1,125 responses, 5 models, 45 Dutch questions: the baseline AI-visibility snapshot of the Netherlands fashion and clothing market

Vertical: Fashion & clothing (mode & kleding), Netherlands
Method: A single point-in-time study of 1,125 responses across five large language models (Claude, Gemini, GPT, Perplexity, Grok), each asked 45 Dutch-language questions five times
Collection window: 11 September 2026

Key terms: AI visibility, AI Visibility Score, clothing brands, fashion brands, mode, kleding, duurzame mode, grote maten, lengtematen, Dutch clothing brands, LLM brand recommendations, ChatGPT/Gemini/Claude/Perplexity/Grok brand recommendation, Generative Engine Optimization (GEO).

This report reproduces the real Dutch-language questions people ask AI assistants about clothing, and measures how often, in what order, and across how many models five large language models name each entity. Because the same questions can be read with or without an origin restriction, and because a question about fabric quality is not the same as a question about which brands exist, the entities are ranked in six separate cuts, defined in Chapter 2. It is a single point-in-time study and should be read within the limitations in Chapter 11.

How to read this report: visibility is not quality

Every number here measures whether an AI assistant names an entity, and how early. It does not measure whether the product is well made, whether it lasts, whether the price is fair, or whether the recommendation was accurate. A brand can be highly visible and mediocre, or excellent and invisible. Read the leaderboards as a map of findability in a new discovery channel — nothing more.

1. Executive Summary

One-sentence takeaway: The most-named clothing entity in the Dutch study is a small Dutch denim brand, but the top of the leaderboard goes to Uniqlo, because being named early counts for as much as being named often.

• The leader and the most-mentioned entity are not the same. Uniqlo tops the general leaderboard at 96.13, appearing in 202 of 1,125 responses (17.96%). Kuyichi is named in 219 (19.47%) — more than anything else in the study — and finishes second at 94.62, because Uniqlo is named earlier when it appears (MRR 0.5531 against 0.4596). Both are named by all five models. The gap between them is 1.51 points; the gap from second to third is 15.19.
• 30 entities qualified for the general ranking. Of 720 dictionary entities, 659 matched at least one response and 30 cleared the 5% mention threshold across all 45 questions: 12 Dutch and 18 foreign. Twenty-four of the thirty are named by all five models.
• Origin depends entirely on which questions you count. On the 33 origin-neutral questions (825 responses), 8 of 28 qualifiers are Dutch and Dutch brands take 3 of the top ten. On the 12 questions that explicitly ask for a Dutch brand (300 responses), 31 of 34 qualifiers are Dutch and the top ten is entirely Dutch — by construction of the question set, not as a finding. The two figures are printed together everywhere in this report and neither is quoted alone.
• Position can outrank reach, and here it does so twice. YOEK ranks seventh in the general cut on a mention rate of 6.31% (71 of 1,125), because it holds the general-cut position maximum at an MRR of 0.5601. Studio Anneloes ranks fifth on 8.71%. The composite is working as designed, but the underlying numbers should be read alongside the score.
• The models differ enormously in how much they say. Grok names 14.11 entities per response against Claude Haiku’s 3.65 — a near fourfold spread from the same questions. They also disagree about the leader: Claude and Perplexity name Kuyichi most, Gemini and GPT name Uniqlo most, and Grok names MUD Jeans most.
• Grounding ranges from 29.3% to 100% of each model’s 225 answers. Claude, Perplexity and Grok searched on every answer, GPT-5.6 on 97.3% of its 225, and Gemini on 29.3% of its 225 — against an overall rate of 85.3% across all 1,125 responses. The two models that ground least are not the two that name fewest Dutch brands, so grounding does not explain the origin spread.
• The information layer is not brand-owned. Reddit is the single most-cited domain in the study, cited in 159 of 1,125 responses, ahead of every brand and retailer site. Retailer domains supply 30.89% of citations and brand-owned sites 17.13%. The most-cited brand site, kuyichi.com, appears in 100 responses.
• Some Dutch-brand questions were not answered with clothing at all. Asked for Nederlandse merken, the assistants returned Unox, Hak, Zonnatura, Koopmans and Mora — Dutch food brands — alongside clothing labels. These are classified Out-of-scope and ranked nowhere, but they are a finding about how the models read the phrase, discussed in Chapter 8.

