The GEO Diaries
On the GEO front:
This is where I work on Generative Engine Optimization questions, one idea at a time, dropping notes “every so often”. And I call it “the GEO front” because the term and related practices are in the painful process of getting defined (or rejected, and swapped) through power plays between platforms, SEO practitioners and new vendors, while academia is overshadowed by industry. Beyond best practices, I’m watching the weird basement where knowledge now gets processed: how AI is changing the path information takes before it reaches us, and what gets … well, lost in translation.
Note 1: "GEO, still SEO": Google’s very convenient truth.
It’s May 26, 2026, and Google’s search bar has never been wider.
A week has passed since Google announced the biggest upgrade to its Search box in more than 25 years, and the news feels surprisingly slow to land with users. The bar has expanded to accommodate longer questions beyond keywords, and an AI experience that ushers us into a new era where the 10 blue-links, while not dead, will start to look like archeology. A wink to the past, and an apt one, since for decades, searching was an act of excavation: we dug, compared, selected, doubted, opened tabs, closed tabs, and built a response for ourselves. Now, preferably it seems, you just ask.
But half of the story is elsewhere. Shortly before the new interface dropped, Google published a document that landed hard in SEO and marketing circles. It finally addressed the inevitable question: how do you optimize for that thing, how do you get cited by AI (and also, how do you not cry over traffic crashing)?
To this, Google has a very convenient answer (for itself), and for SEOs who have every reason to feel the ground beneath them slipping: GEO is still SEO. There will be, Google adds, a great deal of myth-busting around GEO. And it knows its documentation functions as de facto industry standard for a large part of the SEO world.
Looking at it closely:
“AEO” stands for “answer engine optimization” and “GEO” for “generative engine optimization.” These are both terms you may see used to describe work specifically focused on improving visibility in AI search experiences. From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
Google spent a decade making SEO mean whatever Google wanted it to mean. When you are the search engine, the optimization vocabulary is yours by default. Nobody had to strategically claim it, the monopoly simply handed it over.
Now Google is publishing reassuring guidelines basically saying: don’t panic, the world did not change that much.
Google is not wrong about Google: AI Overviews and AI Mode do run RAG (Retrieval Augmented Generation) against Google's own index using Google's ranking systems. Inside that pipeline, SEO rules hold. So they are being honest about their own plumbing.
Is GEO just the continuing pursuit of SEO through other means?
The problem is that Google is no longer the only building on the street: ChatGPT does not sit on Google's index. It may draw on Bing’s, but also relies on OpenAI's own. Not to mention the rest of the players. Claude runs its own crawler infrastructure and routes live search through Brave. Perplexity retrieves differently, cites more aggressively, and weights sources by a logic not fully documented publicly. DeepSeek, Kimi, Grok, Mistral and others all point toward the same future: not one search layer, but many answer environments.
And then there is the training-time question, which is different from live retrieval entirely. Whether a model can talk about your organisation or you or whatever you do without searching is partly a function of corpus representation, in other words what existed in the data at the time of training and later model updates. That means Wikipedia presence, GitHub footprint, academic press, Reddit-oh-reddit, news archives (oops), and other public or licensed traces. That is a visibility problem with different levers from snippet eligibility, and Google does not mention it because Google is not designed to help you with Claude.
"Still SEO" is not a simple technical claim. It also turns the mess into a familiar category. Folding back GEO and AEO into SEO makes the transition easier to understand, and easier to contain. “Still SEO” is a strategic framing, and for Google, perhaps something close to a sovereignty claim.
The retrieval layer is fragmenting. Google still matters, enormously. Nobody can afford to be invisible on the search engine, but nobody can afford to optimize for one provider and call it done. The tactical debates are already here. We may discuss at length whether chunking helps or llms.txt matters or markdown formatting moves the needle. Google says it doesn’t. Microsoft says otherwise. We may argue, and we probably should, but the underlying question remains the same: who gets to set the rules when there is no single landlord anymore.
The new visibility problem cannot be solved by pretending every answer engine is just a Google-shaped machine. It will require understanding how different systems retrieve, cite, remember, and ignore.
“Still SEO” is true inside Google. Outside Google, it starts to look like a border dispute, and the border is moving.
Note 2: Europe takes on the answer machine.
Meanwhile in Europe …
… as the answer layer continues to fragment, regulation catches up. There is a first time for everything, and now it is ChatGPT's turn. Is a chatbot that can search the web in real time a search engine? You bet, says Europe.
On August 31, the European Commission announced the designation of ChatGPT as a Very Large Online Search Engine (VLOSE), a hybrid service that can answer prompts by querying the live web, bringing it under the same regulatory roof as Google Search and Microsoft Bing.
With 159.1 million average monthly active users in the EU, ChatGPT’s search function sits well above the 45 million-user threshold set by the Digital Services Act to trigger the designation. In practical terms, OpenAI will have to comply within 4 months of notification with the obligations imposed on VLOSEs, bringing us roughly to the turn of 2027.
At any rate, if that means zeroing in on systemic risks, the stakes are well worth following. It is easy, and painfully concerning in some respects, to see how AI is fast becoming the default intermediary in how knowledge and information are surfaced. The risks involved are not limited to outright falsehoods: take for instance hallucinations, overconfidence and persuasive framing, irrelevant sources, misattribution, sloppy inference, overgeneralisation or simply dropping the qualifiers that would preserve the nuances of an argument… all those things that were not really “a thing” in the lost Kingdom of blue links, where you were free to wade through a spread of results (and draw your conclusions).
Perhaps for good reason, I have grown interested in AI representation as a challenge potentially more consequential than the narrower question of who gets visibility in an answer. At scale and in aggregate, those small distortions can become a systemic risk, while none of them looks particularly spectacular in isolation.
But compliance with EU regulations may prove more bounded than the diffuse epistemic issue. Under the Digital Services Act, OpenAI is expected to "assess and mitigate systemic risks stemming from [its] service and algorithmic systems". The European Commission specifically points to risks involving "illegal content, negative effects on minors, users' physical and mental well-being, fundamental rights, electoral processes and public security”. Either the taxonomy has no slot for epistemic distortion, or it must be squeezed into the categories of fundamental rights and electoral processes.
(Entry to be continued)
