This is a recurring, neutral study of a question nobody had measured: when someone in the Netherlands asks an AI assistant what to wear or where to buy it, which brands does it name?
We ask five large language models the same Dutch-language questions real people type, repeat each one five times, and record how often, how early, and across how many assistants each entity appears. The result is an AI Visibility Score — a measure of findability, not of quality. A brand ranks highly here because these systems have information about it and surface it readily. That is all it means.
Why this series exists
Dutch clothing is thoroughly measured in aggregate and barely measured at brand level.
The aggregates are solid. CBS puts Dutch household spending on clothing and textiles at €20.839 billion in 2025, provisional, in its national-accounts consumption table — footwear is classified separately there, so this figure is clothing and textiles rather than clothing and footwear. The Thuiswinkel Markt Monitor, executed by GfK for Thuiswinkel.org and Retail Insiders, puts 36% of the clothing market online in calendar 2025, excluding shipping and returns, down from 37% in 2024. CBS counted 12,350 registered clothing-retail locations at 1 January 2026 under SBI 4771 — business registrations rather than storefronts, rounded to multiples of five.
What does not exist is a public brand ranking. We looked for a freely inspectable ranking of clothing brands by Dutch sales or sales market share and could not verify one. The closest current Dutch source is Emerce’s Online Damesfashion 2025–2026 report, which uses Trendata search-intelligence data drawn from around five billion daily searches and finds Zara ranking first in search behaviour. That is a measure of what people look for, not what they buy, the detailed report sits behind a subscription, and it should not be read as a sales ranking.
So a shopper, a brand manager and a journalist all face the same gap. Everyone knows how large the market is. Nobody can point at a public, checkable list of which clothing brands are most visible to Dutch consumers — and increasingly, the place a Dutch consumer meets that list is an AI assistant.
What this series adds, and what it does not
This study does not measure sales, market share or quality, and it is not a substitute for the ranking that does not exist. It measures one specific, newly important thing: which brands five AI assistants name when asked real Dutch consumer questions, how early they name them, and how consistently across models.
Every figure is published with the data pack behind it so that any number can be re-derived.
What the first edition found
| # | Entity | Origin | AI Score | Mention rate | Responses |
|---|---|---|---|---|---|
| 1 | Uniqlo | Foreign | 96.13 | 17.96% | 202 / 1,125 |
| 2 | Kuyichi | Dutch | 94.62 | 19.47% | 219 / 1,125 |
| 3 | Zalando | Foreign | 79.43 | 14.76% | 166 / 1,125 |
| 4 | MUD Jeans | Dutch | 71.41 | 13.87% | 156 / 1,125 |
| 5 | Studio Anneloes | Dutch | 71.01 | 8.71% | 98 / 1,125 |
| 6 | COS | Foreign | 70.02 | 12.98% | 146 / 1,125 |
| 7 | YOEK | Dutch | 69.59 | 6.31% | 71 / 1,125 |
| 8 | H&M | Foreign | 68.38 | 13.42% | 151 / 1,125 |
| 9 | Suitsupply | Dutch | 67.00 | 7.82% | 88 / 1,125 |
| 10 | Vinted | Foreign | 65.44 | 7.38% | 83 / 1,125 |
The leader and the most-named entity are not the same brand. Kuyichi appears in more answers than anything else in the study. Uniqlo appears in seventeen fewer and still takes the top score, because when it appears it appears earlier. The composite weights position at 30% for exactly this reason: being mentioned and being recommended are different things.
Dutch brands hold three of the top ten when nothing asks them to. On the 33 origin-neutral questions (825 responses), 8 of 28 qualifying entities are Dutch. On the 12 questions that explicitly ask for a Dutch brand (300 responses), 31 of 34 are Dutch and the top ten is entirely Dutch — which is a property of the question, not a finding. Both figures appear together everywhere in the edition.
The assistants disagree about how much to say more than about who to name. Grok names 14.11 entities per response; Claude Haiku names 3.65. Three of the five models grounded every answer in a live web search, GPT-5.6 grounded 97.3% of its 225, and Gemini grounded 29.3% of its 225.
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%.
What a high score does not mean
An AI Visibility Score says an assistant named a brand, and how early. It does not say the brand:
- makes better clothes, or clothes that last
- sells more, or holds more market share
- fits any particular body, which is what most of these questions are actually about
- is still trading, or trading in the Netherlands
- was recommended for a good reason, or a correct one
A brand can be highly visible and mediocre. A brand can be excellent and invisible, which is the more common case for small labels with thin web coverage. The score measures a discovery channel, not a product.
What we measure, and how
Each edition asks five large language models 45 Dutch-language questions, five times each, with no system prompt and no locale parameter. The questions are written the way Dutch consumers actually write them, including lower-case openings, English loanwords and imperfect grammar, and no question is rewritten because its answers were inconvenient.
Every entity named in an answer is identified by alias matching anchored on word boundaries, after link targets and URLs are stripped so that a cited domain cannot inflate a brand’s count. Candidate entities are then researched and classified by a human before any number is computed: origin by trademark ownership, entity type, and scope.
The score has three components, computed separately within each cut:
ai_score = 0.45 × mention + 0.30 × position + 0.25 × breadthMention is how often an entity is named, position is how early, breadth is how many of the five assistants name it at all. Scores are scaled against the leader of each cut, which means a score is only meaningful inside its own cut and its own edition. Comparisons across editions use the raw measurements — mention rate, position and breadth — never the score.
