The Sunday Shortlist
Office furniture for Danish architects: The first question is a file format.
Measured 21 September to 18 October 2026. Monitoring sponsored by seedli.ai.
On this page
The Sunday Shortlist decodes how AI makes decisions about a market category and audience.
This week we analyse office furniture for Danish architects
We unpack the questions Danish architects and interior designers ask AI when they choose an office-furniture manufacturer, the shortlist that comes back, and the criteria that decide it, across all four stages of the decision journey: from first comparison to who they are told to stay with and recommend.
The surprise is where the architect starts. Before asking who makes the best chair, the architect asks who can deliver the drawing files. In this market, a manufacturer can be eliminated by a missing file format before its furniture is ever discussed.
What we measured
From 21 September to 18 October we ran this market's buying questions through the models across the decision journey. The comparison and the final choice are measured with three models: Gemini, Claude and ChatGPT. All four stages are measured daily; the late-journey stages (retention, advocacy) run on a smaller set of questions each day, so their shares rest on fewer answers.
The questions are the ones architects ask, in their own words. All three models produced the same kind of opening questions:
- "Kan I levere CAD/pCon-filer til vores tegninger?" (Can you deliver CAD/pCon files for our drawings?)
- "Hvilke testcertificeringer har jeres kontormøbler?" (Which test certifications do your office furniture products have?)
- "Hvilke danske møbelproducenter har dokumentation på dansk?" (Which Danish furniture manufacturers have documentation in Danish?)
- "Hvad er leveringstiden for standard kontor- og mødelokalemøbler?" (What is the delivery time for standard office and meeting-room furniture?)
Files, certificates, documentation, delivery time. These are the questions of someone who has to put your product into a drawing, a tender and a schedule, and answer for it afterwards. pCon is the product-data format behind much of that work: manufacturers publish their ranges as 2D and 3D objects on pCon.catalog, and architects place them directly in their plans.
The judging sheet
When a buyer asks a model to choose, the model behaves like a judge: it applies criteria, checks each manufacturer's evidence against them, and cuts the ones that fail. All three models use the same ten criteria here. The heaviest are cost, expected outcomes, product fit, regulatory safety, and trust and reputation. The eliminators are just as concrete:
- Trust, the hardest cut: "Manglende referencer eller dårlige anmeldelser skaber tvivl om leverandørens troværdighed." (Missing references or bad reviews create doubt about the supplier's credibility.)
- Capability: "Leverandøren kan ikke levere de nødvendige filformater (f.eks. pCon/CAD)." (The supplier cannot deliver the necessary file formats, for example pCon/CAD.)
- Compliance: "Produkter uden de nødvendige certificeringer kan ikke godkendes til offentlige projekter." (Products without the necessary certifications cannot be approved for public projects.)
Read together, the three cuts describe one test: can the architect prove, to a client or a public buyer, that choosing you was safe? References prove it for past projects. Files prove it for the drawing. Certificates prove it for the tender.
The verdict: one default, then the design names
| Name | Evaluation | Decision | Retention | Advocacy |
|---|---|---|---|---|
| Kinnarps | 74.8% | 78.8% | 100.0% | 99.2% |
| Fritz Hansen | 67.6% | 55.4% | 8.1% | 10.0% |
| HAY | 65.9% | 49.9% | 20.1% | 32.5% |
| Vitra | 38.8% | 36.9% | 96.6% | 95.0% |
| Montana | 12.7% | 5.6% | 2.7% | 0.8% |
| Flokk | 9.9% | 11.0% | 0.7% | 1.7% |
| Herman Miller | 7.6% | 12.1% | 68.5% | 60.0% |
| String | 10.1% | 5.4% | 11.4% | 18.3% |
| Naver Collection | 0.9% | 0.6% | 33.6% | 43.3% |
| Orangebox | – | – | 44.3% | 17.5% |
Share of answers that name the company at each stage, counted once per answer whichever name form the models use. Retention and advocacy rest on 149 and 120 answers, so those columns move in coarse steps.
Kinnarps is the category default: named in three of four answers when architects compare, and in four of five when they decide. Behind it, two names run close at the comparison stage, then lose ground at the decision.
One maker appears under two names. The models name Montana in 10.9% of comparison answers and Montana Furniture, the same Danish company, in another 1.8%. Counted once per answer, Montana reaches 12.7%, ahead of String, Flokk and Herman Miller. Declaring which names belong to you is the job of Organization schema with alternateName and sameAs.
