Google just told you to ignore most GEO advice. Here's what matters.

On May 15, Google published its first official guide on optimizing for AI Overviews and AI Mode. The headline is blunt: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Google names llms.txt files, content chunking, AI-specific rewriting, and special schema as tactics you can ignore for its generative AI features.

If you've been pitched any of those in the last six months, you now have the primary source that says otherwise.

This matters for Southeast Asian startups and SMEs because the gap between what Google says works and what gets sold as "GEO services" is wide, expensive, and growing. The guide doesn't introduce new rules. It documents the rules that already applied, and debunks the ones that never did.

What Google actually said

The guide covers five areas: non-commodity content, local and shopping optimization, technical requirements, what to ignore, and early guidance on agentic experiences. The "what to ignore" section is the most operationally useful.

Google says you don't need llms.txt files or other "special" markup to appear in generative AI search. Google may crawl these files, but they receive no special treatment. You don't need to break content into small pieces (chunking) because Google's systems "are able to understand the nuance of multiple topics on a page and show the relevant piece to users." You don't need to rewrite content in a specific way for AI systems. And there's no special schema required for generative AI search, though standard structured data remains useful for rich results eligibility.

The reason this cuts through is simple: these are four of the most commonly promoted GEO tactics in agency pitches and consultant audits. Google is now on record saying they don't apply to its own platforms.

That doesn't settle the question for ChatGPT, Perplexity, or Claude. But if your acquisition funnel runs through Google Search, you've just been handed permission to skip a category of work.

What this means for Series A through Series C teams

Most startups I work with are stretched thin. One or two people carry search. The ops reality is binary: what gets prioritized, and what gets cut. This guide gives you coverage to cut four things that were eating cycles.

You don't need to commission llms.txt files. You don't need to audit every page for "chunk-ability." You don't need a separate AI rewrite pass on existing content. You don't need to hunt for GEO-specific schema beyond what you'd implement for standard SEO.

What you do need is the same list that's been true since 2021: content that only your team can write, structured clearly, with technical hygiene that makes it crawlable and indexable. Google's guide calls this "non-commodity content." The example they give is pointed. Commodity: "7 Tips for First-Time Homebuyers." Non-commodity: "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line."

The difference is specific, first-hand operational knowledge. If an AI engine can synthesize your content from five other sources, you're not adding citation value.

In my consulting work with Southeast Asian startups, the pattern is consistent. The companies that show up in AI Overviews are the ones publishing founder-level insights, customer research they ran themselves, or product decisions no competitor has context on. The ones that don't show up are the ones rewriting industry best practices with minor regional tweaks.

Google's guide formalizes what the data already showed. Original beats optimized.

The three things you should audit this week

If you haven't touched your search practice since this guide dropped, run these three checks.

First, audit for non-commodity content. Open your five highest-traffic pages. Read each one and ask: could a competitor write this if they had access to the same public information we did? If yes, you have a commodity page. It will rank on domain authority and backlinks, but it won't get cited in AI answers. The fix is not a rewrite. The fix is gathering the primary-source material (customer data, internal process notes, founder interviews) that makes the page genuinely unique.

Second, confirm technical eligibility. Google's guide is explicit: pages must be indexed and eligible to appear with a snippet. If you've blocked snippets via meta tags or robots rules, you've also blocked AI Overview inclusion. Check Search Console for pages that rank but don't show snippets. That's your audit list.

Third, verify local and shopping feeds if you run transactions. Google says that Merchant Center feeds and Google Business Profiles directly influence what appears in AI responses for product and local queries. If you're an e-commerce startup or a service business with local presence, those feeds are infrastructure, not optimization. You don't get a second chance to be cited if the feed isn't live.

What to do about non-Google AI engines

This guide applies only to Google Search. ChatGPT, Perplexity, Claude, and Gemini (outside of Google Search) may weight signals differently. The question I get from every startup is: do I need separate strategies?

