, the marketing industry held a séance. Everyone showed up with their own theory about what AI search optimization required, said it loudly enough on LinkedIn, and eventually convinced themselves it was real. Then on May 15, 2026, Google walked into the room and published the official guide. The twist is that Google is no longer the only one running search, and the LLMs that matter most for "AI search" don't take orders from Mountain View.
This is a stranger moment for marketing than it first appears.
How GEO and AEO Became Real By Repetition
There was no founding document for "Generative Engine Optimization" or "Answer Engine Optimization." There was no whitepaper, no research consortium, no platform release notes. There was an observation, which was that AI Overviews started appearing at the top of search results and that ChatGPT was eating informational queries. From that observation, an industry of guesses bloomed.
The guesses came from people who genuinely wanted to figure it out, and from vendors who needed something to sell. Both groups ended up saying the same things. Chunk your content. Add llms.txt files. Rewrite headings as questions. Build FAQ retrofits. Seed authentic mentions across Reddit. Use structured data more aggressively. Optimize for entities. Build topical authority clusters.
None of it was based on documentation from the systems being optimized. It was reverse-engineered intuition repeated until the repetition itself became the evidence. By early 2026, you could buy a six-month GEO retainer from agencies that had no more access to Google's internals than the rest of us.
Marketing teams felt the pressure to act. AI Overviews now appear on approximately 48% of all tracked search queries as of about February 2026, per BrightEdge data, and that pressure was real. The problem was that the action being recommended was, in significant part, a collective hallucination.
Then Google Published the Guide
Google's May 15 documentation is unusually direct. It names tactics. It says, in plain language, that llms.txt files do not influence Google's AI search features. It says content chunking is not required. It says AI-specific rewriting is not necessary. It says inauthentic mention building is already filtered. It says structured data is not a special GEO lever, just standard SEO hygiene.
The official position on the whole vocabulary is this: from Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.
That single sentence retired a category of marketing services. Several major industry publications read the document the same way. Search Engine Journal reported that AEO and GEO are treated as part of SEO, not separate disciplines, for Google Search, and that Google says llms.txt, content chunking, AI-specific rewriting, and special schema aren't needed for its generative AI features. The framing of the guide itself, with its mythbusting section calling out specific tactics, suggests Google viewed the situation as serious enough to require formal correction.
In other circumstances, that would be the end of the story. The authority spoke. Everyone goes home.
Except Google Isn't the Only Authority Anymore
Here is the part the guide cannot solve, no matter how cleanly it's written.
When most marketers say "AI search," they are not actually talking about AI Overviews. They are talking about the broader environment in which people get answers from ChatGPT, Perplexity, Claude, Microsoft Copilot, and increasingly from agents embedded in browsers and operating systems. ChatGPT alone reportedly processes around 2 billion queries per day according to industry data tracking. Perplexity has carved out a defensible position in research and source-citing queries. Claude is the assistant of choice for a meaningful slice of technical and professional work.
Google's market share of traditional search has dipped below 90 percent for most of 2025, according to multiple industry trackers, and informational query volume — the exact slice that drives content traffic — has shifted faster than the headline numbers suggest. The "AI search" market is not Google's market. It overlaps with Google's market.
This is where the guide becomes complicated. Google has written the rulebook for AI Overviews and AI Mode, which run on Gemini. That rulebook is binding inside Google's surfaces. It does not bind ChatGPT. It does not bind Perplexity. It does not bind Claude. It does not bind any of the agentic browsers and assistants that will likely fragment the discovery landscape further over the next two years.
Several of the tactics Google dismissed may genuinely matter for those other surfaces. Anthropic has been reasonably public about supporting llms.txt conventions. Structured content patterns that help ChatGPT extract a clean answer may not help Google's AI Overviews specifically but may absolutely help an OpenAI model selecting a citation. Brand mentions across the web influence the training data and retrieval behavior of LLMs that are not built by Google and that index the web differently than Googlebot does.
So when Google publishes a definitive guide saying "this is what works for AI search," the accurate footnote is "for the portion of AI search we operate." That is still a large portion. It is not the whole.
The Awkward Position Google Is In
Google built its authority over twenty-five years by being the place where everyone went to find things. That authority let Google publish search guidance and watch the entire SEO industry reshape itself around the recommendations. Google's word was, for practical purposes, the law of discoverability.
The AI search transition complicates that authority in a specific way. Google is now one of several systems mediating discovery, and the others have their own logic. When Google publishes a guide that frames AI search optimization as "still SEO," it is making two claims at once. The first claim is correct: for Google's own AI surfaces, the same content quality and technical fundamentals apply. The second claim, the implicit one, is that Google's definition should be the industry's definition. That second claim is the one the market may not accept the same way it did in 2010.