Why it matters. Discovery is moving from search engines and marketplaces toward AI assistants, and clothing is an unusually hard category for that transition: fit, fabric and sizing are the things shoppers actually ask about, and none of them is easy to answer from a product feed. The questions in this study reflect that — inside leg 36, size 46/48 with sleeves, jeans that do not gap at the waistband, a raincoat that does not look like an outdoor jacket. This edition asks a narrow question about it: when someone types a Dutch question into an assistant, whose name comes back, and how early. It is the first (baseline) snapshot, so there is no prior edition to compare against and none is implied anywhere in this report. One reminder: the numbers describe visibility and information availability, not which clothes are best.

2. Methodology

One-sentence takeaway: Five large language models were each asked 45 Dutch questions five times with no system prompt, producing 1,125 responses of which all 1,125 were analyzed; entities were identified by alias matching plus human-gated research and ranked with a three-component (45/30/25) score computed separately within each of six cuts.

2.1 Scope and models

The study queried five large language models directly via API with identical questions. All models received only the user question. No system prompt was used, temperature was not set manually (provider default), and no region or locale parameter was defined. Because the questions are in Dutch, the models infer the Netherlands market from language alone.

45 questions × 5 models × 5 repeats produced 1,125 responses collected. Every question × model × repeat cell exists exactly once and no response returned empty text, so 1,125 responses were analyzed and every per-model denominator is exactly 225. Denominators are computed per (question, model) cell rather than as questions × models × repeats, which is the same thing here only because the grid is complete.

2.2 Web-search configuration and grounding

No model was instructed to search. Grounding is whatever each provider does by default, and the spread is wide: 85.3% of all 1,125 responses were web-grounded, but that average conceals Claude Haiku, Perplexity and Grok at 100%, GPT-5.6 at 97.3%, and Gemini at 29.3%. Because grounding differs so sharply by model, and so does answer breadth, breadth carries only 25% of the composite score. Chapter 5 treats the spread as a finding in its own right.

2.3 Questions and cuts

Questions are written the way Dutch consumers actually ask them, including colloquial phrasing, English loanwords, abbreviations and imperfect grammar. No question was rewritten and none was removed because its answers were inconvenient: new_text equals original_text for all 45 questions, and every quotation in this report reproduces the question as asked, including its lower-case openings and missing punctuation. All 45 appear in Dutch with English glosses in Chapter 12.

Six cuts are published. There is no trend cut in this edition, because there is no prior edition to trend against, and no segment cuts, for the reason given in Chapter 7.

Every question belongs to exactly one question type and exactly one origin scope, so each of the 45 questions contributes to three of the six cuts: general, one of local/open, and one type. Question 45 — “Ik wil een tijdloos pak voor bruiloften en af en toe werk, liefst van een Nederlands merk. Waar zou je kijken?” — is a local, use-case question, and its 25 responses sit in the general, local and use-case denominators and nowhere else.

The denominators reconcile in two independent ways: local (300) plus open (825) equals general (1,125); and the three question-type cuts (375 + 375 + 375) sum to 1,125. Every entity’s response counts decompose the same way.

2.4 Entity resolution

Candidates were extracted from the analyzed responses after stripping markdown link targets, bare URLs, domain-style anchors and citation markers, so that source domains could not inflate brand mentions. This matters more here than it might sound: 100% of Claude’s and Grok’s answers and 97% of GPT’s contain inline links, and the anchor text is frequently a brand name pointing at an unrelated retailer’s page.

2,957 candidates reached a floor of ≥3 responses. 1,117 were sent to human-gated research and the remainder dropped mechanically, each with a stated reason. 266 of the researched rows came back as Non-brand. The final dictionary holds 720 entities carrying 785 aliases — 642 Core, 42 Adjacent, 36 Out-of-scope.

