The answer comes down to two questions, to be taken in this order. One: when your customers ask ChatGPT, Claude or Gemini who to recommend in your field, do you appear — correctly, and in a good position? Two: how do you bring AI into your operations without stacking up subscriptions you don't control? Everything a Swiss SME must decide about AI in 2026 falls under one of these two questions. This note serves as an entry point: it lays out the method and points to the detailed notes.
Why two questions, not one "AI strategy"?
I spent fifteen years in the digital departments of large international groups. I saw many "strategies" born there: elegant, fifty-page, structuring documents, often buried six months later. An SME with five to fifty employees has neither the time to write these documents nor the means to be wrong for long. It needs decisions, not house doctrine.
The decisions that matter naturally split into two movements. The first concerns your demand: how your future customers find you, compare you, and choose you — a journey that now partly runs through generated answers. The second concerns your operations: how you produce, classify, respond, invoice — and the place AI can take there without creating new dependencies. The two movements don't mobilise the same projects, but they are measured and arbitrated the same way: first a stocktake, then decisions ranked by return.
First movement: being cited and chosen when demand is expressed
On complex queries — comparing providers, preparing a purchase, understanding a topic — the generated answer often precedes the consultation of links, and Google I/O 2026 confirmed the extension of this regime with AI Mode and the first search agents[2]. The user reads the summary, remembers the cited sources, and doesn't always scroll further down. A company that is properly indexed but editorially weak can therefore hold its Google positions while remaining absent from generated answers on its strategic queries. I detailed this shift in the note on business visibility after May 2026.
One doctrinal point avoids spending in the wrong place. On 15 May 2026, Google Search Central published its page on optimising for generative features, filed among the fundamentals alongside the SEO guide[1]: generated answers rely on the core ranking and quality systems of classic search. In other words, there is no separate discipline to buy — I explain in GEO vs SEO why this opposition no longer has a doctrine to support it. What changes is the severity: in an answer that cites a few sources, mediocre content no longer gets through.
Concretely, this movement rests on two projects. First, a site that models can read and cite: useful content, clean structure, honest structured data, sharp answers to the questions your customers actually ask — this is the core of my work in website creation and redesign. Then a measurement, because you don't steer what you don't measure.
The GEO Score™ is an aggregated measure across five generative models — ChatGPT, Claude, Perplexity, Gemini, Mistral — built on query baskets anchored in your business and a seven-theme grid; it produces a score out of 100 and a detailed profile that management can track over time. The full method is set out in MCVA Cahier No. 1 and on the GEO Score™ page. The instrument doesn't promise a breakthrough: it documents where you appear, where you don't, and what explains the gap.
Second movement: working with AI without multiplying dependencies
The same technological shift that changed search has changed the cost of software. Open models now run on ordinary business hardware, and assisted development produces in days what used to take weeks. Direct consequence for an SME: the trade-off between the SaaS subscription and the owned tool deserves to be reconsidered from scratch. I documented a real, complete case — FiscalDoc, a local application that replaced a tax document-management subscription — and a broader analysis in SaaS versus custom-built in Switzerland.
The movement doesn't mean bringing everything in-house. Some subscription tools remain the right choice: real network effects, large-scale collaboration, functional complexity that exceeds what a small team can handle. The movement is about taking back control of the trade-off: inventory what you pay for, identify the functions actually used, spot the data that would benefit from staying within a scope controlled under Swiss law, and qualify what belongs to owned custom-built solutions — your own code, your own data, contractual reversibility.
In what order should you move?
The method I apply comes in three steps, and it always starts with measurement.
First step: measure citability. The GEO Score™ pre-audit is free and delivers a quantified answer within a few days — you know where you appear in the five models' answers, and where you're absent. That's the factual answer to the first question, before any commitment.
Second step: map the flows. The AI Usage Diagnostic takes sixty minutes and covers both movements — your demand on one side, your operations and software stack on the other. It produces a prioritisation: what deserves a project, what can wait, what justifies nothing.
Third step: build, then maintain. Depending on the diagnostic, the project takes the form of a site designed to be cited or a custom-built solution — and in both cases, the work is ongoing, following the Diagnose, Build, Maintain logic. Sometimes the diagnostic concludes there's nothing to build this year. That outcome is worth as much as a quote.
What I advise against
Three expenses come up often, and I advise against them in this order. The reflex redesign: redoing the site because it "looks old", without having measured what the models retain from it — you spend in the wrong place. Buying a "GEO discipline" sold as a new profession: the 15 May 2026 doctrine removed the foundation from that narrative, and what was sold under that name belongs to the fundamentals applied seriously. Gadget training: a half-day workshop with no real use case changes no practice — training a team on AI plays out on concrete workflows, not on demonstrations.
Conversely, the least risky expense is the one that produces knowledge: a measurement, an inventory, a diagnostic. It illuminates all the ones that follow.
Key takeaways
— Two questions frame 2026: are you cited and chosen by AI when your customers ask, and how do you work with AI without stacking dependencies. — Google's 15 May 2026 doctrine confirms there's no new discipline to buy: there are fundamentals to apply rigorously, and a measurement to add. — The order of operations: measure (GEO Score™ pre-audit, free), map the flows (AI Usage Diagnostic, sixty minutes), then build and maintain.
FAQ
Where do I start if I have neither time nor a dedicated team? With measurement, because it doesn't require anyone on your side. The GEO Score™ pre-audit takes two minutes to request and the answer arrives within a few days, quantified. You then decide, based on facts, whether a more complete diagnostic is warranted.
My SME is small and local: does this concern me? Yes, and often more so than a large organisation. On a local query — a tradesperson, an accounting firm, a hotel in Valais — the generated answer cites two or three names. Being part of that or not makes a direct commercial difference, and the competitors cited aren't always the biggest.
Is GEO a profession I need to buy? No. Since 15 May 2026, Google's doctrine places optimisation for generative features among the fundamentals of search. What deserves investment is the serious application of these fundamentals and a citability measurement tracked over time — not a rebranded discipline.
What should I do with my current SaaS subscriptions? Nothing in a rush. Inventory what you pay for, what you actually use, and where your data sits. The trade-off between keeping, renegotiating and replacing with an owned tool arises function by function — that's precisely what the AI Usage Diagnostic lays out.
Want a quantified answer to the first question? The GEO Score™ pre-audit measures, free of charge, where your company appears — and doesn't appear — in the answers of ChatGPT, Claude, Perplexity, Gemini and Mistral. Quantified answer within a few days. Request a free pre-audit
Sources
[1] Google Search Central, Optimizing your website for generative AI features on Google Search, 15 May 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide [↩]
[2] Google, Google Search's I/O 2026 updates: AI agents and more, 19 May 2026. blog.google/products-and-platforms/products/search/search-io-2026/ [↩]
Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.