1,125 responses, 5 models, 45 Turkish questions: the first AI-visibility snapshot of Turkey's kitchen & tableware market

Vertical: Kitchen & tableware (mutfak & sofra ürünleri), Turkey
Method: A single point-in-time study of 1,125 responses across five large language models (Claude, Gemini, GPT, Perplexity, Grok), each asked 45 Turkish-language questions five times
Collection date: 1 August 2026

Key terms: AI visibility, AI Visibility Score, kitchenware brands, cookware brands, tableware brands, mutfak markaları, sofra ürünleri, çeyiz, Turkish (yerli) kitchen brands, LLM brand recommendations, ChatGPT/Gemini/Claude/Perplexity/Grok brand recommendation, Generative Engine Optimization (GEO).

This report reproduces the real Turkish-language questions consumers ask AI assistants about pots, pans, dinner sets, cutlery, storage and trousseau shopping in Turkey, and measures how often, in what order, and across how many models five large language models name each brand. Because the same 45 questions can be read by origin scope, by product segment and by question type, brands are ranked in ten separate cuts, defined in Chapter 2. It is a single point-in-time study and should be read within the limitations in Chapter 13.

How to read this report: visibility is not quality

This report measures one thing: how often, how prominently, and across how many AI assistants a brand is named when people ask about kitchen and tableware products in Turkey. A high score means these systems currently have a lot of information available about a brand and surface it readily. It does not mean the brand is better, more durable, safer, cheaper or higher quality. A low score, or a zero, means a brand is currently less visible to these systems, not that it is inferior. AI visibility reflects information availability and discovery, not product quality, materials, price or trustworthiness. Readers and shoppers should not treat an AI recommendation, or this report, as a verdict on whether a brand is “good” or “bad.” This study does not measure the accuracy, quality, or sentiment of any recommendation.

1. Executive Summary

One-sentence takeaway: Turkish kitchen and tableware answers are led by Karaca at a perfect score, but the more useful finding is that three of the six most visible entities belong to a single corporate group, and that a quarter of all answers name a brand from outside the category altogether.

• Karaca leads at 100.00 — the first perfect score in this series. It appears in 57.2% of all 1,125 responses, is named earliest on average (MRR 0.576) and is seen by all five models. Under component scaling a leader normally lands short of 100 because no single entity tops every component; Karaca tops all three simultaneously. That is a property of this cut, not a normalisation artefact, and Chapter 3 explains it.
• 23 entities qualified in the general cut. Of 144 tracked entities, 23 cleared the 5% mention threshold across all 45 questions: 17 Turkish and 6 foreign. Their average score is 58.2, and 19 of the 23 are seen by all five models.
• The top three are one company. Karaca (#1, 100.00), Emsan (#5, 70.11) and Jumbo (#6, 66.76) are all Karaca Group brands. Their combined footprint is 727 of 1,125 responses — 64.6% — and on origin-neutral questions it reaches 72.0%. A reader scanning the leaderboard sees three competitors. Chapter 9 breaks out all five corporate groups in the data.
• Segment leadership genuinely splits. Korkmaz leads Cookware (86.39, narrowly over Karaca at 85.81) and Whole-kitchen & Trousseau (98.32). Karaca leads Tableware & Serving (98.30) and Cutlery & Storage (86.98). No entity wins all four.
• Foreign visibility is concentrated in one segment. Tableware & Serving has exactly one foreign qualifier out of 21. Cutlery & Storage has IKEA at #2 (79.16) and Zwilling at #4 — the strongest foreign showing anywhere in the study.
• The origin-neutral shelf is Turkish-led but broadly contested. On the 23 open questions, mentions split 63.6% Turkish to 36.4% foreign, and 16 Turkish entities qualify against 11 foreign. But 58 distinct foreign entities appear against 54 Turkish ones: the foreign presence is wider and thinner, a long tail of European cookware names each mentioned rarely.
• Marketplaces rank low, and that is one model’s doing. Trendyol (#20) and Hepsiburada (#21) are the only marketplaces to qualify, and 96.1% and 97.4% of their mentions come from Grok alone. Four models barely name them.
• A quarter of all answers drift outside the category. 25.8% of responses name an appliance, small-appliance or cabinetry brand — Arçelik, Bosch, Beko, Arzum — rising to 51.3% on the six whole-kitchen questions. These entities are classified Out-of-scope and excluded from every table, but the drift itself is a finding about how the word mutfak is understood. Chapter 10.
• One model mostly does not search. 83.3% of all answers were web-grounded, but the figure spans 19.6% (Gemini) to 100% (Claude, Perplexity). Gemini searches when a question contains a checkable physical claim — 40.0% on attribute questions, 8.0% on use-case questions — and answers from memory otherwise. Chapter 6.
• Sources are brand-owned first. 44.2% of citations point to brands’ own domains, led by karaca.com, cited by 34.2% of all responses and by all five models.