Three extraction failures are worth naming, because each was found by a check rather than by reading:

1. Brands that only ever appear with a product name attached. “Jack Wolfskin Tempest” existed as a candidate at 15 responses; “Jack Wolfskin” did not exist at all, because the extractor captured the longest capitalised run and never the bare brand. Happy Rainy Days and Rogelli failed the same way. Prefix forms were added to candidate generation and 36 entities recovered. The failure was surfaced by the cited-domain backstop, not by inspection.
2. Aliases colliding with ordinary garment prose. Gap matched “waist gap” across 29 responses, Easy matched “easy care”, and Next matched “next-day” and the unrelated entity The Next Closet. All three were removed after in-context samples were inspected. Costco was separately blocked from attaching to COS by a token-boundary rule.
3. One company under two dictionary rows. Checking whether any two entities shared a verified website found eleven splits. G-Star RAW (38 responses) and G-Star (36) are one brand; their union is 68 and it qualifies in the general cut, where neither half did. De Bijenkorf and Bijenkorf, Zazou and Zazou Basic, Olaf Hussein and OLAF, and seven more were merged the same way. Merges are unions of response sets, never sums, and MRR is recomputed from the earliest first mention among the merged aliases.

Matching is alias-based, Unicode-normalized, case-insensitive, URL-stripped and anchored on word boundaries. Unrestricted substring matching is not used. Dutch compounds are handled only through a controlled brand-plus-product-term rule. All aliases of six characters or fewer were inspected with in-context samples.

Two dictionary entities matched zero responses, which is why 659 entities appear in the metrics and 720 in the dictionary. Both are sub-brand parents — Fear of God and River Island — that the models never name alone: the answers say “Fear of God Essentials” and “River Island Plus” and nothing else. They are reported rather than removed.

Sub-brands are ranked as independent entities, as in every edition of this series. A second full resolution in which sub-brands fold into their parents is published in the data pack as the _families tables, on its own scale. Folding changes very little — the typical parent gains three or four responses, because a sub-brand is almost always named in the same answer as its parent — and the largest movement is Levi’s, from 105 responses to 109. The two resolutions are on different scales and must not be compared.

2.5 The AI Visibility Score (0–100)

Per entity, per cut:

mention_rate = responses / denom
mrr          = sum_rr / responses      (reciprocal of the first-mention rank)
breadth      = distinct models
qualified    = mention_rate >= 0.05 and not Out-of-scope

c_mention  = 100 * mention_rate / max_mention[cut]
c_position = 100 * mrr / max_mrr[cut]
c_breadth  = 100 * breadth / 5
ai_score   = 0.45*c_mention + 0.3*c_position + 0.25*c_breadth

An entity is counted once per response, at its earliest mention.

Worked example, general cut. The scale is drawn from qualified entities only: max_mention is Kuyichi’s 0.194667 and max_mrr is YOEK’s 0.560095, across 1,125 responses. Uniqlo has a mention rate of 0.179556, an MRR of 0.553143 and breadth 5, so c_mention = 100 × 0.179556 / 0.194667 = 92.24, c_position = 100 × 0.553143 / 0.560095 = 98.75, c_breadth = 100, and ai_score = 0.45 × 92.24 + 0.30 × 98.75 + 0.25 × 100 = 96.13. That is the number in the table below, and it reproduces from scale_registry.json on every one of the 2,700-plus rows in the master metrics.

Components are not clipped at 100. 91 rows among unqualified entities exceed 100 on a component, because the maxima come from qualified entities only. They are reported as they are.

2.6 Origin, neutrality and self-exclusion

Origin is trademark ownership. Local means the trademark country is the Netherlands. Belgium and Germany are Foreign, and Benelux is not a rollup. A .nl domain, Dutch-language marketing, Dutch retail availability, a Dutch distributor or a Dutch subsidiary do not make an entity Dutch. Both trademark country and operator country were collected for every researched entity, so the rule can be changed in a later edition without repeating the research. The dictionary stores ISO country codes; the tables in this report use the same codes and roll them up to Dutch/Foreign/Unknown for shares only.

43 entities have unresolved trademark origin. They are retained in the dictionary and excluded from origin shares rather than counted as Foreign. None of them qualifies in any cut, so no published share is affected.

