What the agents read

On May 15, 2026, in the Google Search Central documentation, under a newly created tab that would later come to be called Generative AI fundamentals, a document of a few thousand words appeared, titled Optimizing your website for generative AI features on Google Search. Neither an op-ed nor a product announcement. A documentation page, filed next to the SEO Starter Guide, that cleanly closes two years of uncertainty. Four days later, in Mountain View, the Google I/O of May 19 and 20 rolled out AI Mode at scale, extended Personal Intelligence to nearly two hundred countries, and embedded information agents in the background of search. The following day, Google announced a new Core Update whose rollout, as I write, is just beginning. The sequence is not incidental. It tells us where the traffic of the next eighteen months will go, and therefore what purpose remains in writing on the web at all.

I want to set down here what this sequence changes for Swiss companies, and in particular for the Valais actors who — as the book argues — have a short window to seize in the shift now underway.

What the document says

The guide rests on three propositions, stated without indirection. The first is technical. Google Search's AI features — AI Overviews at the top of the results pages, AI Mode as a conversational interface, agents that crawl the web in the background to prepare concrete actions — do not run on a separate index. They draw from the standard Search index via retrieval-augmented generation and query fan-out: a language model breaks the query into sub-questions, retrieves the relevant passages from pages already ranked by the usual systems, and composes the answer from those excerpts. If your content was not indexable, crawlable, and deemed of quality by yesterday's systems, it will be no more so for today's.

The second proposition is marketing-oriented, and it explicitly targets the agencies and their market. Everything that has been sold since late 2024 under the name GEO — Generative Engine Optimization — or AEO — Answer Engine Optimization — as a framework distinct from SEO rests, according to Google, on nothing technically specific. The same sentence, repeated under different angles throughout the document, shuts the door on parallel frameworks. There is no new discipline; there is an older discipline whose requirements have just moved up a notch.

The third proposition is more brutal. A section titled Mythbusting generative AI search names the techniques to abandon. The llms.txt files — an open standard pushed by part of the technical community to signal to models what they may ingest — are treated by Google like any other text file, with no indexing privilege. Artificially carving content into chunks meant to ease ingestion by LLMs adds nothing: Google's systems decide on their own what they retain from a multi-topic page. The "AI-specific" Schema.org markup, rewrites designed to please the models rather than the readers, paid mentions on third-party sites intended to "train" the AIs — all of this is expressly filed under the useless. The guide does not call for renouncing such practices on principle. It simply notes that they do not serve the cause pursued.

In their place, the document recentres four requirements. Valuable, unique, non-commodity content — the first of these adjectives leading the rest. First-hand expertise, that is, lived, situated, signed by an identifiable human. A technically solid site — accessible, fast, readable by humans as much as by machines. And the E-E-A-T framework, broadened in 2022 — Experience, Expertise, Authoritativeness, Trustworthiness — whose first letter, Experience, becomes the real differentiator now that generative models frame how things are read.

What is closing

To understand what this announcement closes, one must first name what had opened.

Over the past twenty years, organic search has gradually industrialised. A significant share of the content published on the web was published for the sole purpose of capturing a position on a keyword. Aggregator sites, content farms, mass-generated pages around query variants, FAQs stuffed with questions no one ever asked, thousand-word articles whose first eight hundred words serve only to reach the minimum length presumed by the algorithm before the answer finally arrives — all of this has constituted an industry. That industry has employed tens of thousands of people, fed agencies, sustained a nebula of tools, and trained an entire generation of writers to produce text that does not address itself to human beings.

The arrival of accessible language models in 2023 multiplied this industry. Producing a thousand SEO pages cost next to nothing. The outcome was predictable: the web filled up with plausible, empty text, production costs fell, competition rose, and positions were settled in a race the most disciplined actors no longer won.

What is closing, then, is not SEO as a discipline. What is closing is the idea that one could, by industrialising content production, durably capture a share of organic attention without holding any real substance.

What is opening

What is opening, in return, is more demanding and more just.

The retrieval systems that feed the AI Overviews must, by construction, choose which passages to cite. With every query, they select a handful of sources out of millions. They no longer concern themselves with linear ranking: there is no longer a page 1, page 2, page 3. There is the answer, and the two or three sources that fed it. To exist in this new geography of attention, a site must produce what the SEO literature now calls information gain: content that adds something the other sources do not have. An original analysis, a previously unpublished figure, a situated observation, a lived experience, an interpretive frame that exists nowhere else on the web.

This criterion is not reachable by industrial production. It is reachable, all else being equal, by practitioners who speak from inside their subject.

It is here that the link to this essay's thesis becomes direct. The competitive shift I describe in chapter five — the partial erosion of the urban concentration of skilled work, through the leverage of AI tools — has an editorial counterpart that must be named. For twenty years, industrial SEO had neutralised part of the advantage that expert practitioners of a subject held over content producers with no matter of their own. A large agency could publish a thousand pages on Valais tax law without ever having advised a local client, and capture a substantial share of the corresponding queries. This was played out at the level of search engines: on positions, on internal linking, on accumulated domain authority. The four-person trust firm in Sierre, holder of real expertise but with no SEO budget, did not figure in it.