My answer is no, with one caveat. The foundational work overlaps almost completely. Crawlable site, clear structure, original content, entity clarity (via schema or consistent naming). That work benefits every platform.

The divergence is in distribution tactics. If you're actively pursuing ChatGPT citations, you may want to ensure your robots.txt allows the relevant crawlers (CCBot, GPTBot). If you're targeting Perplexity, branded mentions on high-authority sources (Reddit, industry forums, case-study platforms) correlate with citation share. But those are channel-specific plays, not foundational rewrites.

For a startup with limited headcount, the right sequencing is: nail the Google-documented basics first, then layer in platform-specific work only if that platform drives measurable pipeline. I've yet to see a Series A or Series B company where non-Google AI traffic justified separate workstreams before the Google work was stable.

What I'm telling clients this month

Three operational points I'm emphasizing in consultancy engagements and training workshops right now.

One: if you commissioned GEO-specific work in Q1 or Q2 2026 that included llms.txt, chunking, or AI schema, audit the invoice against this guide. You may have paid for work Google now says is unnecessary. That's not a vendor quality issue, it's a timing issue. The guide came out May 15. Pre-May recommendations were based on inference. Post-May, you have a primary source.

Two: the highest-ROI work for the next 90 days is non-commodity content creation, not optimization of existing pages. If you have one writer and 40 hours a month, spend 30 of those hours on original research, founder Q&As, or customer case studies. Spend 10 on technical cleanup (indexability, snippet eligibility, schema validation). Skip the rewrite queue unless a page is factually outdated.

Three: Google's section on agentic experiences is labeled "if this is something that's relevant to your business and you have extra time." That framing is deliberate. Agent optimization (Universal Commerce Protocol, WebMCP) is forward-looking, not urgent. If you're a marketplace or transactional platform, it's worth a read. If you're a SaaS product or content business, it's not your H2 priority.

Why this guide matters more than most Google docs

Google documents a lot. Most of it is incremental. This one is structural.

For the first time, Google has named the tactics that don't work for its AI features and put them in official Search Central documentation. That creates accountability. If an agency pitches you on llms.txt or chunking for Google visibility, you now have the receipt that says otherwise.

It also clarifies the discipline. AEO and GEO have been positioned by some vendors as separate practices requiring separate budgets. Google's guide says no: "from Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

That's not semantic. It's organizational. It means the team that runs your search practice is the team that runs your AI search practice. You don't need a separate GEO hire. You need your existing search lead to read this guide, adjust priorities, and keep shipping.

If your current search consultant or agency hasn't sent you a breakdown of this guide and how it changes your roadmap, that's a signal. The doc is three weeks old. It's the most significant official AI search guidance Google has published. If it hasn't landed in your inbox yet, you're working with someone who isn't monitoring the primary sources.

What to watch next

Google's guide is a snapshot, not a ceiling. Two things are likely to shift in the next six months.

First, the "agentic experiences" section will grow. Google mentions Universal Commerce Protocol and WebMCP as emerging standards for agent-friendly websites. Both are experimental. But if Google is documenting them in May 2026, expect them to become standard by Q1 2027 for transactional sites. If you run booking, checkout, or multi-step workflows, add this to your H2 reading list.

Second, non-Google platforms (OpenAI, Anthropic, Perplexity) will likely publish their own optimization guidance in response. The fact that Google went first sets the template. Expect others to follow with similar debunking of over-hyped tactics and emphasis on original, structured content.

For now, the actionable move is simple: if you've been told you need llms.txt, chunking, or special AI schema for Google, you don't. If you've been told you need a separate GEO strategy from your SEO strategy, you don't. And if you've been sold on optimizing for AI visibility before you've fixed basic indexability and snippet eligibility, the sequencing is backwards.

Google just gave you a primary source. Use it to cut the noise and focus the work.

If you want an audit of where your current practice stands against this guide, book a call or explore the consultancy offering. I run these audits weekly for Southeast Asian startups, and the debrief typically saves three months of misdirected work.