1,125 responses, five assistants, four product segments: which bedding, towel and curtain brands does AI actually name in Turkey?

This is a recurring, neutral study of a question nobody had measured: when someone in Turkey asks an AI assistant which sheets, towels or curtains to buy, which brands does it name?

We ask five large language models the same Turkish-language questions real people type, repeat each one five times, and record how often, how early, and across how many assistants each brand 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

Three facts sit next to each other, and the gap between them is the reason for this research.

1. This is a substantial industry, and unusually well-defined. On the İTKİB/İTHİB basket — the one current national series with a fully disclosed product definition — Turkey exported US$1,828m of home textiles in 2025, against US$1,888m in 2024, and imported US$138.4m against US$124.1m. That places Turkey fifth in the world with 3.0% of global home-textile exports (İTKİB Genel Sekreterliği / İTHİB, Dünya Ev Tekstili Sektörü Dış Ticaret Raporu, 16 July 2026, using ITC Trade Map data). Germany took US$295m of it (16.1%), the United States US$270m (14.8%), then France, Italy and the United Kingdom.

That basket must be named whenever the figure is used. It is built from specified GTIP codes: bed linen, table linen, towels, curtains, bedspreads, blankets, quilts and pillows, plus bathrobes coded as apparel and HS 5805 wall tapestries. It excludes floor carpets and rugs, and excludes home-textile fabrics. Combining carpets into the total is the principal way much larger “home textile” figures reach the press, and we do not do it.

Production is genuinely clustered, and one province can be measured precisely. Denizli registered US$750.7m of exports in 2025 across bathrobes, terry towels and sheets/duvet covers alone — 20.90% of all DENİB-registered exports (Denizli İhracatçılar Birliği, December 2025 monthly report; towels alone accounted for 11.39%). Bursa hosts SETEK, a formally recognised sustainable home-textile cluster supported under the Ministry of Industry and Technology’s clustering programme (BEBKA), and holds the registered Bursa Havlusu geographical indication (TÜRKPATENT no. 1519, registered 9 January 2024). Buldan Bezi (no. 194, 2016) and Şile Bezi (no. 272, 2017) are registered as well. Uşak is defensibly a blanket and recycled-textile ecosystem, though the most product-specific official study we found dates from 2019 and we do not repeat its estimates.

Two things we could not establish, and will not estimate: home textiles as a percentage of Turkey’s total merchandise exports on any published basis, and a sector employment figure with a stated reference year. A current association statement puts employment above a rounded threshold but does not say for which year, so we leave it out.

2. The category is precisely defined and almost entirely unmeasured. This is the fact that makes the series worth publishing, and the shape of the gap is unusual.

TÜİK can identify home textiles narrowly — COICOP basic heading 05200, “Ev tekstili”, covering curtains, sheers, bedspreads, quilts, blankets, duvet sets, pillows and towels. But that is a consumer price index heading, not a market measurement. There is no published 2025 Household Budget Survey value isolating it. The retail sales volume index bundles it into “Tekstil, giyim ve ayakkabı.” Neither the industrial production index nor the turnover index exposes a home-textile series publicly. Relevant businesses span at least five NACE codes across manufacturing, wholesale and retail, so SGK publishes no consolidated employment total.

In e-commerce it is worse. The Ministry of Trade’s Türkiye’de E-Ticaretin Görünümü (12 May 2026) places home textiles inside “Ev, Bahçe, Mobilya ve Dekorasyon,” worth TRY 215.57bn in 2025 — a category that also contains garden, furniture and decoration, and cannot be described as an online home-textiles market. That headline grew 50.0% year on year, but this is nominal growth in current lira: average consumer prices rose 34.88% over the corresponding twelve-month comparison (TÜİK, 5 January 2026), and the Ministry publishes no real growth rate. It publishes no online-penetration rate for the category either — the report’s penetration chart covers six selected sectors, and this is not one of them.

