
by Ema Fulga
Ema is an AI Search Content Strategist and GEO (Generative Engine Optimisation) expert. She's also the founder of decipher., an AEO agency that helps brands appear where people are now searching: AI-powered platforms like Perplexity, ChatGPT, Gemini, and others. With a background in copywriting and creative strategy, she’s on a mission to turn messy messaging into clear and structured content that helps brands get mentioned and cited in AI searches.
Connect with Ema.
Last updated: 05.10.2026
TL;DR: Google has published its own official guidance on optimising for AI Overviews and AI Mode, and the headline message is refreshingly unsentimental: there’s no secret AI trick. Solid SEO, genuinely useful content, and clean technical foundations are still what drive visibility, because Google’s generative features are built directly on top of its existing Search ranking systems. Google also used the guide to explicitly debunk several popular “AEO/GEO hacks”, including llms.txt files, content chunking, keyword stuffing, and chasing inauthentic mentions. Let’s break down what Google’s tools actually are, what the guide tells you to do, and the missing piece brands need to hear.
Google has been the gamemaster of search for decades now. When AI search came along, it wasn’t going to give up the top spot easily. So naturally, it incorporated AI into its existing search experience to keep people Googling. One thing about Google is they love a bit of mystery. For a while now, the industry has been swapping theories about what actually gets a brand surfaced in AI Overviews or AI Mode. Some of it’s been solid. A lot of it has been guesswork dressed up as certainty. So when Google published its own official guide on the subject, I simply had to stop to read it, because for once, we didn’t have to reverse-engineer the answer.
Here’s what’s in it, translated out of tech-nerd documentation-speak.
First, what are we even talking about?
Google’s generative AI search features come in two flavours:
AI Overviews are the AI-generated summaries that appear directly inside regular Google Search results, above or alongside the usual blue links, answering your query with a synthesised response and source citations, without you having to leave the normal search results page.
AI Mode is a separate, fully conversational search experience inside Google, closer to chatting with an AI assistant than scanning a results page. You can ask follow-up questions and go back and forth, and it pulls in web results as it goes.
Neither is a separate search engine bolted onto the side of Google. Both are built directly on Google’s existing Search ranking and quality systems, using two techniques worth understanding:
Retrieval-augmented generation (RAG), sometimes called grounding, is how Google pulls in fresh, relevant pages from its Search index to base an AI answer on, rather than letting the model just generate an answer from what it already “knows.” That’s why AI Overviews show clickable source links: the answer is built from retrieved pages, not invented from nothing. They weren’t getting rid of the list of links we’re used to that easily!
Query fan-out is the process where the model quietly generates a batch of related sub-queries around your original question to gather a fuller picture before answering. Ask “how do I fix a lawn full of weeds” and behind the scenes it might also be querying “best herbicides for lawns” or “how to prevent weeds in lawn”, then pulling from whichever pages answer those best. This makes it distinct from optimising for keywords like we did in pure SEO. If your content only answers the exact question someone typed, you might be invisible to the fan-out queries running underneath it.
So does “AEO” or “GEO” even apply to Google?
Yes, but not in the way you might think. This is where Google’s framing is genuinely useful. Google’s own view is that optimising for its generative AI features isn’t a new discipline. It’s still SEO, just applied to a newer surface. We’re not starting from scratch. We’re refining and adapting, which, to be honest, Google has done with its algorithm throughout its history.
Google isn’t dismissing GEO as a category. Just the need to throw all your toys out of the pram. Even if it did, Google can only speak on Google Search. Other AI search engines (ChatGPT, Perplexity, Claude) are a different story, and we’ll come back to that. (If AEO and GEO are still new terms to you, our AI search hub is a good place to get oriented.)
What Google actually told you to do
Strip away the documentation formatting, and Google’s advice comes down to three pillars.
1. Create genuinely non-commodity content. Google draws a sharp line here between content it wants to cite and content it will gloss over. An article like “7 Tips for First-Time Homebuyers” is commodity content, generic, replicable, and something an AI could churn out itself. Something like “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line” is non-commodity. Why? It’s a specific, first-hand experience nobody else can produce. Google’s systems are explicitly looking for the latter. A genuine point of view, built on real expertise or experience, not a summary of what’s already out there.
