The Dawn of the Living Dead
Because web agencies that do not speak to AIs are already dead — and so are their clients.
There is something unsettling in a world where everything still seems to work. The old SEO brings traffic. Google indexes the pages. Agencies sell campaigns. Clients pay. The cycle goes on. But underneath, something has already changed. And those who have not seen it yet are not late — they are already out.
The problem no one sees because it doesn’t hurt yet.
In 2024 Google introduced AI Overview in search results. By 2025 it became mainstream. In 2026 millions of people ask ChatGPT, Claude, Gemini, and Perplexity — not Google — what the best solution for their problem is. And these AIs do not read your site like Google does. Google reads your pages, counts the links, evaluates the keywords. It’s a system you know, that you can optimize, that you can beat with budget and patience. An AI agent does something different. It looks for your structured files — llms.txt, llms-full.txt, the JSON-LD graph. It reads them. It reconstructs who you are, what you do, for whom you do it. And if it doesn’t find them, or finds them empty, or finds them inconsistent — it infers. Inference is the point. When an AI does not find enough structured data, it doesn’t stop. It invents. It fills the gaps with statistical patterns. And the result is a poorly reconstructed corporate identity, distributed to millions of queries, impossible to correct later.
What happens concretely when the AI searches for you?
Let’s take a real local SME. A plumbing distributor in Bologna. It bills 1.5 million, has 8 employees, is the exclusive distributor for two major brands in the area. No llms.txt, no JSON-LD, a well-built site but semantically mute. A potential customer asks ChatGPT: “Find me a local distributor in Bologna that has condensing boilers of brand X in stock and sells to VAT number.” The AI faces an informational blackout. And it reacts in three wrong ways: The cousin effect. The company is called “Idraulica Bolognese S.r.l.” but has no semantic data specifying the brands dealt with. The AI statistically associates it with any other plumbing company in Bologna. Result: it could tell the customer that this SME sells a brand that actually belongs to a competitor. The geographical hallucination. Without a clear PostalAddress node in the code, the AI confuses the legal office — perhaps at a commercialist’s office — with the operational showroom. The customer gets sent to the wrong place. Brand devaluation. In the absence of certain data, the AI responds generically: “It’s a company that deals with trade…” — depriving the SME of its real strengths: exclusive distributor, 24-hour delivery, specialized technical consulting. Three clients lost. For a JSON-LD file that does not exist.
The old SEO is the problem, not the solution.
The paradox is this: the old SEO works so well that it hides the problem. An agency that optimizes well for Google brings traffic. The client is satisfied. The reports show growth. No one has reason to look beyond. Meanwhile, every day that passes without an AI-readable structure is a day in which the competitor who has done their homework ranks better in AI agents’ responses. The gap is not immediately visible. It accumulates. Silently. Until one day the client notices that leads are decreasing. That when they ask ChatGPT “what’s the best supplier of X in my area” — they don’t appear. The competitor appears. And by that time it’s already too late to catch up.
The trap of intent searches.
Users — even in B2B — no longer search just on Google by typing keywords. They use AI to perform complex screenings: “Find me a local distributor in Bologna that has condensing boilers of brand X in stock and sells to VAT number.” This is an intent search. Specific, commercial, ready to buy. If the SME has not structured its data with JSON-LD using schemas like LocalBusiness, WholesaleStore and properties like knowsAbout or offers — the AI will never find it. Literally. It’s not a positioning problem. It’s an existence problem. And while the local SME does not exist for AI agents, large marketplaces and national distribution giants — who have SEO teams that meticulously take care of semantics — are proposed to the customer next door. The business model based on proximity and local consulting silently dissolves.
What is concretely needed.
For a local marketing SME, just a few Schema.org tags are needed, but the right ones:
[Your SME] │ ├── @type: "WholesaleStore" or "LocalBusiness" │ → Tells the AI who you are │ ├── areaServed: "Emilia-Romagna" │ → Tells the AI where you operate │ ├── brands: ["Brand A", "Brand B"] │ → Tells the AI what you exactly sell │ └── priceRange: "$" → Helps qualify the targetFor a large organization, semantics is branding.For a local SME, semantics is the only way to be found by AIs when a local customer is ready to buy.
Agencies are not ready — and it’s not their fault.
It’s not a critique of agencies. It’s a snapshot of the market. Those who have built their competence on keyword research, backlink building, on-page optimization — have sharpened tools for a problem that is becoming secondary. The llms.txt files have existed as standard for less than two years. The JSON-LD @graph protocol with linked nodes is still a niche. Most agencies have never heard of it. The problem is not malevolence. It’s the natural latency between the emergence of a technology and its adoption in the market. Except this time, the latency has an invisible cost that is paid in the future — and the future has already arrived.
The cemetery is silent. The living dead in the title are not the failed agencies. They are the agencies that work, bill, produce results — and do not know that it is
👉 lake8.dev/lagotto-meter — gratis
Sources
Author: Fogli Giantommaso
Publication Date: 2026-07-01
Rights and Attribution
Images, logos, and photographs are the property of their respective owners. Used for commentary purposes.
← Back to journal