Danish IT outsourcing: The hardest cut is a question nobody answers.

Somewhere in Denmark right now, an owner-manager with 40 employees and no IT department is asking an AI model which IT partner to trust. The model answers in seconds: a shortlist, the risks, a favourite. No provider hears about that conversation. We measure it.

What we measured

Since 7 August we have run the Danish IT-outsourcing market's buying questions through the models, across the decision journey from first comparison to advocacy. Three measurement days across nine days. Measured with three models: Gemini, Claude, and ChatGPT.

The questions are the ones buyers ask, in their own words:

• "Hvad koster det at outsource vores IT-drift?" (What does it cost to outsource our IT operations?)
• "Hvordan finder man en pålidelig IT-partner i Danmark?" (How do you find a reliable IT partner in Denmark?)
• "Hvad sker der, hvis IT-partneren ikke leverer?" (What happens if the IT partner fails to deliver?)

The judging sheet

The verdict: the default is also loved

How to read the table: the percentages show how frequently each brand appears in the answers at each stage of the decision journey, measured across the full window, 7 to 16 August. A dash means the sample for that stage is too small to report.

Sentia is the window's upward converter: 29% of evaluation answers become 35% of decision answers, and it holds through the later stages. The two hardest falls from comparison to choice: Logica, 25% to 8%, and Fujitsu, 14% to 0.4%. Present in the comparison, nearly absent from the verdict.

And the table adds a pattern beyond compared-versus-chosen: chosen and kept are different columns. A high decision share does not carry into retention by itself, and the brands holding the later stages are not always the ones winning the earlier ones. Every stage is its own contest, judged on its own evidence.

For challengers, that closes one door. There is no advocacy crack to compound against here. The opening in this market is a different one, and it is standing wide open below.

One more pattern, stated in aggregate: a share of the market's largest consolidated players appears in the answers only through the brands they acquired, not under their own name. An acquisition transfers the customers and the revenue. It does not transfer the recommendation. The recommendation capital sits on the brand name the models learned, and it stays there until the evidence moves.

The shortlist is frozen

Across three measurement days spanning nine days: 0% brand churn, 100% criteria stability. Identical shortlists, every time.

That matches what network-traffic research found this summer: Suganthan Mohanadasan, reading ChatGPT's self-written search queries, observed that settled categories keep their shortlists between runs while contested ones wobble (one account, directional).¹ This category is settled in the models' heads. The answer is not churning, so waiting for randomness to open a door is not a strategy. The door opens when the evidence changes.

Nobody owns a criterion

Ten criteria, zero owners. No brand is the default answer for transparent pricing, for response time, for industry expertise, for documented outcomes. The default wins on coverage and reputation, not on owning the questions.

And the buyers feel it, in near-identical words to the ones we heard in the US CRM market:

"Alle lover det samme, men hvad er reelt forskellen?"

(Everyone promises the same, but what is actually the difference?)

a buyer hesitation, surfaced repeatedly in the monitoring data

Two categories, an ocean apart in product and price, one identical vacuum. When nobody owns a criterion, buyers cannot tell providers apart, and the judge falls back on the default.

The eliminator nobody answers

The most severe elimination trigger in this category is not price. All three models cut providers on conflict of interest, and the buyer phrase behind it is blunt:

"Hvis de arbejder tæt sammen med en konkurrent, kan vi ikke vælge dem."

(If they work closely with a competitor, we cannot choose them.)

Look at the category's structure and the trigger becomes obvious. An IT partner serves dozens or hundreds of companies. In a market Denmark's size, the provider you are evaluating plausibly already serves someone who competes with you. The buyer knows it, asks the model about it, and gets eliminated providers back.

Here is the finding: no provider addresses this anywhere. Client lists are confidential, so the entire category has settled on silence, and the silence reads as risk. The most severe cut in the market fires against a question with no answer page in it.

That makes this the cheapest criterion takeover we have measured so far. The first provider that publishes how it handles competing clients, named policy, real mechanisms, wins the model's answer to a question every rival is contractually shy of touching.