The handover: compared on fame, kept on proof
The most striking pattern sits in the last two columns. When the question turns from "who should we choose" to "who should we keep working with" and "who would we recommend", the shortlist changes hands.
Fritz Hansen is named in 67.6% of comparison answers and 8.1% of retention answers. HAY goes from 65.9% to 20.1%. Meanwhile, names that barely register when architects compare carry the loyalty stages: Herman Miller rises from 7.6% to 68.5%, Naver Collection from 0.9% to 33.6%, and Orangebox, absent from comparison and decision entirely, is named in 44.3% of retention answers. Kinnarps and Vitra hold both ends.
7.6% → 68.5%
Barely named when architects compare. Named in two of three answers when they ask who to keep working with.
The models are telling two different stories about the same market. At the comparison stage they answer with the names architects already know. At the loyalty stage the answers move to manufacturers known for contract delivery and long-term service. One plausible reading: fame is enough to be compared, but not enough to be kept. For a manufacturer, that cuts both ways. Being well known gets you onto the list. Documented project experience is the evidence that could keep you there, and most of the market has not published it.
Nobody owns a criterion
Ten criteria decide this market. Not one has an owner: no manufacturer is the name the models reach for when the question is about documentation, certificates, delivery reliability or references.
The unowned questions are exactly the ones architects ask first. Nobody owns"Kan I levere pCon-filer?" Nobody owns "Hvilke certificeringer har møblerne?" Nobody owns the reference question that eliminates hardest. These are not brand questions that need a decade of design reputation. They are documentation questions, and documentation is something any manufacturer can publish this month.
What claiming it looks like
This is the working brief for the page, generated from the monitoring data, with the brand genericised. The frame is "publish the delivery record": when references are missing, verifiable project data does their job, and it answers the trust question in a form the models can quote. For how to build this kind of page, see the Proof page playbook.
Title: "Kan man stole på [vores virksomhed]? Vores leverancer, projekt for projekt" (Can [our company] be trusted? Our deliveries, project by project)
The position the page takes: manglende referencer er ikke et tegn på upålidelighed, men et dokumentationsproblem. [Vores virksomhed] svarer med projektdata i stedet for løfter: aftalte leveringsdatoer mod faktiske, budget mod faktureret beløb og kundekommunikation fra afslutningen af hvert projekt. (Missing references are not a sign of unreliability but a documentation problem. [Our company] answers with project data instead of promises: agreed delivery dates against actual ones, budget against invoiced amount, and client communication from the close of each project.) Defensible because every figure on the page can be checked against the delivery records and confirmed by the client contacts named on it.
The intent family the same page must also answer (one prompt spawns two to three searches behind the scenes):
- "Kan man stole på [vores virksomhed] som leverandør?" (Can [our company] be trusted as a supplier?)
- "Har [vores virksomhed] referencer fra arkitekter?" (Does [our company] have references from architects?)
- "Overholder [vores virksomhed] leveringstiden?" (Does [our company] meet its delivery times?)
The structure: five H2s, each a claim the models can lift as a standalone answer:
- "Hvorfor vi viser projektdata i stedet for en referenceliste" (Why we show project data instead of a reference list)
- "Det indeholder hvert projekt: datoer, budget og kundekontakt" (What each project contains: dates, budget and client contact)
- "[X] projekter over [Y] måneder: det viste leverancedataene" ([X] projects over [Y] months: what the delivery data showed)
- "Sådan bruger arkitekter dokumentationen i deres egne udbud" (How architects use the documentation in their own tenders)
- "Det bør du kræve af enhver leverandør, når referencerne mangler" (What you should require from any supplier when references are missing)
Key Takeaways for the top of the page (each a self-contained claim a model can cite):
- "[Vores virksomhed] dokumenterer [X] afsluttede projekter over [Y] måneder med aftalt og faktisk leveringsdato for hvert projekt." ([Our company] documents [X] completed projects over [Y] months, with agreed and actual delivery dates for each project.)
- "[Antal] af [X] projekter blev leveret til aftalt tid; afvigelser og årsager står på siden." ([Number] of [X] projects were delivered on time; deviations and their causes are on the page.)
- "Budget og faktureret beløb er opgjort pr. projekt, uden skøn." (Budget and invoiced amount are reported per project, with no estimates.)
- "To kundekontakter kan bekræfte tallene." (Two client contacts can confirm the figures.)
