By Gecko Studio · updated October 2026
On 15 May 2026 Google published on developers.google.com/search its official guide to optimising for AI Overviews and AI Mode. The headline is uncomfortable for part of the industry: there is no new discipline. What many people called GEO (Generative Engine Optimization) or AEO (Answer Engine Optimization) does not exist as a separate category according to Google. It is SEO.
The guide's wording is direct: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO". That sentence sinks the separate offering of several agencies and consultants who have spent two years selling "GEO packages" as if they were services distinct from traditional SEO.
And the guide does not stop there. It has a whole section titled "Mythbusting generative AI search" debunking practices that have been sold as mandatory for appearing in AI Overviews. We go through the seven that circulate most.
What exactly Google's official guide says
The page /search/docs/fundamentals/ai-optimization-guide starts from a technical premise worth having clear before we get to the myths.
Google's AI Overviews and AI Mode draw on the same search index as the classic results page. The guide puts it this way: "These features rely on AI techniques to highlight content from our Search index". There is no "parallel index" for AI. There is no independent retrieval system. If a page ranks well in Search, it is eligible to be cited in AI Overviews. If it is not indexed, it does not appear. Full stop.
The practical corollary: SEO best practice stays the same. Crawlability, useful content, page experience, topical authority. What changes is the shape of the result, not the system that decides which result deserves to be there.
The seven myths Google debunks
1. Creating llms.txt files or other special markup for AI
Google is emphatic: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search". The llms.txt proposal (the equivalent of robots.txt for language models) has been circulating for months as good practice. Some agencies add it to their audits as a deliverable.
According to the official guide, it adds nothing. Google does not read it, and the main commercial models have not adopted it systematically either. If you add it, it does not penalise you, but it does not help.
2. "Chunking" content into small pieces
The idea of splitting content into short fragments so AI models can "digest it better" is another myth Google debunks: "There's no requirement to break your content into tiny pieces for AI to better understand it". Models process full semantic structure. Forcing articles into ultra-short lists or breaking paragraphs into loose sentences does not improve eligibility.
There is an important nuance: clear structure does help. Logical H2/H3 headings, paragraphs that make sense, lists where they apply. That is ordinary SEO, not chunking.
3. Rewriting content specifically "for AI"
Another direct quote: "You don't need to write in a specific way just for generative AI search". There is no "AI-friendly" style distinct from solid SEO writing. The tips circulating on social media about "short declarative sentences", "avoid aspirational content" or "reformulate as chained FAQs" have no official backing.
What AI (and Google) does reward is content with a unique perspective. The guide gives a clear example: a generic summary such as "7 tips for buying a flat" loses to a first-hand account such as "Why we skipped the survey and saved money". The originality of the experience is what an AI Overview cites, not the formal structure of the text.
4. Special Schema.org for AI
"Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add". There is no "AI-ready" schema. There are no new properties that add extra eligibility for AI Overviews.
A caveat: classic schemas (Product, LocalBusiness, Organization, Article and so on) remain useful because they generate rich results and reinforce the knowledge graph. But none is specifically designed for AI. The guide recommends using schema.org as part of general SEO strategy, not as a separate AI lever.
5. Seeking inauthentic brand mentions
Some GEO strategies recommend seeding brand mentions on forums, blogs and Reddit to "train" the models. The guide is clear: "Seeking inauthentic 'mentions' across the web isn't as helpful as it might seem".
The problem is twofold: forced mentions tend to look forced, which reduces their effectiveness even if models did read them. And breaking platform policies (spam in subreddits, fake comments on blogs) has a real reputational cost.
6. Mass-generating content variations
The idea of multiplying pages with automatic variations to cover the long tail (X in London / X in Manchester / X in Bristol, using an identical template) falls squarely under the scaled content abuse spam policy. Google has penalised this since March 2024 and the guide recalls it explicitly in the AI Search context.
It makes no difference whether generation is 100% human, 100% AI or a mix. What Google evaluates is the main purpose of the content: if it seems to exist to manipulate rankings rather than help the reader, it falls. Using AI to generate copy is not illegal; replicating pages with the same skeleton and a different city is.
7. Obsessing over long-tail keywords
"AI systems can understand synonyms and general meanings... you don't have to worry that you don't have enough 'long-tail' keywords". Models understand synonyms and related concepts. The old tactic of listing 200 long-tail variants to cover every way of asking the same thing is unnecessary.
What does matter is topical depth: covering the subject with enough informational density for the model to extract the answer. That is measured by semantic coverage, not by number of keywords.
What Google does recommend for appearing in AI Overviews
The guide concentrates its positive recommendations in four blocks:
Content with a unique perspective. Not summaries of information available elsewhere. First-hand accounts, analysis with your own data, informed opinion backed by experience. This is the point the guide stresses most.
Standard technical quality. Indexable, crawlable, no blocks in robots.txt, semantic HTML, JavaScript that does not hide main content, reasonable latency, responsive. Nothing new.
Integrated multimedia. Relevant images and video increase the chance of being cited, because AI Overviews include visual elements in their answers. Here there is a change from classic SEO: video and images are no longer "decoration", they are a citation surface.
Adjacent business strategy. Google Merchant Center for products, Google Business Profile for local. These are structured sources that the AI consults directly. They are not SEO in the strict sense, but they are part of the assessable digital presence.
What Google says is not what applies in other AI engines
Google's guide covers only Google AI Overviews and AI Mode. But AI searches today are split across five main engines: Google AI Overviews, ChatGPT Search, Perplexity, Bing Copilot and, to a lesser extent, Claude (which does not yet have a public search product with citations). The immediate question is whether what Google debunks stays debunked in the others.
We cross-checked the seven practices against each engine's official position as of May 2026. (The comparison matrix we published in the original article is available in the Spanish version.) Sources: Google's official guide (15 May 2026), Bing Webmaster Guidelines (2026 revision), public OpenAI and Perplexity docs, and a server-log analysis by Kai Spriestersbach (Medium, April 2026).
The comparison shows two non-obvious findings:
Bing Copilot is Microsoft's exception to Google's doctrine. In February 2026 Microsoft rewrote the Bing Webmaster Guidelines, explicitly adding the term "Generative Engine Optimization (GEO)" as an official concept. They recommend block-level parsing with clear mini-titles on each block, exactly what Google calls "unnecessary chunking". Same practice, opposite stance.
llms.txt is widespread theatre. Server-log analyses show that GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended routinely ignore the /llms.txt file and crawl HTML directly. Stripe, Vercel, Cloudflare, Anthropic, Coinbase and Pinecone publish it anyway, but more as a cultural gesture to developers than as a real inclusion mechanism. The frankest quote circulated in April on Medium: "The llms.txt is dead. More precisely: a dud."
Other smaller divergences with operational impact:
- ChatGPT rewards recency 3.2x more than old content (content updated in the last 30 days). Google does not stress this factor as strongly.
- Perplexity requires an internal quality score of 0.75 or more to enter top placements, and gives a 3x multiplier to content on AI, science or marketing.
- 65% of ChatGPT citations come from third parties (Reddit, Quora, independent reviews, trade publications). This is not an invitation to buy mentions (still counterproductive) but it does reinforce the importance of legitimate digital PR.
- Bing launched its AI Performance Report in February inside Bing Webmaster Tools, showing how many times Copilot cites a URL. It is the first official AI citation metric available. Google has no public equivalent.
Why this guide matters more than it seems
The real impact of this publication is not technical. It is commercial.
For two years, part of the SEO industry has sold GEO audits, AEO packages, Perplexity optimisation consulting, and specific deliverables such as llms.txt. Some in good faith, others knowing it was smoke. The line between the two was hard to prove while Google had not spoken officially.