ChatGPT does not need Google's approval to surface a brand. Perplexity does not consult Google's guidelines when selecting citations. The agentic browsers emerging from Anthropic, OpenAI, and others are building their own conventions for how websites should expose information to AI assistants. The Universal Commerce Protocol that Google referenced in its own guide is one example of a multi-party standard that Google participates in but does not control.
This is not a criticism of Google's guide. The guide is genuinely useful. It clears away a lot of paid nonsense that has been wasting marketing budgets. The point is narrower than that. Google has the authority to define the rules of Google. It has less authority than it used to in defining the rules of "AI search" as a category, and that gap is going to keep widening.
What Marketing Teams Should Actually Do
The practical implication is that the answer to "how do we optimize for AI search" has split into at least two questions, and possibly more depending on which surfaces matter to your business.
For Google's AI surfaces, the guide is the authoritative source. Build crawlable, indexable pages. Invest in non-commodity content grounded in firsthand experience and proprietary data. Stop paying for llms.txt files, chunking projects, and FAQ retrofits if they are being sold as Google AI levers. Treat technical SEO foundations as the floor, not the differentiator.
For the non-Google AI surfaces, the answer is messier and worth thinking about on its own terms. ChatGPT, Perplexity, and Claude weight signals differently from Google. The brand mention economy that Google's spam systems filter may actually influence how non-Google LLMs surface and cite your content. Structured documents that read cleanly for an LLM doing retrieval may be worth investing in even if they do nothing for Google AI Overviews. Some of what Google dismissed is genuinely valuable elsewhere.
The harder question is governance. Marketing teams that have a single "AI search optimization" budget line and a single vendor running that work are now structurally mismatched with how the discovery landscape actually operates. Google's guide will help one surface. The rest of the work needs its own thinking, its own evaluation criteria, and possibly its own vendors.
That feels harder than the previous model, in which one authority published one guide and everyone aligned. It is harder. The previous model is also not coming back.
The Quiet Lesson Underneath
There is a lesson here about the season we just went through, and it is worth naming before the next acronym arrives.
The GEO and AEO industry grew to the size it did because people convinced themselves a new discipline existed before any of the platforms it was supposedly optimizing for had confirmed that it did. The repetition was the proof. The proof was the repetition. By the time Google walked in and said "this isn't a thing," significant budgets had already been committed to it being a thing.
This pattern is going to repeat. The next AI surface, the next agent protocol, the next browser-native discovery experience will arrive, and within months an industry of consultants and audits will exist around it. Some of those consultants will be genuinely useful. Some will be selling repeated guesses back to a market that has manifested its own demand for them.
The defense is not skepticism for its own sake. The defense is the discipline of asking, before signing the retainer, what the actual mechanism is. What system is being optimized for. What evidence exists that the proposed tactic affects that system. Whose documentation supports the claim. Whether the documentation comes from the platform itself or from a vendor selling services to optimize for it.
Google's guide is useful because it makes that question answerable for Google's own surfaces. The unfinished work is doing the same disciplined evaluation for every other surface that matters, without an official guide to lean on.
Practical Takeaways
For marketing leaders trying to translate this moment into actual decisions:
- Treat Google's guide as authoritative for Google's AI surfaces and nowhere else. The temptation will be to read it as guidance for all of AI search. It isn't.
- Stop paying for tactics Google explicitly dismissed when those tactics are being sold as Google AI levers. That is a clean budget reallocation backed by official documentation.
- Keep an open mind about whether some of those same tactics may have value for ChatGPT, Perplexity, Claude, and other LLM-driven surfaces. Different systems, different rules.
- Invest in non-commodity content that AI systems on every surface have a hard time replicating. Original data, firsthand experience, named expert bylines, proprietary research. This is the rare advice that works equally well for Google and for every system Google does not control.
- Build the internal habit of asking "what is the mechanism" before signing the contract. The next acronym is already being drafted by someone, somewhere, hoping you will manifest it into existence with them.
For most of the last two decades, optimizing for search meant optimizing for Google. The implicit assumption underneath the GEO and AEO panic was that this would still be true in the AI era, just with new tactics. The more interesting reality is that Google is one authority among several, the rulebook for AI search is going to be written by multiple parties with different incentives, and the era of waiting for one company to publish one guide and reshaping everything around it may be quietly ending. Google's guide is excellent. It is also, in a sense that would have been unthinkable a decade ago, partial.