What the sequence of May 15 to 21, 2026 sets in motion is, to my reading, the partial inversion of this mechanic. The generative models that synthesise the answers need sources where the matter is dense. They favour, at equal technical quality, content that brings something irreducible: a lived case, a local figure, an original line of reasoning, an identifiable signature. The Sierre trust firm that has been publishing, for three years, signed notes on the taxation of second homes in tourist zones now finds itself, in 2026, in a position its means had never afforded it. It can be cited. Not for its size. For its matter.

There remains a too-comfortable reading to be guarded against. The example of the trust firm publishing for three years presupposes precisely what most SMEs lack: a writing discipline established before the window opens. Generative models do not, in a single day, distinguish the situated practitioner from the opportunistic newcomer. They identify, through authority signals distributed over time, those who have been writing on their subject for a while. The window this sequence opens is therefore not mechanical. It presupposes a run-up. SMEs that begin in 2026 — signed and patient — will reap the benefit in eighteen to twenty-four months. Those who wait until others have proven the return will arrive, as so often, when the positions are already frozen.

What companies must do — and stop doing

This observation must be translated into operational instructions. In the following terms.

The first front is content production. Anything outsourced to an agency whose deliverable is called a "1,500-word SEO article" should be reconsidered. Not because such agencies are dishonest — many do clean work — but because the commodity they produce is precisely what Google now signals it no longer wishes to promote. The opposite direction is clear: have practitioners write, in their own voice, on what they actually know, signing by name. This requires reorganising the internal production chain. It also requires accepting to publish less and longer. It is not a comfort. It is a budget reallocation.

The second front is editorial discipline. The content that AI systems will retain in 2026 meets a few testable criteria. It is visibly dated, because a recent answer prevails over an older one. It is signed by an identifiable person, because the models evaluate the Experience of the producer. It cites its sources, because the models distinguish assertion from rumour. It takes a position when it is honest to do so, because the information gain of a text that never cuts is zero. It owns its language, because a text uniformised by a rewriting tool ends up losing the mark that the models read as a signal of authenticity. And it avoids the patterns that agencies have over-represented these past eighteen months: artificial FAQs, bullet lists parachuted into contexts that do not call for them, marketing chunking, opening formulas in which one announces what one is about to say rather than saying it.

The third front is measurement. The classic SEO tools — positions, search volumes, backlinks — remain useful but no longer suffice. One must start instrumenting visibility inside AI answers: tracking citations in Google's AI Overviews, in the responses of ChatGPT, Claude, Perplexity, Mistral. Several platforms offer this monitoring. None of them is yet perfect. That does not exempt one from looking. The metric that counts is no longer only the click — which erodes as AI answers take up the page — but the citation, and the authority that citation builds through repetition.

Set apart, a task of a different order: editorial sovereignty. The sequence of May 15 has a subtext that needs naming. Google explains that the llms.txt files are of no use to it — which is technically accurate for Google Search. But this standard remains less a ranking lever than an act of documentary declaration: a way, for a site, to state what it considers readable, stable, citable. Whether the large models honour it tomorrow or not, the act of writing it and publishing it in the open belongs to whoever runs the site. It is not a technical argument. It is a political act. And this site, which publishes one, knows it.

What this entails for the Valais

I have written elsewhere that the territorial window of AI would open in favour of those actors capable of orchestrating the shift rather than enduring it. The sequence of May 15, 2026 is, on my reading, one of the clearest manifestations of that shift applied to the visibility of companies on the web.

For Valais SMEs — trust firms, lawyers, doctors, wineries, craftspeople, tourism operators, business advisors — a path reopens. Not by imitating the SEO agencies of Zurich or Geneva. By doing the opposite. By publishing what they actually know, in a voice that belongs to them, at a sustainable pace. One note a month — signed, dated, situated, on a subject they truly command — is now worth more than ten generic articles produced in series. This is not a consolation prize for the smaller players. It is their comparative advantage reasserting itself.

For the canton itself, as an institution, the stake is twin. Public content — that of the administrations, of the universities of applied sciences, of the bourgeoisies, of the professional associations — is precisely what AI models cite willingly, because it carries a situated expertise and a domain authority that few commercial actors can match. There lies an under-exploited asset, one that calls not for a classic communications budget but for a rigorous editorial discipline and an explicit policy of openness toward the AI crawlers of the publishers by whom one consents to be read.

In the background remains the political question this site tries to keep open. When AI agents fetch what we write, they hand it back to their users without those users ever visiting our sites. The "free content financed by click advertising" business model is fading, page by page, as the answers form outside our pages. What territories and businesses now derive from their content is less monetisable traffic than distributed authority. That authority has a value. It does not yet have a price.

This is precisely the wager of this site, and the reason its robots.txt explicitly authorises by name the agents that will come and read it. To be cited, in 2026, is not an advertising win. It is an inscription in the conversation these models hold in our stead. At this level of stakes, the question is no longer one of clever tagging. It is a matter of having, on one's subject, something to say that no one else has written.