Why it matters. Kitchen and tableware buying in Turkey runs on two distinct occasions — routine replacement and the çeyiz (trousseau) purchase — and both start with a broad question about which brands are worth considering. Whether a brand appears in an assistant’s answer to that question is becoming a discovery channel in its own right. This report captures the first snapshot of that shelf. One reminder: the numbers below describe visibility and information availability, not which brands are best.

2. Methodology

One-sentence takeaway: Five models were asked 45 Turkish questions five times each, producing 1,125 responses with none excluded; entities were discovered from the responses and reviewed by hand, and every score is computed against a per-cut scale published once and never recomputed.

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 Turkish, the models infer the Turkish market from language alone.

Scale: 45 questions × 5 repeats × 5 models = 1,125 responses; none were excluded (no model returned an empty completed body). Single collection window, 1 August 2026, 05:57–10:54 UTC. Denominators are computed per question × model cell, never as questions × models × repeats.

2.2 Web-search configuration

Web search was offered as an enabled tool to all five models; each decided for itself whether to use it (Perplexity’s Sonar is search-grounded by design). Whether a response drew on web search or the model’s own knowledge is inferred from whether the provider returned any citation or source: a citation means “web-search,” none means “own knowledge.” This is a reasonable proxy, not direct proof of the model’s internal process.

In this run the grounding rate is 83.3% overall but wildly uneven — 100% for two models and 19.6% for one. That spread matters for interpretation: one assistant answers mostly from its parameters while the others answer mostly from retrieved pages, so a brand’s visibility is partly a function of which model is asked. Chapter 6 examines it directly.

2.3 Questions, segments and scope

The 45 questions reproduce the real Turkish queries people bring to an assistant about kitchenware: general discovery (“iyi bir yerli tencere markası var mı”), specific attributes (“bulaşık makinesinde deseni çıkmayan yemek takımı hangi marka”), and use cases (“yeni eve çıkıyorum”, “çeyiz için tek markadan yerli mutfak seti alacağım”). They are balanced across three behavioural types — 15 discovery, 15 attribute, 15 use_case — and split 22 “local” (explicitly asking for Turkish / yerli brands) against 23 “open” (no origin restriction).

Questions were retained as asked, including ambiguity. No question was rewritten in this edition. Six whole-kitchen questions using the bare word mutfak were resolved by several models as kitchen cabinetry or white goods rather than kitchenware. Those questions were not rewritten and not removed — rewriting a question because its answers are inconvenient is selection on outcome. The off-market entities are classified Out-of-scope, excluded from every table, and the drift is reported as a finding in Chapter 10.

Because the question-type mix is not balanced within every segment, the number of distinct brands in one segment should not be compared with another. Whole-kitchen & Trousseau carries six questions and no attribute questions at all, and is marked indicative throughout.

2.4 Entity resolution and origin

Entities were discovered from the responses rather than from a pre-built list. Candidates were extracted case-insensitively from URL-stripped answer text with Turkish-aware case folding, passed through a single human review gate, then researched in two external passes.

399 candidates were decided: 144 Keep, 135 Merge, 120 Drop. Of the kept entities: 99 brands, 28 retailers, 6 marketplaces, 7 sub-brands and 4 licensed brands. A further 61 entities are real companies in an adjacent market — white goods, small appliances, kitchen cabinetry, sinks — and are recorded as Out-of-scope rather than deleted, which is what makes Chapter 10 measurable.

Matching is alias-based, word-boundary anchored with a Turkish letter class, tolerant of Turkish suffixes, and run on URL-stripped text so that a link target cannot inflate a brand’s count.

Origin rule: trademark. An entity is Turkish (Local) when the trademark owner is Turkish, and Foreign otherwise. A foreign mark produced or sold in Turkey under licence is Foreign, tagged Licensed-brand. Both trademark country and operator country were recorded for every entity so the rule can be changed in a later edition without re-researching. Of the 144 kept entities, 84 are Turkish and 60 foreign.

