Google dismisses llms.txt and GEO tactics in new AI guide

What happened

Per Search Engine Journal, Google has published a new AI Search guide that tells site owners to stop treating Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) as separate disciplines. The official line: it is all still SEO. The guide also names specific tactics Google considers a waste of time, including llms.txt files, manual content chunking, and AI-specific schema markup.

Matt G. Southern's write-up makes Google's position explicit. The company is arguing that the same crawlable, well-structured content that wins in classic search is what surfaces in AI Overviews and AI Mode. No parallel stack required.

This is Google staking out territory. The GEO consultancy market has been selling a separate playbook for eighteen months. Google just told CMOs they do not need to buy it.

Why it matters for your brand

The first thing to understand: Google's guidance applies to Google. It does not apply to ChatGPT, Perplexity, Claude, or Copilot. Those systems use different retrieval architectures, different source weightings, and in several cases different crawlers entirely. A CMO who reads Google's guide and concludes "we can ignore AI-specific work" is making a category error. Google is telling you how to win in Google's AI surfaces. It is not telling you how to win in the AI surfaces your buyers actually use to shortlist vendors.

For financial services brands, this matters because procurement-adjacent research is increasingly happening in ChatGPT and Perplexity, not Google. When a treasury team at a Fortune 500 asks an LLM "who are the leading custodians for tokenised assets," the citation set is built from a mix of trade press, analyst notes, and Reddit threads. llms.txt may be dismissed by Google, but Anthropic and OpenAI have both signalled interest in publisher-controlled access signals. Dropping the file because Google said so is premature.

For multilaterals and policy institutions, the implication is sharper. UN agencies, the World Bank ecosystem, and standards bodies like ISO and IEEE produce the exact kind of authoritative, long-form, well-cited content that LLMs love to quote. Google's guide effectively confirms what we have argued: the technical optimisation theatre matters less than whether your content is structurally trustworthy and properly attributed. For a CGAP or a UNDRR, this is good news. The work to make a research paper citable in Claude is largely the work to make it readable and indexable, full stop.

For major industrial groups, the risk is different. Companies like HOLCIM or Adecco often have sprawling, multi-domain, multi-language web estates where the basics (crawlability, canonical tags, structured data for products and locations) are broken in ways that no GEO tactic can fix. Google's guide is a useful internal weapon here: it gives the CMO ammunition to push budget back into fundamentals rather than letting an agency sell a "GEO transformation" on top of a broken foundation.

The content strategy implication: stop building two production pipelines. One pipeline, optimised for citation across every surface that matters, beats a "regular content" team and a separate "AI content" team. The brands winning citations in our tracking are the ones treating every page as a potential answer, not the ones running parallel GEO experiments.

The signal in context

Google's framing fits a pattern. Whenever a new optimisation discipline emerges around Google's surfaces (think mobile-first, think Core Web Vitals, think E-E-A-T), Google's eventual public position is "this was always just good SEO." The company has a structural interest in collapsing AEO and GEO back into SEO, because a separate discipline implies separate metrics, separate vendors, and eventually separate budgets that Google does not control. Calling llms.txt unnecessary is consistent with this: Google does not want a publisher-side standard that fragments how AI systems negotiate access.

The broader trend worth watching is divergence between platforms. ChatGPT cites Reddit and Wikipedia heavily. Perplexity weights primary sources and recent news. Google AI Overviews favour its own index and YouTube. A single "AI visibility" strategy that ignores those differences will underperform. Google saying "it is all SEO" is true for Google and misleading for everywhere else. Senior marketers should treat the new guide as one input among many, not as the settled definition of how AI search works.