Is SEO still relevant with AI search? Five takeaways from Google's new guide

With AI now sitting at the heart of how people search — and "GEO" (generative engine optimisation) being marketed as the next big thing — a fair question keeps coming up: is SEO still worth the effort

Spoiler: yes. And we no longer have to guess; Google has spoken!

Google has just released an official guide on optimising for generative AI features on Search, covering AI Overviews, AI Mode and the broader generative experience [1]. If you’ve been wondering whether to bin your SEO playbook and pivot to “AEO” or “GEO” hacks, this is the clarifying read you’ve been waiting for.

Here are the five takeaways I think matter most — especially if you’re a researcher, academic, founder or technical expert trying to separate signal from noise.

---

#1. GEO and AEO are not separate disciplines — they are SEO

Let’s start with the headline. In Google’s own words: optimising for generative AI search is optimising for the search experience, and thus still SEO [1]. google

That single line settles a lot of debate. AEO (answer engine optimisation) and GEO (generative engine optimisation) have been marketed as new disciplines — sometimes with their own workshops, certifications and agency packages. From Google Search’s perspective, they are not.

The reason is technical, and worth understanding. Google’s generative AI features sit on top of the same ranking and quality systems that already power classic Search [1]. Two mechanisms do the heavy lifting:

• Retrieval-augmented generation (RAG), which pulls relevant pages from the Search index to ground AI responses in real, indexed content [1].

• Query fan-out, which generates a set of related sub-queries to fetch additional results for a single user question [1].

In plain terms:
AI Overviews don't go shopping somewhere else. They shop in the same index your site is competing in. If your SEO is solid, you're already in the running.

---

#2. Foundational SEO is still the bedrock

The guide is explicit — the SEO Starter Guide and Search Essentials continue to apply, because the generative features are rooted in the same core systems [1, 2].

In practice, that means the same fundamentals you’ve (hopefully) been investing in:

• A site that is crawlable, indexable and meets Google’s technical requirements.

• A clear technical structure with sensible HTML, no JavaScript blocking issues, and well-handled redirects.

• A good page experience — Core Web Vitals, mobile-friendliness, low latency.

• Reduced duplicate content and a clean URL structure.

There’s a helpful nuance for those of us who like clean code: Google notes that you don’t need perfect semantic HTML. The web in general isn’t valid HTML, and Google can interpret what you’ve got [1]. Use semantic HTML where you can — mostly because it helps real users, including screen readers, assistive tech and, increasingly, browser-based AI agents.

Semantic HTML simply means using the right code "label" for each part of your page — a heading is marked as a heading, a list as a list, a button as a button. It tells browsers, search engines and assistive tech what each element actually is, not just how it looks.

---

#3. Create non-commodity content that only you can produce

This is where the guide gets a little philosophical — and where I think it speaks loudest to academics, researchers and technical experts.

Google introduces a useful phrase: non-commodity content [1]. Commodity content is the kind anyone (or any LLM) could produce — a generic “7 tips for first-time homebuyers”. Non-commodity content offers a viewpoint that can’t be cheaply replicated: a first-hand account, an expert teardown, a counter-intuitive case study from your own work [1]. google

For my audience, that translates directly.

If you’ve spent ten years on a specific research problem, your post on it is non-commodity content by default. If you’ve run a clinic, a lab or a consultancy, your anonymised case notes are non-commodity content. AI systems are designed to surface this kind of material. They are not designed to reward yet another rephrased explainer.

Google’s quick test is worth memorising: is this content my visitors would find satisfying? If yes, you’re on the right track [1, 3].

---

#4. Skip the hacks: no llms.txt, no chunking, no fake mentions

This is the section that quietly dismantles a lot of the “GEO toolkit” being sold right now.

Google explicitly lists what you do not need to do for AI search [1]:

• No llms.txt files or special AI markup. Google does not give these any special treatment for Search ranking. You can safely ignore them as a Google visibility lever.

• No “chunking” content into tiny pieces. Google’s systems already understand multiple topics on a single page. There is no ideal page length — write for your audience, not for an imagined LLM parser.

• No rewriting for AI. Google’s systems understand synonyms and meaning. You don’t need to chase every “long-tail” variation of how someone might phrase a question.