Self-exclusion. The publisher’s own domain was checked for in the cited-source data. herm.io appears in one of 1,125 responses, and that single citation is excluded from the source figures and from the published domain list. No other publisher-owned domain appears.

2.7 What a future edition will and will not be able to compare

This is edition 1. A second edition will be able to compare mention_rate, mrr and breadth for any entity present in both, because those are absolute measurements against a stated denominator. It will not be able to compare ai_score, because the scale is drawn from the leader of each cut and moves whenever the leader moves. If Kuyichi’s mention rate falls and it remains the leader, every other entity’s c_mention rises without anything having changed for them.

Two structural things will also change: this edition has no segment cuts, and edition 2 will carry a subcategory assigned at collection time; and the entity dictionary will begin from this one rather than from discovery, which shortens the review gate but risks missing new entrants. Both are noted so that a reader of edition 2 knows which differences are market movement and which are method.

3. Overall Visibility Leaderboard

One-sentence takeaway: Thirty entities qualify, the top two are separated by 1.51 points and disagree about which of them is more visible, and twenty-four of the thirty are named by all five models.

3.1 General leaderboard: 30 qualified entities

Reading. Uniqlo tops the table at 96.13 and Kuyichi follows at 94.62, and the two are measuring different things: Kuyichi is named in 17 more responses, Uniqlo is named earlier in the ones it appears in. Below them Zalando (79.43) is the highest-placed pure retailer, and the board then flattens — fourth to thirtieth spans 27 points across 27 entities.

The composition is worth noting. Ten of the thirty are Brand+Retailer entities that design and sell under their own name, four are pure retailers, one is a marketplace, and the remaining fifteen are brands that sell through others. In a market where the leading names operate their own stores, separating “brand” from “retailer” would misdescribe most of the board.

3.2 The same cut, ranked by mention rate

The reordering shows what the position component is doing. Kuyichi takes first place on volume. Arket climbs and YOEK falls out of the top ten entirely — YOEK is named in 71 responses but almost always first, while Arket is named in 122 and usually later. Both clear the bar, and they describe different things: one is a name that comes up as the answer, the other a name that comes up in the list.

3.3 Brand-only view (marketplaces removed)

One of the thirty qualified entities is a marketplace: Vinted at #10. It is kept in the headline ranking because that is how shoppers meet the second-hand shelf, but removing it gives a pure brand-and-retailer picture:

The brand-only files strip marketplaces only, by name, and there are fourteen on the general board’s underlying entity list. Own-label and vertically integrated retailers — Zalando, Wehkamp, De Bijenkorf, HEMA, Zeeman — are retained in every table. Removing them would delete most of how Dutch shoppers actually buy clothes.

3.4 Above the threshold but off the board

One entity clears 5% in a cut and is still not ranked: Coolblue, at 5.33% of the attribute cut (20 of 375 responses). It is classified Out-of-scope — an electronics retailer that the models named when asked about returns and refund speed, not about clothing. It is tracked, excluded from every leaderboard and every origin share, and reported here so the exclusion is visible rather than silent.

3.5 Below the threshold

Six Core entities sit within twelve responses of qualifying in the general cut: Massimo Dutti (55 of 1,125, 4.89%), MS Mode (51, 4.53%), Filippa K (49, 4.36%), WE Fashion (48, 4.27%), Girav (46, 4.09%) and Joe Merino (45, 4.00%). Three of the six are Dutch. Girav is the interesting one: at an MRR of 0.6775 it would hold the general-cut position maximum if it qualified, and it ranks second in the local cut, where the denominator is smaller and its 29 responses clear the bar.

4. Origin: the constrained and origin-neutral shelves

One-sentence takeaway: Dutch brands hold 3 of the top ten when the question does not ask for them and 10 of 10 when it does, and only the first of those two numbers is a finding.

4.1 The origin-neutral shelf (33 questions, 825 responses)

This is the cut that carries the origin finding, because nothing in these 33 questions asks for a Dutch answer. Uniqlo leads at 96.37 on 201 of 825 responses, followed by Zalando, COS and H&M — three foreign entities and one foreign-owned retailer before the first Dutch name. Studio Anneloes is the highest-placed Dutch entity at fifth, on 55 responses and an MRR of 0.6295, which is the position maximum for this cut.

