Does technical AI readiness predict AI search visibility for Danish businesses?

The benchmark scored 50 Danish businesses on two independent instruments, a technical readiness score and an AI visibility score, then tested whether they move together. The sample was stratified so no single sector dominated, spanning København, Aarhus, Aalborg, Odense, Vejle and Fyn.

Readiness scored each site on 12 objective signals (1, 0 or not applicable): structured data presence, Organization or LocalBusiness schema, FAQPage schema, Product schema, breadcrumb schema, robots.txt, whether robots.txt allows the AI crawlers (GPTBot, ClaudeBot, PerplexityBot, CCBot), llms.txt, sitemap.xml, title and meta description, mobile viewport, and HTTPS. Readiness % is checks passed divided by checks that applied.

Two measurement choices matter. First, the checker reads only the server sent raw HTML and does not run JavaScript, because the major AI crawlers do not run JavaScript either, so schema injected on the client is invisible to them and is correctly scored as absent. Second, it reads one page per business, the homepage. That second choice is why 0 of 12 ecommerce homepages exposed Product schema: product markup lives on product pages, which this single page scan did not visit, not on the homepage. A separate product page spot check (same raw HTML method) found that most of these retailers do publish Product schema on their product pages, so the 0% is a homepage scope figure and should not be read as evidence that Danish ecommerce lacks product markup.

Visibility used 3 customer style Danish queries per business, written with category phrasing and never the brand name (for example "bedste smørrebrød i København", not the restaurant's name), so the test measured genuine discovery rather than echoing the brand. Each query ran on two engines, ChatGPT (GPT-4o with web search) and Google Gemini (gemini-2.5-flash with Google Search grounding, a proxy for AI Overviews). A business counts as named when a model based judge confirms it was cited or recommended in the answer. Visibility % is times named divided by checks run. Perplexity was not run, so visibility reflects two engines.

Danish businesses have near universal web plumbing but a thin AI legibility layer. HTTPS, mobile, crawlability and sitemaps are almost everywhere, while the machine readable identity signals that help an AI understand a business are mostly missing. Average readiness was 61.9% (median 61%, range 36% to 82%).

AI engines named the Danish businesses in 52% of checks on average, and Google Gemini named them more often than ChatGPT. Gemini, with live Google Search grounding, cited a business in 57% of queries (85 of 149), against ChatGPT's 47% (70 of 150). Six businesses were named in every query (Geranium, Aamanns, Falsled Kro, Matas, Proshop and Nemlig) and five in none (DAHL, Roesgaard, Falck, Kemp og Lauritzen, Caverion).

On its own, AI readiness was a weak predictor of AI visibility. At n = 50 the correlation was r = 0.12, not statistically significant (p is roughly 0.4), and the top half on readiness was named 1.17 times as often as the bottom half. Read that correctly: readiness is the eligibility test, not the tiebreaker. It gets a business into the set an engine can cite; it does not decide which one the engine picks. What does decide it is in the next section — and the technical work that moves the needle is the legibility work, not the checklist.

Sector and brand authority drove AI visibility far more than technical readiness. Visibility swung from 72% for ecommerce down to 31% for local services while average readiness barely moved (59% to 66%). Established consumer brands like Matas, Nemlig and Noma get named whatever their markup, and small business to business service firms struggle even when technically tidy.

Not all readiness is equal — and the parts that make a business machine legible did track higher AI visibility, even though the blunt composite score did not. Sites carrying machine readable identity and content (an llms.txt, Organization or LocalBusiness schema, any structured data) were named more often than those without. The composite was diluted by universal signals everyone already has (HTTPS, mobile) and by the brand authority confound. So the right technical work still matters; it just has to be the legibility work that helps an engine understand and trust a business, not a generic checklist.

The llms.txt group is only 5 sites, so treat that number as directional, not settled.

For Danish businesses the practical lesson is that technical AI readiness is necessary but not sufficient. Structured data and crawler access help an AI understand and trust a business, but they cannot manufacture the third party signals (reviews, mentions, brand search) that make an engine choose to name it. Fix the cheap, widely missing AI legibility layer first, then pair it with brand authority work, because schema alone will not overcome a weak brand or a hard category. The biggest opportunity sits where readiness is low and the brand is already real: the technical fix is cheap and the authority is there to convert it.

And note who this sample was: established, mid to large brands that already carry authority. For the smaller, lower authority businesses that actually hire help getting found by AI, technical legibility likely matters more than these averages suggest — for them it is often the difference between an engine being able to cite you at all and being invisible. Track per engine too, since grounded Gemini and ChatGPT behave differently and one combined number hides that.

The clearest limitation is that the sample leans toward established, mid to large businesses, which likely understates how much readiness matters for the low authority SMBs that most AI-visibility clients actually are. Because brand authority was the strongest driver we found, a sample weighted toward genuine SMBs is the priority follow up. The rest: