Danish spend management: AI's answer is taken. The reasons are not.

Somewhere in Denmark right now, a bookkeeper in a company with 30 employees is asking an AI model how to get control of receipts, cards, and out-of-pocket expenses. The model answers in seconds: a shortlist, the risks, a favourite. No vendor hears about that conversation. We measure it.

What we measured

This decode is a baseline snapshot: two measurement days, 26 and 30 August, across the decision journey, measured with three models: Gemini, Claude, and ChatGPT. A snapshot photographs the market's standing answers; it cannot yet show drift. The late-journey stages (retention, advocacy) are sampled weekly rather than daily, because loyalty moves slower than choice, and a single weekly sample is below our reporting floor, so they appear as dashes until the second measurement week. What a snapshot does show, sharply, is who the models already believe.¹

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

• "Hvordan kan vi få bedre kontrol over vores udgifter?" (How do we get better control of our expenses?)
• "Hvad koster det at implementere en udgiftsstyringsløsning?" (What does it cost to implement an expense management solution?)
• "Understøtter det danske momsregler og SKAT-rapportering?" (Does it support Danish VAT rules and SKAT reporting?)

The judging sheet

The verdict: the default is the local name

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 snapshot window, 26 to 30 August. A dash means the sample for that stage is too small to report in a five-day baseline.

Pleo leads both stages and converts comparison into choice: 64% of evaluation answers become 71% of decision answers. In three editions of this series, that conversion is what we have watched brands fail at. The top three take 63.7% of all decision mentions between them.

Spendesk stands at 57% of evaluation answers, tied with Dinero and well clear of the rest of the field. At decision it carries 22%. A 35-point fall, the largest in the category: present in the comparison, mostly gone from the verdict. Lunar shows the same shape from a lower base, 27% to 10%.

But the widest gap in the table is not between any two brands. It is between what the default has and what it owns. That gap is the rest of this decode.

The category buyers do not believe exists

Look at the table again. Dinero, E-conomic, and Billy are accounting software. In the models' decision answers for spend management, bookkeeping systems hold two of the top four positions and nearly half the shortlist.

The hesitation data explains why, in the buyer's own words:

"Er det bare en ny måde at sige bogføring på?"

(Is this just a new way of saying bookkeeping?)

And its twin: "Hvad betyder 'udgiftshåndtering' egentlig for min lille virksomhed?" (What does 'expense management' actually mean for my small company?)

The models are not confused. They are faithful. They draw the category the way buyers think, not the way the industry maps it, and Danish SME buyers think in bookkeeping. The mechanism is visible right in the buyer language: one of the most common questions in the data is "Kræver det integration med vores regnskabsprogram (f.eks. e-conomic, Dinero)?" (Does it require integration with our accounting system, e.g. e-conomic, Dinero?). The buyer names the adjacent tools inside the question, and the answer promotes them to candidates. The neighbour you integrate with becomes the rival you are compared to.

For every category-creator, this is the finding to sit with: your real AI competitors are not your category. They are the buyer's mental model of your category. Either you teach the boundary, with content that answers "what is this, and why is it not bookkeeping", or you compete inside the neighbour's shortlist on the neighbour's terms.

The boundary-teaching page, in three lines: the title in the buyer's question shape, "Er udgiftshåndtering bare bogføring? Forskellen forklaret" (Is expense management just bookkeeping? The difference explained); the position, that expense management is what happens before bookkeeping, control at the moment of spending rather than registration after it; the slug, /er-udgiftshaandtering-bare-bogfoering. One page, and the category's most common hesitation has an answer with your name on it.

Nobody owns a criterion

Ten criteria, zero owners. That is now four markets in a row: US CRM platforms, Danish IT outsourcing, UK sustainable architecture, Danish spend management. Not one of the forty measured criteria across four categories has a brand attached as its default answer.