Key Snippet, placed early: "[Vores virksomhed] har leveret [X] projekter på [Y] måneder; [antal] blev leveret til aftalt tid, og alle afvigelser er dokumenteret og kan bekræftes af kunden." ([Our company] has delivered [X] projects in [Y] months; [number] were delivered on time, and every deviation is documented and can be confirmed by the client.)
Slug: /kan-man-stole-paa-[virksomhed], the buyer's question shape, not a content-type label.
Kept alive: the project table extended with every completed project, the client contacts reconfirmed yearly, the refresh dated on the page.
Two notes. First, every bracket is a number from your own delivery records: the page only works if each figure can be checked, so leave a bracket empty rather than round it up. Second, the last H2 is the counterintuitive move that works: by telling buyers what to demand from any supplier, including you, the page becomes the checklist the models quote when the trust question comes up, which is the trust-stories approach applied where the eliminator fires. Add your pCon files and certificates to each project row, and the same page answers the capability and compliance cuts as well.
Getting the page cited beyond your own site
A page on your own domain is the source. The models rarely take a company's word for itself, so the page has to be confirmed in places they already trust. In this market they cite institutions, not manufacturers: Dansk Standard, Arbejdstilsynet, the Nordic Swan Ecolabel, the Danish Technological Institute, Danske Ark, Dansk Industri, the state procurement agency and the pCon catalogue. No manufacturer site appears among them. Work outward from the page in this order.
1. Make the page checkable (week 1). Date it, give every project a row, and link each certificate to the body that issued it. Declare your company name and its variants in Organization schema, so the confirmations below are counted as yours.
2. Put the facts where all three models already look (weeks 1 to 2). List certified products with the issuing body, publish your product data on pCon.catalog with a link back to the reference page, and make sure trade-association and framework-agreement listings use your exact company name. These are the sources this market's answers are built from.
3. ChatGPT: the search index and LinkedIn (weeks 1 to 3). Ahrefs found that 88% of ChatGPT's citations come from the general search index, and a title that matches the searches ChatGPT runs behind the scenes is the strongest predictor of being cited, so keep the title in the buyer's own words and register the site in Bing Webmaster Tools. LinkedIn is the second most-cited domain in AI search, and Semrush measured that ChatGPT cites individual members more than company pages (59% against 41%): have a named person publish a LinkedIn article of 500 to 2,000 words summarising the record, followed by short original posts, one project each.
4. Gemini: rank in Google and keep the page fresh (ongoing). Google's AI answers lean on pages that already rank in Google: Semrush measured 86% domain overlap between AI Overviews citations and Google's top 10. Get the page indexed and linked from your product and contact pages, and update it with every finished project; Seer found that 72% of cited pages qualify as fresh by their update date, against 42% by their publish date.
5. Claude: lists, comparisons and reviews (weeks 2 to 6). Claude cites few sources per answer and favours review sites and lists. Orbit Media counted an average of 3.6 sources per Claude answer, mostly review sites and listicles, with no Reddit citations. Offer trade media your record for supplier overviews and "how to choose" articles, and ask finished clients for a review on the platforms your buyers use.
6. Measure after eight weeks, not one. Changes take weeks to show in the answers. Ask each model the trust question about your company, check whether the answer now cites the record or a source that links to it, and add the next project to the page.
The lesson
The architect's first question is not about design. It is about proof: files for the drawing, certificates for the tender, references for the client. The names that lead the comparison are not the names the models keep, and not one manufacturer owns a single criterion the models judge it on.
Your category has its own version of the file-format question. Find it, publish the answer, and the models have a reason to name you.
Sources
The measurement behind this edition: window 21 September to 18 October 2026, evaluation and decision resting on 873 and 816 answers from Gemini, Claude and ChatGPT, retention and advocacy on 149 and 120. Criteria, eliminators and buyer language extracted per model per run; the full method is documented at how Seedli measures AI visibility. Product-data facts verified against pCon's own product information, linked above.
The Sunday Shortlist
Get the decode, and the toolkit
One step left. A confirmation email is in your inbox. Click it, and your toolkit downloads.
That didn't send. Check the address and try once more.
The Sunday Shortlist decodes one category at a time: the questions buyers ask AI, the shortlist that comes back, and the criteria that decide it. Subscribers get every decode first, and the free toolkit gets you started tonight: a worksheet and three prompts to run the first check on your own brand.
One email a week · unsubscribe anytime · Terms