So a Turkish home-textile brand can find its category in the CPI basket and in the customs nomenclature, but cannot look up its category’s domestic market size, its online penetration, its retail sales trend, or its employment. Until now it could not look up its AI rank either.

3. Turkey uses AI assistants heavily — and leans on the one that answers from memory. On the most recent GWI country cell we could verify directly, 39.7% of Turkish internet users had used ChatGPT in the previous month, 11th of 54 markets against a worldwide 26.5% (DataReportal/Kepios using GWI Q2 2025, published 15 October 2025). A newer wave exists — GWI Q4 2025, reported 22 April 2026, with a worldwide figure of 31.2% — but its Turkey row is not publicly exposed, so we quote the older cell and say so. Turkey’s own statistics office puts generative-AI use lower, at 19.2% of recent internet users aged 16–74 (TÜİK, 1 October 2025). These measure different products, different age groups and different recall periods; they are not competing estimates and must not be averaged.

The distinctive Turkish fact is which assistant. On Statcounter’s July 2026 AI-referral data, ChatGPT accounts for 76.31% of Turkey’s AI-originated web referrals against 77.92% worldwide — slightly below the global average — while Gemini takes 18.18% in Turkey against 9.90% worldwide. Turkey’s Gemini share exceeded the worldwide figure in all four months from April to July 2026. This is a measure of referrals, not of people: Statcounter’s denominator is traffic, not users.

That matters for this study more than it looks. Gemini is the assistant whose behaviour diverges most in our data: it grounded only 30.7% of its answers against essentially 100% for the other four, has the narrowest vocabulary of the five at 111 distinct entities, and names Taç in 89.8% of its responses. The assistant Turkey over-weights is the one that answers most from memory.

The gap. A fifth-placed export industry with a precise customs definition and measurable regional clusters. A shopping occasion — the çeyiz — that furnishes an entire household at once. A population that has adopted AI assistants quickly and over-weights the one that retrieves least. And no public measurement of which home-textile brands those assistants actually recommend. That is what this series publishes.

What the first edition found

The August 2026 baseline produced five findings worth reading in full.

→ Read the August 2026 edition

What a high score does not mean

A high AI Visibility Score means one thing: these systems currently have information about a brand and surface it readily when asked. It is not a statement that the brand is better made, better value, more durable, more popular, better selling, or larger in market share.

This is worth more than a disclaimer here, because in this category the comparison mostly cannot be attempted.

• Quality is not measured. We record whether a brand is named and how early. We do not assess whether the recommendation was accurate, whether the description was positive, or whether the product performs. There is no sentiment layer and no product testing. Thread count, GSM, fibre content, weave and colour fastness are all real product attributes, and none of them is in our data.
• Sales are not measured, by us or by anyone in this category. No published study estimates the incremental sales caused by a brand being named in an AI answer. Nor does anything isolate home textiles: the closest category evidence we found is Adobe’s, which places “Bedroom Linens/Furniture” in a qualitative “strong boost” tier for AI-referral growth without publishing a numeric rate, and combines textiles with furniture in every cut it does publish. Similarweb’s home-goods figures are domain-level visits, not product traffic. Both companies sell AI-visibility products.
• Market share cannot be publicly checked. Our research pass looked for a ranking of Turkish home-textile brands by value share with a disclosed market definition, brand-level values, a stated period and a described method. It found none that meets all four. The honest qualification is that Euromonitor publicly advertises a paid “Brand Shares … % Value 2021–2024” table inside its Home Furnishings in Turkey report, so credible commercial brand-share data plainly exists — it is simply not inspectable, and its market boundary cannot be matched to home textiles from the free material. What we found in public instead were rankings measuring something else: a Nielsen 2019 study of awareness and liking, a 2025 consumer study of “ev yaşam mağazaları”, and a digital-reputation ranking. None is a sales-share measurement, and none should be reported as one.