What claiming it looks like

This is the working brief for that page, generated from the monitoring data, with the brand genericised:

Where the evidence must live: who listens where

The models name their sources, and this week's list is unusually concrete. Across the three models, the answers cited Datatilsynet, Erhvervsstyrelsen, CVR-registret, Dansk Industri, IT-Branchen, Trustpilot, Version2, and LinkedIn. Regulators, registries, the industry association, a review platform, the trade press, and a professional network. Not one provider blog.

That list is not random, and the research explains the mechanism behind each move you should make:

1. Your own page carries the facts, and it must parse

Network-traffic captures show the models going to the official page first for factual claims and giving up when content hides behind JavaScript. In one recorded reasoning trace, ChatGPT wanted a vendor's own numbers, could not parse the page, and cited a third-party review site instead.²

The conflict-of-interest policy must therefore exist as a dated, plain-HTML page: the policy text itself, not a PDF, not a claim rendered by script.

The same captures show ChatGPT appending "official" to its fact-seeking searches.³ For a policy question, your policy page is the official source it is hunting for. Be findable as exactly that.

2. The verdict about you needs third-party surfaces

Models do not cite you for the judgment. The research is blunt: vendor pages get cited for their own facts, while the verdict gets cited to third parties.²

For this category the third-party surfaces are already named in our data. Your Trustpilot profile and every reply you write there is indexed text a model can read. A reply to a review that references your conflict policy, in the policy's own words, is an echo on a surface all three models cited this week.

The trade press (Version2, Computerworld) and the industry association (IT-Branchen) carry the professional layer.

3. Match the surface to the engine

The per-engine differences are measured.

ChatGPT leans on text surfaces: Reddit threads, LinkedIn articles under named authors (ChatGPT cited LinkedIn in this week's run), review platforms. Video is close to citation-dead there, because search fetches a video's metadata, not its transcript.^2,5

Claude leans on professional and editorial surfaces: LinkedIn articles under named authors, the trade press, the industry association. The same text-bound rule applies, so the written policy and its echoes carry the weight, not video.

Perplexity is the inverse: it quotes video heavily, rewards fresh dated listicles, and cites vendors' own comparison pages, so a "[my brand] versus [rival]" page you write honestly yourself is a Perplexity asset.⁴

Gemini reads Google's surfaces, YouTube transcripts included.

One policy, one set of words, placed per engine: the page itself for the fact, LinkedIn and review replies for ChatGPT and Claude, a spoken version for Gemini and Perplexity.

4. Registries are evidence you already own

The models cited CVR and Erhvervsstyrelsen this week. Your registrations, your certifications, your filed accounts are machine-readable trust signals that cost nothing to keep clean and current. Check what the registries say about you before the models do.

5. Date everything

In an analysis of 250 million AI responses across eight answer engines, half of all top-cited content was under 13 weeks old.⁶ The recency bias is structural, and the researchers reading the engines' network traffic reached the same standing rule from the other direction: date every claim.³

For the policy page that means four signal layers working together: a dateModified that moves when the substance moves, temporal markers on the time-sensitive claims, a visible last-verified line, and a sitemap that agrees with the schema. The full implementation is documented in how to signal temporal authority to AI models.

A policy page with a visible review date beats an undated one twice: once with the buyer, once with the judge.

The lesson

Sources

The network-traffic findings below come from one researcher's logged-in accounts and are labelled directional by the author; the mechanisms are reproducible, the percentages are not population measurements. Capture dates matter: the plumbing changes faster than the mechanisms.

1. Suganthan Mohanadasan, ChatGPT Already Knows Who It'll Recommend Before It Searches, suganthan.com, August 2026.
2. Suganthan Mohanadasan, How ChatGPT Actually Picks Sources (I Read the Network Traffic, Not the Outputs), suganthan.com, June 2026.
3. Suganthan Mohanadasan, ChatGPT Changed How It Picks Sources While You Were Reading My Last Post, suganthan.com, July 2026.
4. Suganthan Mohanadasan, How Perplexity Actually Picks Sources (I Read the Stream, Not the Answers), suganthan.com, July 2026.
5. Ahrefs, Why ChatGPT Cites Pages, analysis of 1.4 million ChatGPT prompts, 2026.
6. Josh Blyskal, Profound, We Analyzed 250 Million AI Search Results, analysis across eight answer engines, 2025.

The Sunday Shortlist

All editions