Five entities in the dictionary were never mentioned in any response — A101, Alkapıda, Boyner, Evidea and Primanova. They are retained and reported rather than deleted; the brand index in Chapter 14 lists the 139 entities that were matched at least once.

Definitions.

• Mention: a validated entity (or alias) appears in the answer text, counted once per response — presence, not frequency.
• Position (MRR): mean reciprocal of the entity’s first-appearance rank within each answer (first = 1.0, second = 0.5, …), averaged only over the responses where it appears.
• Breadth: number of the five models (0–5) that mention the entity within that cut.

2.5 AI Visibility Score (0–100)

Score = 0.45 × Mention + 0.30 × Position + 0.25 × Breadth

Metrics are computed in two separable stages. Stage 1 is raw and scale-free: mention rate, MRR and breadth, each computed within a cut. Stage 2 is the scale layer: each component is divided by the maximum reached by a qualified entity within that cut, and those maxima are published once in a scale registry that every downstream file reads. Nothing recomputes them.

Qualification (≥5% rule): only entities mentioned in at least 5% of that cut’s responses are ranked. This limits the risk that a rare entity which happens to appear first in a few answers inflates the position component. Unqualified entities are still scored, on the same cut maxima, so the appendix stays on one scale — which means some exceed 100 on a component. 21 rows, 2.3% of all metric rows, do. They are flagged and never clipped: a component above 100 is a true statement that the entity beats the published leader on that component within that slice.

2.6 The ten published cuts, and what cannot be compared across them

Every entity is scored ten times, once per cut, each with its own denominator and its own published maxima.

A scaled score is not comparable across cuts. 100 in one cut is a different absolute number from 100 in another, because the scale moves whenever the leader moves. Comparing a brand’s Cookware score with its Tableware score says nothing. Cross-cut and cross-edition comparison must use the raw layer — mention rate, MRR and breadth — which is why every table in this report publishes those alongside the score.

2.7 Neutrality and self-exclusion

This is a market-wide, neutral study with no focus brand. To prevent any conflict of interest, all citations to Herm.io’s own domains were excluded from the source data before analysis, so the company’s own content neither appears among the most-cited domains nor influences any figure.

3. Overall Visibility Leaderboards

One-sentence takeaway: Karaca tops every cut, but the shape underneath changes completely — the local shelf is entirely Turkish, the open shelf puts a foreign brand at #2, and the marketplaces sit near the bottom.

As a reminder, these rank visibility, not product quality, durability, price or safety.

3.1 General leaderboard: 23 qualified entities

Average score of the 23 qualified entities: 58.2. Origin split: 17 Turkish, 6 foreign.

Why the leader reaches exactly 100.00. Under component scaling, each component is divided by the best qualified performance in that cut. A leader reaches 100 only if the same entity is simultaneously the most-mentioned, the earliest-named and present in all five models. That rarely happens — the mention leader and the position leader are usually different entities, as they are in eight of this study’s ten cuts. Here Karaca holds all three: the highest mention rate (57.2%), the highest MRR (0.576) and full breadth. The perfect score is a description of that coincidence, not a ceiling imposed by the method.

Reading. Karaca (100.00) leads on volume and prominence together. Korkmaz (80.90) is second on a much lower mention rate (33.6%) but almost identical prominence — an MRR of 0.565 against Karaca’s 0.576, giving it 98.1 on the position component. Third and fourth are the porcelain houses, Kütahya Porselen (70.95) and Porland (70.37), both ranking on position rather than volume. Emsan (70.11) inverts that: it is mentioned more often than either of them (35.6%) but named markedly later (MRR 0.329), which costs it the position component and drops it to fifth.

The highest foreign entity is Tefal at #7 (66.55), on only 17.2% mentions but an MRR of 0.538 — when Tefal is named, it is named early. IKEA follows at #8. Six foreign entities qualify in all: Tefal, IKEA, WMF, Fissler, Zwilling and Le Creuset — every one of them a cookware or knife name, none of them in tableware.

Two entities to read carefully. Trendyol (#20) and Hepsiburada (#21) are the only marketplaces that qualify, and both rank on volume with very late positioning (MRR 0.201 and 0.161). Their visibility is also almost entirely one model’s: 96.1% of Trendyol’s mentions and 97.4% of Hepsiburada’s come from Grok, and both have a breadth of 4 rather than 5. Read as a market signal, the marketplaces are not a general feature of this shelf; they are a feature of one assistant’s answering style.