• No buying inauthentic mentions. Mentions on low-quality, low-traffic sites don’t help, and Google’s spam systems are designed to filter them out [1, 4].

• No over-engineering structured data. Structured data is still useful for rich results in classic Search, but it isn’t a requirement for generative AI features [1].

For anyone who has been pitched a “GEO audit” that mostly amounts to generating an llms.txt and seeding mentions across obscure forums — this is your permission slip to politely decline.

---

#5. Look ahead: agentic experiences are coming

The final takeaway is more of a heads-up than an instruction.

AI agents — autonomous systems that can book a reservation, compare specifications or pull data on a user’s behalf — are starting to interact with websites directly [1]. They may render pages, read the DOM and inspect the accessibility tree to complete tasks.

You don’t need to overhaul your site for this today. But two existing investments pay off here:

• Semantic HTML and a strong accessibility tree make your content easier for agents to parse — and easier for assistive technologies, which has always been the right call regardless.

• Clear page structures and well-labelled interactive elements help agents understand what your site actually offers.

Emerging protocols like the Universal Commerce Protocol will likely shape how Search agents interact with sites in the coming years [1]. Worth keeping on your radar, particularly if you run an ecommerce or booking-based business.

---

What this means in practice

If you take one thing from this guide, take this: good SEO and good content are the strategy. Not a hack. Not a separate AI playbook.

In practice:

• Keep your site technically clean and fast.

• Write from your actual experience and expertise — the kind of content that can only come from you.

• Structure your pages for humans first.

• Resist the “GEO hacks” being sold around llms.txt and content chunking.

• Stay aware of where agentic search is heading.

If your foundations are solid and you write for your target audience by adding genuine value — you’re already optimising for generative AI search. The work isn’t different. It’s the same work, done well.

---

---

Frequently asked questions

• Is SEO dead because of AI search? No. Google’s official guidance is that generative AI features in Search are built on the same ranking and quality systems as classic Search. Solid SEO foundations remain the best way to be visible in AI Overviews and AI Mode [1].

• What is the difference between SEO, AEO and GEO? From Google’s perspective, there isn’t one. AEO (answer engine optimisation) and GEO (generative engine optimisation) describe the same goal — being visible in AI-powered search experiences — but the underlying work is SEO [1].

• Do I need an llms.txt file for my website? Not for Google Search. Google does not require or give special treatment to llms.txt files for visibility in generative AI features [1]. It may still be useful for some third-party tools, but it shouldn’t be your priority.

• Should I “chunk” my content into smaller pieces for AI? No. Google’s systems understand multiple topics within a single page. There is no ideal page length — focus on writing for your readers [1].

• Does structured data still matter? Structured data isn’t required for generative AI search, but it remains useful for rich results in classic Google Search. It’s worth keeping in your SEO toolkit, just not at the expense of strong content and a clean technical foundation [1].

• What should I do first if I want to be visible in AI search? Audit the basics — indexing, page experience and content quality. Then ask the simple test Google itself suggests: would my visitors find this content satisfying? If yes, you’re on the right track [1, 3].

---

References

[1] Google Search Central. (2026). Optimising your website for generative AI features on Google Search. Last updated 15 May 2026. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide

[2] Google Search Central. SEO Starter Guide. https://developers.google.com/search/docs/fundamentals/seo-starter-guide

[3] Google Search Central. Creating helpful, reliable, people-first content. https://developers.google.com/search/docs/fundamentals/creating-helpful-content

[4] Google Search Central. Spam policies for Google web search. https://developers.google.com/search/docs/essentials/spam-policies

This article is part of the Strategy360 newsletter — provides strategic communication insights for scientists, researchers, and technical professionals building their digital presence with purpose.

Go further

1/ Explore how I help scientists build strategic communication to showcase their experience and build their authority online → Visit tiffanysiegel-sci.com
2/ Follow along for weekly strategic insights → Connect on LinkedIn

If you’re interested to learn how to use AI as a tool to enhance your writing and communication while staying true to yourself - read this:

---

Thank you for reading! ⇢ Did this resonate? A ♡ helps more entrepreneurs, founders, scientists and researchers find this newsletter.