Company accounts do not fill the gap either. Of the sixteen most prominent brands we checked, only two sit inside a clearly identifiable listed issuer: Yataş (Borsa İstanbul, YATAS — audited FY2025 consolidated revenue TRY 22,658,162,029) and Soley (BRMEN, on the Pre-Market Trading Platform — TRY 200,225,505). Both figures are issuer-level and cover multiple brands; Yataş Group alone spans Yataş Bedding, Enza Home, Divanev and Puffy. Taç and Linens sit within Zorluteks, consolidated into Korteks, whose FY2025 audited revenue of TRY 16,295,837 thousand covers several companies and is not either brand’s revenue. Özdilek, Karaca, Chakra, English Home, Madame Coco, Bella Maison, Cotton Box, Penelope, Brillant, Ecocotton, Hamam and Sarev publish no accessible audited revenue at all.

Store counts are worse. Only one survives a basic sourcing test: Yataş Group reported 100 company-owned stores and 709 domestic dealer-owned stores at year-end 2024 (KAP, 10 March 2025) — and that is a group network, not a single brand’s. Everything else we found was undated, approximate, or mixed owned stores with franchises and sales points without saying which.

So when a brand ranks first here, the honest reading is narrow and specific: it is the most findable, not the most bought. Claims of market leadership in this category should be treated as vendor claims unless they arrive with a named research provider, a market definition, a period and a percentage.

What we measure, and how

• Five models, queried directly. Claude Haiku 4.5, Gemini 3.6 Flash, GPT-5.6 Luna Pro, Perplexity Sonar and Grok 4.3 — via API, no system prompt, provider-default temperature, no locale set. The questions are in Turkish; the models infer the market from language alone.
• Real questions, repeated, and never rewritten. Each edition uses Turkish-language questions of the kind people actually type, asked five times each to capture run-to-run variation. Ambiguous questions are kept as asked — rewriting a question because its answers are inconvenient is selection on outcome.
• Entities discovered from the answers, then reviewed by hand. Candidates are extracted from the responses themselves, passed through a human review gate, and researched externally before any metric is computed. Origin is assigned by trademark ownership, so a foreign mark made or sold in Turkey under licence stays foreign.
• A three-component score. AI Visibility Score = 0.45 × Mention + 0.30 × Position + 0.25 × Breadth — how often a brand is named, how early it appears, and how many of the five assistants know it.
• Ten cuts, each on its own published scale. Every edition is read across product segments, origin scope (questions that demand Turkish brands versus questions that don’t), and behavioural types (discovery / attribute / use-case). Each cut has its own denominator and its own scale, fixed once and never recomputed. Turkish-versus-foreign comparisons are made only on origin-neutral questions — the only fair basis.

What we do not measure. Quality, accuracy, sentiment, price, durability or safety. A score is not an endorsement.

Editions

The wider context: is AI discovery real yet?

We think the honest answer is “real, measurable, growing from a small base — and smaller than the vendors say.” The evidence, weighed:

AI answers demonstrably suppress traditional clicks. The strongest independent measurement remains Pew Research Center, which observed 68,879 Google searches by 900 US adults in March 2025: when an AI summary appeared, users clicked a traditional result in only 8% of visits against 15% without one, and clicked a source inside the summary just 1% of the time. A July 2026 preprint using Comscore desktop clickstream data from 45,386 households found an outbound visit in only 5.2% of ChatGPT conversation sessions, against 31.1% of Google search sessions. Much of the time, the assistant answers and the click never happens.

Referral traffic is growing fast, but the conversion story is vendor-owned and it has moved. Adobe reported AI-referred US retail traffic up 138% year on year to May 2026, converting 54% better than non-AI traffic. That is worth reading against Adobe’s own earlier data: in February 2025 the same channel converted 9% worse than other traffic, and in July 2024, 43% worse. Similarweb separately estimated ChatGPT-referred retail visits converting at 11.4% against 5.3% for organic search. Adobe and Similarweb both sell analytics and AI-visibility products. The direction is credible; the magnitudes are not settled, and the base is small and undisclosed.