3.2 Brand-only view (marketplaces removed)

Per study scope, marketplaces are stripped from the brand-only view by name — Trendyol, Hepsiburada, n11, Amazon, Çiçeksepeti and PttAVM. Own-label retailers such as Karaca, Jumbo, IKEA and English Home are retailers, not marketplaces, and remain.

Because only two marketplaces qualified in the general cut, and both sit at #20 and #21, the brand-only top ten is identical to the general top ten. The removal takes the table from 23 rows to 21 and changes nothing above it. That is itself worth stating: unlike some verticals in this series, the Turkish kitchenware shelf is not aggregator-led. Assistants answer these questions with manufacturers.

3.3 Local leaderboard: 21 qualified entities (Turkish-only questions)

When the 22 questions that explicitly ask for Turkish / yerli brands are isolated, every one of the 21 qualified entities is Turkish. Not a single foreign name survives the filter, which shows the assistants treat yerli as a real constraint rather than a stylistic hint.

Average score of the 21 qualified local entities: 54.6. All 21 are Turkish.

Reading. Karaca (87.14) still leads, but its margin collapses: Korkmaz is at 83.88, and on the position component Korkmaz is ahead — an MRR of 0.707 against Karaca’s 0.520. When the question narrows to domestic brands, Korkmaz is the name assistants reach for first.

The local cut also surfaces an entity the general cut hides. Pirge ranks #6 (62.62) on just 7.8% mentions, carried by an MRR of 0.911 — the highest position score in the cut and the reference maximum against which every other local position score is measured. It is a knife maker, and when a question asks for a Turkish knife, it is named first almost every time. In the general cut Pirge falls just below the threshold at 4.00% and is not ranked at all.

4. Segment Leaderboards

One-sentence takeaway: Four segments, two different leaders, and radically different openness to foreign brands — Tableware & Serving has one foreign qualifier out of 21, while Cutlery & Storage has IKEA at #2.

Scores below are scaled within each segment and cannot be compared across segments. Mention rates can.

4.1 Cookware — 14 questions, 350 responses, 20 qualified

The closest race in the study. Korkmaz (86.39) edges Karaca (85.81) by half a point — and it does so while being mentioned less often (52.3% against 56.0%), winning on position instead (MRR 0.608 against 0.496). This is the only cut in which Karaca is not the mention leader and not the winner. Tefal (#4) and WMF (#5) give this segment the strongest foreign presence outside Cutlery & Storage.

Note Paşabahçe at #8 on a 5.7% mention rate: it clears the bar largely on an MRR of 0.942, the highest position figure anywhere in the study and the reference maximum for this cut. Read its rank cautiously — it is a glassware house being named first in the handful of cookware answers that reach for oven-safe glass.

4.2 Tableware & Serving — 15 questions, 375 responses, 21 qualified

The most domestic segment in the study, and by a distance. Of 21 qualified entities, exactly one is foreign — Villeroy & Boch. The top four are Karaca (98.30), Kütahya Porselen (92.29), Porland (89.28) and Güral Porselen (74.04), and the three porcelain houses between them define what an assistant considers a Turkish dinner set.

Kütahya Porselen is the mention leader here at 70.7% — higher than Karaca’s 68.0% — and is the reference maximum for the mention component in this cut. Karaca takes the segment on position (MRR 0.677 to 0.503).

4.3 Cutlery & Storage — 10 questions, 250 responses, 29 qualified

The most contested segment, and the only one where a foreign brand is genuinely near the top. IKEA sits at #2 (79.16) on a 33.2% mention rate, and Zwilling at #4 (70.08). This is the strongest foreign showing in the study, and the pattern is legible: storage containers and knives are categories where European and global names carry established reputations that assistants reproduce.

It is also the segment with the most qualifiers (29 from 250 responses), which reflects a wide, fragmented field rather than a deeper market — the qualifying bar is 12.5 responses here against 56.25 in the general cut. Pirge (#3) is the position leader at MRR 0.911; Sürbisa (#7) and Nehir (#6) are Turkish knife makers that appear nowhere in the general leaderboard.

4.4 Whole-kitchen & Trousseau — 6 questions, 150 responses, 28 qualified · indicative

Read this table as indicative, not reportable. Six questions, 150 responses, and no attribute questions at all, so the segment’s question mix is not comparable with the others.

