In June 2026 we audited 38 Danish SMB websites across five verticals and pushed each one through the same automated pipeline: HTML and answer-engine readability, schema coverage, PageSpeed, CrUX field data, AI citation checks, and agent readiness. The results are anonymized. No site is named, no owner was contacted. We call the dataset the Danish AI Visibility Index.
The headline finding fits in one sentence: Danish SMB sites are readable, not transactable. Average AI-visibility came in at 83 out of 100. These sites are not invisible to AI. An assistant like ChatGPT or Claude can find them, parse them, and describe what the business does. Years of decent CMS defaults and ordinary SEO work got them there, mostly by accident.
Agent-readiness tells the other half of the story. The average was 48 out of 100. Only 26 percent of the sites publish an llms.txt file. And the number that stopped us: 0 of 38 publish structured prices, availability, or opening hours. Not one.
Why the gap matters
The two scores measure two different futures. AI-visibility is about being described: when someone asks an assistant for a recommendation, can the model read your site well enough to mention you accurately? Most of these businesses pass that test today.
Agent-readiness is about being acted on. An AI agent booking a table, comparing quotes, or placing an order needs machine-readable answers to three questions: what does it cost, is it available, and when are you open. If those answers only exist as pixels in a hero image or text in a PDF menu, the agent moves on to a competitor that publishes them as data. Readable gets you mentioned. Transactable gets you chosen.
The pattern across verticals was consistent: prose is fine, structure is missing. Sites describe their offers in fluent paragraphs and then hide every number an agent would need. That is not a technology gap. Schema.org has supported offers, prices, and opening hours for over a decade. It is an intent gap: nobody asked the sites to be machines' counterparties until now.
What closing the gap looks like
The fixes are unglamorous and fast. JSON-LD that carries actual prices, not just the organization name. An llms.txt that tells AI crawlers what the site contains and what they may use. Crawler directives in robots.txt that welcome the agents you want. A machine-readable product or service feed, so an agent does not have to scrape your DOM and guess.
We ran this playbook on a live client site this week: a jewelry e-commerce brand in Casablanca. The before scores were 81.7 for AI-readability and 37 for agent-readiness, almost exactly the Index pattern. After one day of fixes (structured product data with real prices, llms.txt, robots directives, a products.json feed), the re-audit came back at 100.0 and 82. The design did not change. No pixel moved. The site simply started answering the questions agents ask.
Caveats, because they matter
38 sites is a small sample. One country. Five verticals. The scores come from our own audit methodology, which is documented at citetome.com, and other methodologies will weight things differently. We are not claiming Danish SMBs are losing measurable revenue to agents today; agent-driven commerce is still early. The claim is narrower and harder to dismiss: when an agent does come asking, 0 of 38 sites in this sample can currently answer with a price. That is fixable in days, and the businesses that fix it first will be the ones agents can choose.