Eight of 28 qualifiers are Dutch. Dutch brands take three of the top ten. For a market of this size against Swedish, Japanese, Spanish and American competition that is a real presence, but it is a minority one, and it is the number to quote when someone asks how Dutch clothing brands are doing in AI answers.

4.2 The constrained shelf (12 questions, 300 responses)

Here Kuyichi leads at 88.38 on 124 of 300 responses — a 41.33% mention rate, the highest anywhere in the study — and the top ten is entirely Dutch. That is not a finding. These twelve questions ask for Dutch brands, so Dutch brands answer them. The cut is published because it shows which Dutch names the assistants reach for when pushed, and the answer is not the same set that qualifies on the open shelf: Girav, Joe Merino and MS Mode all qualify here and nowhere else.

Three foreign entities qualify in this cut anyway. They are named in answers to questions that asked for Dutch brands, which is worth knowing on its own.

4.3 Reading the two together

The 12/33 split is itself a limitation. The constrained shelf rests on 300 responses against the open shelf’s 825, so its qualifier list is thinner evidence per name, and the 5% threshold means 15 responses there against 57 in the general cut. Every Dutch-visibility claim in this report is quoted with both figures and with the question count behind each, and any single-number version of “how visible are Dutch clothing brands in AI answers” would be choosing which of the two to report.

5. Differences Between Models

One-sentence takeaway: The five models broadly agree about who belongs on the shelf and disagree almost completely about how many names to put on it — Grok gives nearly four times as many as Claude.

5.1 Per-model behavior

The spread is the story. Grok names 14.11 entities per response and Claude Haiku 3.65. Asked the same 45 questions the same number of times, one model hands back a long list and another hands back a short recommendation. Neither is wrong, and the composite deliberately limits how much this can distort the ranking — breadth counts distinct models, not mentions, so a model that names forty brands cannot inflate any one of them beyond one response.

5.2 Grounding: one model barely searched

Gemini grounded 29.3% of its answers. Claude, Perplexity and Grok grounded 100% and GPT-5.6 97.3%. This is the widest grounding spread the series has recorded, and it means roughly seven in ten Gemini answers about Dutch clothing were produced from training data alone.

The tempting reading is that ungrounded answers should favour large international brands, since those dominate training data, while grounded answers should surface Dutch names from the live web. The data does not support it. Gemini’s Dutch share of mentions is 37.5%; Grok’s, at 100% grounding, is 36.7%. The two models with the highest Dutch share are Perplexity (49.2%) and Claude (48.2%), one of which grounds everything and neither of which is an outlier on breadth. Whatever drives the origin mix, it is not whether the model searched.

5.3 Each model’s most-named entities

Three different leaders across five models. Claude and Perplexity lead with Kuyichi, Gemini and GPT with Uniqlo, and Grok with MUD Jeans. The Dutch sustainable-denim cluster — Kuyichi, MUD Jeans, Kings of Indigo — is strong in every model; the international high-street names are strongest in the two that ground least and name most.

5.4 When a rank is really one model’s habit

Twenty-four of the thirty general-cut qualifiers are named by all five models, and the remaining six by four. No general-cut ranking in this edition rests on fewer than four models. That is not true of the smaller cuts:

Suitable’s attribute ranking is one model’s habit. It qualifies on 19 responses from a single assistant; the other four never name it in an attribute answer. Every one of these seven is published with its breadth column intact rather than quietly dropped, and each rests on at least 15 responses. Róhe appears twice, which makes it the clearest case in the edition of an entity visible to part of the panel and invisible to the rest.

6. Question-Type Ownership

One-sentence takeaway: Discovery questions produce nearly twice as many qualifying entities as the other two types, and the Dutch share of qualifiers moves by fifteen points depending on what is being asked.