In the first three editions, the vacuum was the challenger's opening. This week it reads differently, because this week the category has a strong default, and the vacuum sits under the default's own feet. Pleo's lead rests on coverage, familiarity, and name gravity, not on owning "no hidden fees" or "fastest bookkeeping close" or "cleanest SKAT compliance" as evidenced, citable territory. A lead like that is real, and it is rented. The first challenger that reads the last three editions of this series and publishes hard evidence against a single criterion starts pulling answers in a market where the incumbent has left every door unlocked.

The eliminator: the independence question

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

"Leverandøren sælger også finansielle produkter, der kan påvirke rådgivning."

(The provider also sells financial products that can influence advice.)

Readers of edition 2 will recognise the trigger: conflict of interest was also the hardest cut in Danish IT outsourcing, where the fear was competitor exposure. Different market, different fear, same shape, and that is the emerging cross-category pattern: in B2B categories where the vendor has adjacent revenue, independence is becoming the sharpest knife in the drawer. Here the category runs on cards, payments, and financial services, so the buyer asks: when this platform advises me on my spending, whose interest is speaking?

The supporting criterion is measured too: "Er der nogen skjulte incitamenter for leverandøren, der kan påvirke deres anbefalinger?" (Are there any hidden incentives for the provider that could influence their recommendations?)

The takeover play is the same one edition 2 mapped: the first vendor that answers the independence question in writing, plainly, on a page (how the platform makes its money, what it earns from cards and payments, what it will and will not recommend and why) claims a criterion the whole category's buyers are already asking about. For a default, that page is defence. For a challenger, it is a crowbar.

What claiming it looks like

This is the working brief for that page, generated from the monitoring data, with the brand genericised. The frame is "acknowledge the problem": a fear-shaped eliminator is defused by validating the concern first, without reservation, and only then showing the structure that removes it. Evidence presented before acknowledgement reads as protesting too much.

Where the evidence must live: who listens where

The models name their sources, and this week's list reads like a Danish finance department's bookmarks. Across the three models, the answers cited SKAT, Erhvervsstyrelsen, Virk, Dansk Industri, IT-Branchen, Børsen, and the accountant layer: revisor directories and the auditor associations. The tax authority, the registries, the trade organisations, the business press, and the profession that actually keeps SME books. Not one vendor 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 source instead.²

For this category the fact under the most pressure is pricing: "skjulte gebyrer" (hidden fees) appears in the eliminator data and in the cost criterion alike. A total-cost page, plain HTML, every fee named, is both the answer to the most-asked cost question and the evidence a model can lift. The same captures show ChatGPT appending "official" to fact-seeking searches; for a pricing question, your pricing page is the official source it hunts for.³

2. The accountant is this category's verdict layer

Models cite vendors for their own facts and third parties for the judgment.² This week the third party has a profession: the models cited revisor surfaces, and the buyer language asks for accounting-system integration by name. In Danish SME buying, the bookkeeper is the trusted recommender, and the surfaces where accountants read and write, the directories, the association guidance, the integration marketplaces of e-conomic and Dinero, are where a vendor's claims become someone else's words.

3. Match the surface to the engine

The per-engine differences are measured.

ChatGPT and Claude lean on text surfaces: trade press, LinkedIn articles under named authors, review platforms and directories. Video is close to citation-dead in ChatGPT, because search fetches a video's metadata, not its transcript.^2,5

Gemini and Perplexity read the spoken word, transcribed: YouTube walkthroughs of real workflows, fresh and dated, are assets there.⁴

One set of claims, placed per engine: the page itself for the facts, the professional and editorial layer for ChatGPT and Claude, a spoken version for Gemini and Perplexity.

4. The registries are evidence you already own

The models cited Erhvervsstyrelsen and Virk this week. In a regulated category, licences are public text: registration status, filed accounts, supervisory approvals 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 Pricing and compliance claims age fastest of all; a fee table with a visible verification date beats a timeless 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