Within those limits it is the segment Korkmaz owns outright: 98.32, on an 84.0% mention rate — the highest mention rate recorded anywhere in this study. When someone asks a broad question about kitchen brands or trousseau sets, Korkmaz is named in more than four answers out of five. Karaca follows at 81.71. These are also the questions where off-market drift is heaviest (Chapter 10), so the segment is measuring a blurrier question than the other three.

5. Differences Between Models

One-sentence takeaway: Four of five models agree that Karaca leads; they disagree enormously about how many brands an answer should contain, and Grok alone accounts for nearly all marketplace visibility in the study.

5.1 Per-model behaviour summary

Grok names about 8.15 entities per answer and reaches the widest vocabulary (96 distinct entities); GPT-5.6 Luna Pro (4.20) and Claude Haiku (4.24) are roughly half as generous. Gemini sits at 6.41 despite searching on only a fifth of its answers, which is worth noting on its own: breadth of recall is not a function of retrieval here.

The Turkish share of mentions is remarkably stable — 72.6% to 80.9% across five models with very different answering styles. Whatever else the assistants disagree about, they agree that this is a market answered mostly with Turkish names.

5.2 Each model’s most-named entities

Karaca tops four of the five models — Perplexity most heavily, naming it in 66.7% of its answers. The exception is Grok, whose top two are Trendyol (87.6%) and Hepsiburada (84.9%), with Karaca third at 82.7%. Grok is the reason marketplaces appear in this study’s general leaderboard at all; strip its responses and the shelf is manufacturers all the way down.

The models also differ in which Turkish names they favour. Claude reaches for Emsan, Tefal and Schafer; Gemini for Korkmaz, Jumbo and Porland; GPT for Jumbo and the porcelain houses. Only Karaca, Korkmaz, Emsan and Kütahya Porselen appear in every model’s top five or six.

6. Search vs. Memory: What Retrieval Changes, and What We Cannot Prove

One-sentence takeaway: Gemini searches when a question contains a physically checkable claim and answers from memory otherwise; the shelf it produces in each mode is visibly different, but the two modes answer different question mixes, so this study cannot show that retrieval caused the difference.

Previous editions in this series omitted this chapter because search was near-universal. Here it is not: 937 of 1,125 responses were grounded and 188 were not, a large enough ungrounded segment to look at. One caveat governs everything below, so it comes first: 96.3% of all ungrounded responses in the study are Gemini’s. A study-wide grounded-versus-ungrounded comparison is therefore very close to a Gemini-versus-everyone-else comparison.

6.1 When Gemini searches

Gemini decided per call whether to search. It did so on 44 of its 225 answers, and the pattern is not random.

Gemini searches when the question contains a claim someone could check, and answers from memory when asked which brands are good. It grounded on 40.0% of attribute questions against 10.7% of discovery and 8.0% of use-case questions. The three questions it grounded on all five repeats are, in translation, a dinner set whose pattern survives the dishwasher, gold trim that does not degrade in the machine, and a Turkish knife that stays sharp. The 25 questions it never grounded on are dominated by open requests to name good brands.

Grounding is also partly stochastic rather than purely a property of the question: 25 questions sat at zero grounded repeats out of five, 3 at five out of five, and 17 somewhere in between. If grounding were purely a question property every question would sit at 0 or 5; if it were purely random the distribution would cluster around the mean. Neither holds.

When Gemini did search, its answers were longer (3,069 characters against 2,155) and named more entities (7.07 against 6.08).

6.2 The two shelves are different

Retrieval surfaces marketplaces and brands with a strong own-site presence: Trendyol (+18.7 pp), Hepsiburada (+17.7 pp), Schafer (+14.4 pp), Karaca (+11.8 pp). Parametric memory surfaces established domestic manufacturers and European cast-iron names: Korkmaz (−17.1 pp), Hisar (−14.4 pp), Lava (−12.4 pp), Porland (−11.8 pp), Le Creuset (−9.2 pp), Staub (−9.1 pp).

That is a real difference in output. It is not, on this data, a demonstrated effect of retrieval.

6.3 Why we stop short of a causal claim

Gemini chose when to search, and it chose by question type. So its two subsets are not two ways of answering the same questions — they are answers to different questions.

68.2% of Gemini’s grounded answers are attribute questions, against 24.9% of its ungrounded ones. Korkmaz being more visible in the ungrounded set may mean that searching pushes it down — or simply that Korkmaz is a name that comes up when someone asks broadly which brands are good, and broad questions are the ones Gemini did not search. Both explanations fit the data equally well, and nothing in this study separates them.