AI’s influence is larger than its clicks — probably. A June 2026 working paper matching opt-in users’ ChatGPT, Claude and Gemini conversations to their browsing found that when an assistant recommended an unfamiliar brand, the probability of subsequently Google-searching that brand rose 4.3 percentage points and of visiting the brand’s own site 2.4 points. Referral statistics therefore understate AI’s effect on which brands people go looking for. (Not peer-reviewed; the authors are affiliated with a company selling AI-visibility products.)

But the assistants are not yet good shopping advisers. Two 2025–26 benchmarks are sobering: one scored leading models at 11.22% and 3.92% on a composite shopping benchmark of 120 expert-curated tasks, documenting reliance on promotional misinformation; another found pass rates of 57–77% across 525 shopping missions, weakest on multi-turn dialogue and optional requirements.

And home textiles has no category evidence at all. This is the honest position, and it is worth stating rather than filling. We found no study — academic or behavioural — of AI-assisted shopping for bedding, towels or home textiles specifically. We found no Turkish evidence comparing online and in-store return rates or purchase confidence for the category. The temptation is to reason from product attributes: home textiles is specification-rich, and peer-reviewed work does show that AI recommenders fare better on functional attributes than experiential ones (Longoni & Cian, Journal of Marketing, 2020; Ruan & Mezei, Journal of Retailing and Consumer Services, 2022) — but those studies tested other categories, and a 2025 study of search-versus-experience products used a laptop and a VR headset. Bedding is both specification-rich and tactile. Nothing published tells us which side wins, and we are not going to guess.

Which is precisely why a baseline matters now. The measurement is most valuable before the behaviour becomes mainstream. If AI-assisted home shopping in Turkey grows the way general AI shopping has, the brands visible when it happens will not be the ones that start optimising afterwards.

What the research actually says about how models pick brands

We are careful not to sell certainty here, because the academic literature does not supply it.

There is no established model of how commercial assistants decide which brands to name. What exists is evidence for individual mechanisms, tested separately and on different models. Brand-origin bias is real but not simple: an EMNLP 2024 study found global brands disproportionately associated with positive attributes, while a FAccT 2026 study found geographic preference varies by model and that stating the user’s location moves rankings more than stating a brand’s origin — so there is no universal foreign-over-local rule. Position matters, conditionally: the well-known “lost in the middle” finding was qualified by EMNLP 2025 work showing that in realistic retrieval pipelines, sophisticated reordering did not beat random shuffling. Content optimisation is weaker than it is sold as: the founding GEO paper (KDD 2024) reported 30–40% visibility gains, but supplied sources to the model after retrieval rather than testing whether an edited page gets retrieved from the live web at all — and a NeurIPS 2025 benchmark found most conversational-SEO rewrites ineffective or negative, with gains shrinking as more competitors adopt them.

One gap is directly relevant to our own method, and this edition sharpened it. Because Gemini grounded only 30.7% of its answers, we could for the first time compare one model’s grounded answers against its own ungrounded ones. The result was that grounding barely changed the answer: Gemini’s grounded top eight and ungrounded top eight are the same brands in nearly the same order, with a slightly wider repertoire when it searched (77 distinct in-scope entities against 66). That is a finding about one model in one vertical, not a general result — and it still does not solve the identification problem, because Gemini chose when to search. No peer-reviewed study isolating the effect of retrieval on brand recommendations was found.

What none of this establishes is which mechanism dominates in a live shopping answer, or that more AI mentions cause more sales — for which there is almost no independent evidence at all.

So this series does the one thing that can be done rigorously: measure the output. We do not claim to explain why a model names a brand. We record that it does, how often, and how early — and we publish the method so anyone can check us.