Discovery questions open the shelf. Asked “wat zijn goede Nederlandse kledingmerken?” or “welke kledingmerken zijn nu populair bij twintigers en dertigers?”, the models list. 410 distinct entities appear in 375 discovery responses and 48 clear the threshold — but this is partly arithmetic, since the threshold is a rate and the cut is a third the size of the general one. distinct_entities is not compared across cuts of different sizes anywhere in this report.

The Dutch share moves with the question type, and not in the obvious direction. Dutch entities are 48.3% of use-case qualifiers, 43.8% of discovery qualifiers and 35.7% of attribute qualifiers. Attribute questions — durable jeans, cotton that does not shrink, wool that does not pill — pull toward international names with established material reputations. Use-case questions — a wedding dress in size 46/48, trousers in inside leg 36, a commuter raincoat — pull toward Dutch specialists. The assistants appear to treat “who makes good fabric” and “who fits my body in this country” as different questions.

Three different leaders. Uniqlo takes discovery, Kuyichi attribute and Studio Anneloes use-case. Studio Anneloes leads no other cut in the study.

7. Why There Are No Segment Cuts

One-sentence takeaway: This edition publishes no product segments, because no honest three-way split of the question set exists, and saying so is better than publishing an arbitrary one.

Previous editions in this series rank entities within product segments as well as overall. This edition does not, and the reason is in the question set rather than in the data.

The questions export carries no subcategory. Assigning one after the fact produces four natural groups rather than three — quality and materials, fit and sizing, occasion and use, and webshop service — and the fourth holds only four questions. Three questions have no product content at all: “waar koop je tegenwoordig nog kleding van écht goede kwaliteit?” asks about quality in the abstract. Roughly eight questions sit on two axes at once: question 32 asks for a festive dress and specifies size 46/48, which is an occasion question and a sizing question in the same sentence.

A three-way split can be forced with a precedence rule — size constraint first, then occasion, then everything else — and it yields 13, 8 and 24 questions. The eight-question segment sits exactly at the floor below which this series treats a segment as indicative only, and the 24-question group is a residual rather than a category.

Publishing that would mean publishing a segment ranking whose composition was decided by a tie-break rule invented after the answers were collected. Product category and audience were instead recorded as attributes of the entity dictionary, where they describe what a brand sells rather than what a question asked. They are not a measured cut and no leaderboard in this report is built from them.

Edition 2 will carry segments assigned at collection time, which is the only point at which they can be assigned honestly.

8. Marketplaces, Retailers and the Out-of-Scope Shelf

One-sentence takeaway: The assistants answer Dutch clothing questions with shops as readily as with brands, and when asked for “Nederlandse merken” some of them answer with soup.

8.1 Retailers and marketplaces are ranked

Four pure retailers and one marketplace qualify in the general cut, and ten more entities are Brand+Retailer. They are ranked, because they are what the assistants actually name: asked where to find trousers that a tailor can alter, or which webshop refunds quickly, the answer is a shop.

Wehkamp (86 responses), Vinted (83) and De Bijenkorf (62) qualify. Marktplaats (30), Project Cece (28), HEMA (27), Decathlon (25) and Vestiaire Collective (19) do not. The second-hand layer is unusually visible for a clothing study: Vinted ranks tenth overall, and Project Cece, Sellpy, United Wardrobe, Depop and IJ-Hallen all appear, which is a direct consequence of the questions that ask about tweedehands kleding and about clothes that last.

8.2 The entities that answer the wrong question

36 entities are classified Out-of-scope and appear in no leaderboard and no origin share. They fall into three groups, and the second is a finding rather than a classification chore.

Non-fashion brands returned to Dutch-brand questions. Asked “wat zijn goede Nederlandse kledingmerken?” and “ik koop liever iets van een Nederlands merk dan van een grote internationale keten”, the assistants named Unox, Hak, Zonnatura, Koopmans and Mora — Dutch food brands — alongside clothing labels, together with Philips, Miele, Bosch, IKEA, Jumbo and Fairphone. The models appear in these cases to answer “which brands are Dutch” rather than “which clothing brands are Dutch”, substituting the more famous question for the harder one. It is not frequent enough to distort a leaderboard, and none of these entities clears 5% in any cut, but it is a specific and reproducible failure mode on exactly the questions this market cares most about.