Now it has. And the position is that there is no separate discipline for Google. That puts any "extra GEO package" billed to a client who already pays for SEO in a service whose legitimacy is in question, at least as far as Google is concerned. Not illegal, but questionable.
For companies with an SEO audit commissioned from a serious agency: this changes nothing about work done well. For companies that paid for specific GEO/AEO packages in the last two years: a review conversation is reasonable.
The critical caveat, already visible in the comparison: Google's position is not the position of the whole industry. Microsoft does recognise GEO. The sensible strategy is to build on the common fundamentals (the ones all engines agree on) and treat each platform's specific recommendations as a fine-tuning layer, not a separate discipline.
How to apply this in a serious SEO audit
An audit that takes this guide seriously includes three specific reviews that were not there two years ago.
A review of deliverables sold as GEO/AEO. If any exist, assess whether they add real value (semantic structure, well-implemented schema, solid E-E-A-T content) or are decorative llms.txt and chunking. The first is useful SEO with a confusing commercial name. The second is an invoice with nothing behind it.
A multi-engine visibility review. Even though the guide is Google-centric, a serious AI Search audit measures the brand in at least four engines (Google AI Overviews, ChatGPT, Perplexity, Bing Copilot). Behaviour differs by engine, and practices do not transfer 1:1.
A review against recent spam policies. Site Reputation Abuse, Scaled Content Abuse, Expired Domain Abuse and Policy Circumvention are 2024-2026 policies that penalise practices some agencies still recommend out of habit. Auditing for them is fifteen minutes' work and avoids serious problems.
SEO optimisation for AI Search: what changes and what does not
What does not change: the basics. Indexability, useful content, topical authority, page experience, well-implemented schema. We have had twenty years of those pillars and they are still the pillars.
What changes: the shape of the result and the citation surface. You no longer compete only for the ten blue links; you compete to be cited in a generative paragraph that answers the user's question before they click. That calls for thinking about content originality more than keyword repetition, and about semantic coverage more than density.
What ends: the separate offering sold as "GEO" or "AEO" to clients who already pay for SEO. If the SEO is done well, you are already working for AI Search. If it is done badly, the problem will not be fixed by a fashionable extra package.
Google's full guide is at developers.google.com/search/docs/fundamentals/ai-optimization-guide, published on 15 May 2026. Open and without a paywall. It is worth reading in full before accepting the next quote that includes the word GEO in bold.
Four questions to ask before buying "SEO for AI"
If you are weighing a proposal sold as GEO, AEO or "SEO for AI", four questions separate a proposal with substance from one that has only changed its cover.
- What will they measure, and against what? An honest answer admits there is no Search Console for AI models: no dashboard tells you how many times you have been cited. Anyone who shows you an exact "AI visibility" percentage is showing you a tool's estimate, and should say so. What is measurable is something else: a set of real questions from your sector, asked periodically and always in the same way, and what comes out.
- What would they do differently from a normal audit? If the answer is clearer content, verifiable data, identifiable authorship and a site that crawlers can read, it is well-done SEO, and that is fine. If the answer is a special file or new markup, ask them to show you who uses it.
- Who decides what gets answered? Models do not cite in the same way: some rely on the search engine and others on their own sources. A serious proposal distinguishes between them instead of talking about "the AI" as if it were one thing.
- Have they checked that the crawlers can get in? This is the boring part and the one most often broken: if your server or provider blocks AI crawlers, everything else is moot. See why a site can be invisible to ChatGPT and how to check whether your brand appears.
The final test is simple: if the proposal still makes sense once you remove the acronyms, it is real work.
An SEO audit with AI Search visibility included
At Gecko Studio we have applied this approach since 2024: solid SEO as the base, measurement across four AI engines (Google, ChatGPT, Perplexity, Bing Copilot), and a transparent review of recent spam policies. No invented extra packages.
Want an audit that applies this standard? See our SEO audit, or get in touch or call +34 871 00 60 52.