Data, caveats and sources

• Point-in-time. Each edition is a snapshot from a single collection window. Model versions and web indexes shift; trend claims become possible only from the second edition onward.
• Visibility is not quality, sales or market share. Repeated because it matters, and because in this category none of the three can be checked against public data.
• Market figures in this hub carry different confidence levels. Trade, e-commerce, price, marriage, household and geographical-indication figures are official or official-derived (TÜİK, Ministry of Trade/ETBİS, TÜRKPATENT, KAP, SGK, DENİB, İTKİB/İTHİB, development agencies). AI referral and conversion statistics are commercial or vendor-published and are labelled as such where used. Turkish-lira growth rates should be read cautiously: in a high-inflation economy, nominal growth is not real growth, and we give the matching inflation comparator rather than subtracting one from the other.
• Different bases are not interchangeable. The US$1.83bn export figure is the İTKİB/İTHİB ITC Trade Map coded basket, and should always be named as such. It excludes floor carpets and home-textile fabrics but includes bathrobes coded as apparel. Larger “home textile” totals in circulation generally include carpets or undisclosed product sets; at HOMETEX 2026, senior speakers gave materially different rounded figures for the same year without disclosing baskets. We report the one series whose definition is published.
• Some things are simply not public, and we say so. There is no published Household Budget Survey value for home textiles, no home-textile-specific monthly retail series, no consolidated SGK employment total, no ETBİS penetration rate for the category, and no representative national estimate of çeyiz spending or of the home-textile share within it. There is no statistic on how many households buy a complete set from one brand. We flag these gaps rather than filling them with estimates.
• The çeyiz occasion is real; the “one brand for the whole home” story is not established. TÜİK’s Family Structure Survey found 60.3% of households still practise çeyiz serme (2021 fieldwork, 19,430 households) — but that measures displaying a trousseau, not buying one. Peer-reviewed work on pre-marriage purchasing (n=463, snowball sample, published 2022) measured information search for a pooled basket including home textiles: 52.5% visited physical stores, 43.2% used the internet, 36.9% consulted family or friends. That is research behaviour, not channel share, and the sample is not nationally representative. When one brand appears in 56% of AI answers, that is a fact about the assistants, not a measurement of what Turkish households buy.
• The demand picture is genuinely mixed, not a simple decline. Marriages fell to 552,237 in 2025 from a revised 569,983 in 2024 (TÜİK, 24 February 2026) — but 2024 had risen from 2023, so there is no sustained monotonic decline. Meanwhile household units rose to 26,977,795 while average household size fell to 3.08 (TÜİK, 12 May 2026). More homes, fewer people in each. Both are demand context; neither is a sales forecast.
• A note on the İstanbul “wedding cost” figures sometimes quoted. The İstanbul Planning Agency’s 2024 basket puts trousseau shopping at TRY 170,099, up 68.7% on 2023. That is a constructed 79-item İstanbul basket, not a survey of what couples spend, its trousseau component mixes home textiles with appliances, decoration and kitchenware, and its growth is nominal — against a contemporaneous annual CPI rate of 71.60%. It cannot yield a home-textile share and we do not use it as one.

Frequently asked questions

About Herm.io & disclosure

Herm.io is a consumer behaviour and marketing data company. We study how people discover and choose brands to help businesses reach the right customers. This report forms part of our public research and is conducted quarterly.

Disclosure & neutrality

The rankings and metrics in this report are entirely impartial and are based on the methodology described within the report. No brand can pay or provide sponsorship to be included in the report, improve its ranking or influence how it is described in the content.

A brand’s score in the report is not an endorsement or assessment of quality; it is solely a measure of its current visibility within AI models. No service or product offered by Herm.io is sold for the purpose of directly changing the results of this report or guaranteeing a ranking. To preserve objectivity, all references to Herm.io’s own domain are removed from the source data before analysis. This ensures that the company does not appear in, measure itself through or benefit from its own study.

Brands seeking to understand their position in the data can schedule a consultation for an impartial assessment of the findings. The consultation is advisory and free of charge.

Quarterly series · Baseline edition published 17 August 2026