Service and platform brands. Coolblue, PostNL, Klarna, PayPal, Riverty and AfterPay appear in answers about returns, delivery and refunds. Coolblue is the only one to clear 5% anywhere.

Publishers and certifications. ELLE is the instructive one. It came back from entity research classified as a clothing brand, which is defensible — there are licensed ELLE apparel lines — but elle.com is cited in 106 responses as a magazine, and counting it as a brand-owned source would have inflated that bucket by a fifth. It was reclassified Out-of-scope after the source audit. GOTS, OEKO-TEX, Fair Wear and Tencel were dropped as certifications and ingredient trademarks before research.

9. The Source Ecosystem

One-sentence takeaway: A forum is the most-cited domain in the study, editorial outranks brand-owned sites, and the gap between being cited and being named is wide enough to be its own finding.

Reddit leads, and no brand site comes close. reddit.com is cited in 159 of 1,125 responses. The next four are whensarasmiles.nl (107), elle.com (106), kuyichi.com (100) and cosmopolitan.com (81). The most-cited brand-owned domain in a study of clothing brands is a Dutch denim label’s own site, in fewer than one answer in ten.

Retailers supply the most citations, editorial comes second. Retailer domains account for 30.89% across 230 domains and editorial for 21.49% across 177, against brand-owned at 17.13%. What an assistant says about a Dutch clothing brand is mostly assembled from what retailers and magazines wrote about it, not from what the brand published.

Being cited and being named are different measurements. houseofeinstein.nl is cited in 43 responses; House of Einstein is named in 2. hetbroekenhuis.nl is cited in 34 and Het Broekenhuis named in 1. overhemden.com is cited in 42 and named in none. A domain can carry an answer without the organization behind it appearing in that answer, and this report measures the second thing. This is also why the cited-domain backstop exists: it checks that no frequently cited commercial domain lacks a dictionary entry, and in this edition it found no missing ranked entity, but it did find the extraction defect described in Chapter 2.4.

“Other” is 25.46% of citations and that is higher than this series likes. It is 721 domains, almost all cited once or twice, and the classifier was improved twice before publication rather than tuned until the number looked acceptable. The figure is published as it stands.

10. What the Patterns Suggest

One-sentence takeaway: Five readings are available from this data, and each has something in it that cuts the other way.

Dutch sustainable denim is the most visible Dutch cluster. Kuyichi, MUD Jeans and Kings of Indigo all qualify in the general cut, and Kuyichi is the most-named entity in the study. Reading: a small group of Dutch brands with a clear, well-documented sustainability story has achieved visibility out of proportion to its retail footprint. Against it: three of the four best-covered Dutch names occupy the same narrow niche, and on the origin-neutral shelf the first Dutch entity is fifth. Depth in one story is not breadth in a market.

Position and reach measure different things, and this market separates them sharply. YOEK is named in 71 of 1,125 responses (6.31%) and holds the position maximum; Arket is named in 122 of 1,125 (10.84%) and ranks eleventh. Reading: assistants have both a “first answer” shelf and a “also consider” shelf, and the two are populated by different brands. Against it: the composite deliberately rewards position at 30%, so some of this separation is the instrument rather than the market, which is why every table prints mention rate and MRR alongside the score.

The shelf is retail as much as brand. Zalando is third overall, Vinted tenth, and second-hand platforms appear throughout. Reading: for Dutch clothing questions the assistants route discovery through where to buy rather than what to buy, more than a pure-brand study would suggest. Against it: several questions explicitly ask about webshops, returns and second-hand, so the question set invites the answer.

Fit and sizing questions surface specialists. The use-case cut has the highest Dutch share of qualifiers at 48.3%, and names like Girav, Joe Merino, MS Mode and Zizzi appear on tall, plus-size and length-specific questions. Reading: local specialists win where body-specific knowledge matters, which is the hardest thing for an international feed to answer. Against it: these are the cuts with the smallest denominators and the thinnest qualifiers, and two of the seven single-model-dependent rankings sit here.