2. Keep your technical foundations clean. It can’t cite what it can’t see. None of this works if your pages aren’t indexed and eligible to appear with a snippet in regular Search results in the first place. Please note that a site also needs to be opted into generative AI features inside Search Console to be eligible. It breaks my heart when I see this mistake on a brand’s website! Keep in mind crawlability, semantic structure people (and screen readers) can actually navigate, solid JavaScript SEO if you’re using it, good page experience, and minimal duplicate content. This is the unglamorous SEO checklist, and Google is explicit that it’s still the foundation.
3. Get your local and ecommerce data structured properly. If you’re a local business or a retailer, Google points specifically to Merchant Center feeds and Google Business Profiles as the mechanisms that let your product listings and business details surface inside AI-generated answers, not just standard results. So get on that!
What Google says you can stop worrying about
AI/SEO hackers, I’m about to burst your bubble. With all the love in the world, these “tactics” aren’t going to earn you millions in AI traffic. That’s coming from the horse’s mouth. Don’t shoot the messenger! Google explicitly called out several popular tactics as unnecessary for Google Search specifically.
The rumour | What Google actually says |
“You need an llms.txt file” | Google Search doesn’t use llms.txt or similar markup files at all. Create one for other platforms if you like, it won’t help or hurt on Google. |
“Content needs to be chunked into tiny pieces for AI” | Not required. Google’s systems understand nuance across a full page and surface the relevant section. There’s no ideal page length. |
“You need to rewrite content specifically for AI systems” | Google’s models understand synonyms and intent without exact keyword matches, so there’s no need to obsessively cover every phrasing variant. |
“Chase mentions across blogs, forums, and videos to build AI trust” | Seeking inauthentic mentions isn’t effective. Google’s ranking systems are built around genuine content quality, not mention volume. |
“Structured data is essential for AI visibility” | Not required for generative AI search specifically, though it’s still worth using for rich results eligibility in standard Search. |
Worth flagging: Google saying structured data isn’t required for its own AI features doesn’t mean schema markup has stopped mattering everywhere. Other AI engines lean on it far more heavily to verify and quote your brand accurately. We’ve gone deep on that in our guide to the technical side of AI brand visibility.
Where this leaves you in your AI optimisation efforts
Google’s guide feels like finding a golden ticket. But that doesn’t mean Willy Wonka is going to give you the keys to the chocolate factory. It’s worth being clear-eyed about what it is and isn’t.
It’s guidance for one generative search surface. It tells you how to show up in AI Overviews and AI Mode, both of which sit inside Google. It says nothing about ChatGPT, Perplexity, Claude, or Gemini, engines that source and weight content very differently from Google (we’ve written before about how those engines behave when it comes to brand mentions on ChatGPT specifically). Following Google’s advice to the letter won’t automatically translate into visibility on other platforms people are increasingly using to search.
It also, understandably, doesn’t tell you how to measure yourself against competitors, benchmark performance across engines, or diagnose why you’re being cited on one platform and not another. Google does point you to the Generative AI performance report in Search Console for its own surfaces, which is a genuinely useful (and free) starting point, but it’s a single-engine view of a multi-engine problem.
Key takeaways
Google’s generative AI search features are built on its existing Search ranking systems, using retrieval-augmented generation and query fan-out, not a separate ranking mechanism.
The core advice is unglamorous but real: non-commodity, first-hand content, clean technical SEO, and structured local/ecommerce data.
Google explicitly debunked llms.txt files, content chunking, AI-specific rewriting, inauthentic mention chasing, and structured-data obsession as unnecessary for its Search features.
This guidance covers Google specifically. It doesn’t address how ChatGPT, Perplexity, Claude, or Gemini source and weight content, which can differ substantially.
Search Console’s Generative AI performance report is a solid, free way to monitor Google-specific visibility, but it won’t tell you how you’re doing anywhere else.
One platform down. Onto the next…
Google’s advice is a good foundation for search optimisation: valuable content and clean technical SEO, applied consistently. But “visible on Google” and “visible in AI search” aren’t quite the same claim anymore. The other engines your customers are asking don’t play by identical rules.
At decipher., we help brands build on exactly this kind of solid foundation, then extend it across the full AI search landscape, not just Google’s corner of it. Take a look at what we do, or if you want to know where you actually stand across ChatGPT, Perplexity, Gemini, and Claude, not just inside Search Console, let’s talk.