Editions
| Edition | Period | Responses analyzed | Entities tracked | Qualified (general) | Leader |
|---|---|---|---|---|---|
| 1 (baseline) | September 2026 | 1,125 | 720 | 30 | Uniqlo (96.13) |
The series is quarterly. Edition 2 will be able to compare mention rate, position and breadth for any entity present in both, and will add product segments assigned at collection time — this first edition publishes none, because the question set does not divide into them cleanly and inventing a split after the answers arrived would not have been honest.
The wider context: is AI discovery real yet?
Enough to measure, not yet enough to dominate.
On the consumer side, Adyen’s survey with Censuswide fieldwork, run 10 February to 12 March 2025 across 2,000 Dutch consumers aged 16 and over, found 25% had used ChatGPT or another AI assistant while shopping in the previous year. That is a quarter of shoppers already in the channel, from fieldwork now more than a year old.
On the traffic side, Statcounter reports that ChatGPT supplied 81.67% of Dutch AI-chatbot referrals to websites in August 2026, with Gemini at 8.73%, Perplexity at 3.23%, Copilot at 3.19% and Claude at 3.18%. That measures referrals reaching websites, not which assistants Dutch people use.
For volume, a Similarweb-derived analysis of click-producing AI referrals to ecommerce, finance and travel sites put the Netherlands at 2.9 million AI-referred visits in March 2026, against 8.4 million each for the UK and Germany, 5.5 million for France, 4.7 million for Italy and 3.9 million for Spain. The Netherlands is a smaller market than all five, and the source publishes no Dutch share of the European total, so none is calculated here.
Read together: the channel exists, one assistant dominates the referrals it produces, and the volumes are still small beside conventional search. That is precisely why a baseline is worth taking now — the answers these systems give are being formed while the traffic is still modest.
Data, caveats and sources
Every edition publishes its full data pack: the metrics table with one row per entity and cut, the entity dictionary with aliases and trademark origins, the decisions log, the dropped-candidate list with reasons, the published scale registry, every leaderboard and its brand-only variant, per-model and per-question-type breakdowns, and the full cited-domain list. Any figure in any edition can be re-derived from those files.
Three caveats travel with every edition:
45 questions cannot cover a clothing market. They span discovery, attribute and scenario questions, and they lean toward quality, durability, fit and sizing because that is what Dutch shoppers ask. They are not a sample of Dutch search demand, and no weighting was applied to make them one.
A single collection window is a snapshot. Edition 1 was collected on 11 September 2026 in a window of about two and a half hours. Assistants change their answers as the web changes beneath them.
Origin is trademark ownership, and it surprises people. Local means the trademark country is the Netherlands. Belgium and Germany are Foreign, Benelux is not a rollup, and a .nl domain, Dutch marketing or Dutch availability do not make a brand Dutch.
Corrections policy
Corrections are logged rather than silently amended. Edition 1 shipped with three entity-resolution defects found and fixed before publication, each described in its methodology chapter, including one that changed a published ranking.
Error reports about any edition are welcome and are credited when they lead to a correction.
Frequently asked questions
A 0–100 measure combining how often an AI assistant names a brand (45%), how early in the answer it names it (30%), and how many of the five assistants name it at all (25%). It is scaled against the leading entity within each cut, so a score is only comparable inside its own cut and edition.
In edition 1, Uniqlo leads the composite at 96.13. Kuyichi is named in more responses — 219 of 1,125 against Uniqlo's 202 — but later in the answer. Which one you call 'most visible' depends on whether you mean named most often or named first.
No. It is a ranking of which brands AI assistants name. No freely inspectable ranking of clothing brands by Dutch sales or market share could be verified, which is part of why this series exists — but this study does not fill that gap and should not be read as if it does.
Quarterly. Each edition repeats the same questions against the same five assistants, so mention rate, position and breadth can be tracked over time. The composite score cannot be compared across editions, because its scale moves with the leader.
Because the assistants name them. Asked which webshop refunds quickly or where to find clothes a tailor can alter, the answer is a shop rather than a brand. Zalando ranks third in edition 1 and Vinted tenth. A brand-only companion table strips marketplaces by name for readers who want a pure brand view.
No. Nothing in these reports is sponsored, and no entity named in any edition commissioned it, reviewed it before publication or paid for placement. The questions are fixed before collection and the entity list is discovered from the answers.
About Herm.io & disclosure
Herm.io publishes quarterly AI Visibility studies measuring how large language models answer real consumer questions in specific markets and verticals. This series is independent research. No entity named in any edition commissioned it, reviewed it before publication, or paid for placement.
Method, data packs and corrections are published in full so that any figure can be re-derived. The publisher’s own domain is checked for in the cited-source data of every edition; in edition 1 it appeared in one response of 1,125 and that citation was excluded from the source figures.
Author: Mert Can Elkaya · Series: Quarterly · Latest edition: September 2026
Mert Can Elkaya
Contributor
I'm a product builder working at the intersection of product, fintech, and growth. From martech and venture capital to leading product at a proptech platform and co-founding a fintech startup, I help teams—and shoppers—make smarter, more confident decisions.
More from Mert