Nobody’s own website is doing the work. Brand-owned sites supply 17.13% of citations while retailers, editorial and forums supply 55.24% between them. Reading: brands have less control over what an assistant says about them than the assistant’s fluency implies. Against it: citation share is not influence share — this study records which domains a model cited, not which shaped its answer, and 29.3% of one model’s answers cited nothing at all.

No recommendations follow from any of this. This is a measurement, not a strategy document.

11. Limitations and Notes

Single collection window. 1,125 responses were collected on 11 September 2026 between 10:07 and 12:38 UTC, and all 1,125 were analyzed. This is a snapshot, not a trend. It is the first edition of this series, so no cross-edition comparison exists and none is implied.

Miss Etam is a boundary case for the origin rule. The Dutch chain was founded in Amsterdam in 1923 and coined the Miss Etam name in 1974, but the ETAM root is a German company of 1916 that continues today as the French Etam Groupe. Under the declared trademark rule Miss Etam is Dutch and is published that way, at medium confidence. A reader who thinks of it as a foreign name is describing its lineage, not its Dutch trademark. It does not qualify in the general cut, so no headline figure turns on it.

One brand exists in the dictionary only because a sub-brand pointed at it. Rogelli was never named alone in any response — the answers say “Rogelli Commuter Core Regenjas” — and entered the dictionary as a promoted parent inheriting its sub-brand’s research. Fear of God and River Island are the same shape and matched nothing at all. Where a market names lines rather than houses, this series will undercount the house.

Seven qualifiers rest on fewer than three models, listed in full in Chapter 5.4, and one of them — Suitable in the attribute cut — rests on a single model. Their visibility is model-specific rather than market-wide. Every one is published with its breadth column intact.

No segment cuts. Explained in Chapter 7. This is the most significant structural limitation in the edition and it is a limitation of the question design, not of the data.

The Dutch-brand cut is thin. Twelve questions and 300 responses against 33 and 825 on the origin-neutral side. Every origin claim in this report carries both figures, and the constrained cut’s 5% threshold is 15 responses.

Smaller cuts qualify more entities. The 375-response discovery cut qualifies 48 entities against 30 in the 1,125-response general cut, because the threshold is a rate. That is arithmetic.

43 entities have Unknown trademark origin. They are excluded from origin shares rather than counted as Foreign, and none of them qualifies in any cut, so no published share is affected.

Three extraction defects were found and fixed before publication, described in Chapter 2.4: brands that only ever appear with a product name attached, aliases colliding with ordinary Dutch garment prose, and eleven cases of one company sitting under two dictionary rows. The third moved a published figure — G-Star RAW qualifies in the general cut only once its two halves are merged. All three were found by automated checks rather than by reading the output, which is the argument for running them.

The tail was not researched individually. 905 candidates appearing in three or four of 1,125 responses were triaged by evidence band; the 387 with the strongest proper-noun evidence were researched and the remainder dropped with reasons. None of them could have qualified in any cut, since 5% of the smallest cut is 15 responses. The measured hit rate of the band rule was 64% for the strong band against 58% for the middle one, which is a weaker separation than the rule assumes, and is the reason both bands were researched rather than only the first.

“Other” is a quarter of source citations. 25.46%, across 722 mostly single-citation domains. The classifier was improved but not tuned to clear a threshold.

The question set is fixed and partial. 45 questions cannot cover a clothing market. They were written to span discovery, attribute and scenario, and they lean toward quality, durability, fit and sizing because that is what Dutch consumers ask. They are not a sample of Dutch search demand and no weighting was applied to make them one. A brand strong in a category the question set touches lightly — footwear, accessories, childrenswear — will be under-represented here.

Never compare scaled scores across cuts or editions. A score of 100 in one cut is a different absolute level of visibility from 100 in another, because the scale moves when the leader moves. Cross-edition comparison must use mention_rate, mrr and breadth.

Frequently Asked Questions

12. Appendix & Data

12.1 All 45 questions, as asked

Questions are reproduced exactly as collected, including lower-case openings, missing punctuation and English loanwords. The English column is a gloss for readers, not a translation used anywhere in the analysis: every entity match and every ranking was computed on the Dutch text.

12.2 All tracked entities (720)