We report the difference and decline the causal claim. Identifying a retrieval effect would require running the same questions with search forced on and forced off in the same model, which is a design change for a future edition rather than something recoverable from this one.

7. Question-Type Ownership

One-sentence takeaway: Karaca leads all three question types, but the field behind it thins dramatically — 38 entities qualify on discovery questions against 20 on attribute questions.

Reading. Karaca leads all three, most heavily on use-case questions (58.1%), where a shopper describes a situation — moving house, buying a trousseau — and assistants reach for a brand that covers a whole kitchen.

The more interesting number is how many entities qualify. Discovery questions produce 38 qualified entities; attribute and use-case produce 20 each. When the question is open (“which kitchen brands are actually good”), assistants spread their answers across a long list. When the question specifies a property — dishwasher-safe, stays sharp, does not warp — the field narrows sharply to the names associated with that property. Attribute questions are also where foreign cookware performs best: Tefal is third overall on attribute questions at 24.0%, and second by score.

Pirge is the position leader on both discovery and attribute questions, despite qualifying in neither the general cut nor the use-case cut. A specialist with a narrow, strongly associated category is named first when named at all.

8. Open Market: Turkish vs. Foreign

One-sentence takeaway: On origin-neutral questions, mentions split 63.6% Turkish to 36.4% foreign — but more distinct foreign entities appear than Turkish ones, so foreign visibility is broad and shallow rather than absent.

This chapter relies only on the 23 open questions with no origin restriction (575 responses), the only fair basis for comparing Turkish and foreign brands. The 22 local questions cannot surface foreign brands by construction, and mixing them would manufacture a Turkish majority.

8.1 Origin share

Counting every entity that appeared at least once, the open cut splits 2,355 Turkish mentions to 1,348 foreign — 63.6% to 36.4%.

Three other framings of the same question, stated together so that none can be quoted alone:

The last row is the finding. More distinct foreign entities appear in the open cut than Turkish ones — 58 against 54 — yet they collect little more than a third of the mentions. Turkish visibility is concentrated in a handful of names that appear constantly; foreign visibility is spread across a long tail of European cookware and porcelain houses that each surface occasionally. Restricting the count to qualified entities removes most of that tail and moves the split to 68.3% / 31.7%, which is why both figures are published here.

For contrast, the general cut — which includes the 22 questions that demand Turkish brands — yields 17 Turkish qualifiers to 6 foreign. That figure describes the study’s question design as much as the market, and should never be quoted as a market share.

8.2 Open-market top 15

8.3 The pattern: a Turkish core, a European periphery

Karaca leads the open cut at 97.87 on a 67.7% mention rate — higher than its general-cut rate, because origin-neutral questions are where it is most dominant. Tefal is second at 70.26, the highest a foreign entity places in any general-population cut in this study, and IKEA is sixth.

The foreign names that qualify follow one pattern almost without exception: they are cookware, knives or storage — Tefal, IKEA, WMF, Zwilling, Fissler, Le Creuset, Lock&Lock, Tupperware. In tableware and serving, where Turkish porcelain manufacturers have deep domestic reputations, foreign brands are nearly absent. So when a Turkish shopper asks an origin-neutral kitchenware question, the answer is Turkish-led overall, but genuinely mixed as soon as the subject turns to pans and knives.

One brand sits exactly on the boundary and is worth naming for what it shows about thresholds: Villeroy & Boch appears in 4.98% of general-cut responses — one response short of the 5% bar. It is not ranked. A threshold is a line, and entities land on both sides of it by margins that carry no meaning.

9. One Owner, Three Leaders: Corporate Groups in the Data

One-sentence takeaway: Karaca, Emsan and Jumbo are one company covering 64.6% of all responses, and on origin-neutral questions a second group — Tefal and WMF’s owner — reaches 41.9%; the leaderboard shows brands, not owners.

A leaderboard ranks brands. Shoppers meet brands. But a reader drawing conclusions about competition from the table in Chapter 3 would be reading three Karaca Group brands as three independent competitors.

Karaca Group appears in 727 of 1,125 responses: 64.6%. That is a union, not a sum. Adding the three members’ mention counts would give 1,379 and double-count every answer that names two of them — which is precisely the error a reader makes scanning the leaderboard. On the open cut the group reaches 72.0%: nearly three in four origin-neutral kitchenware answers name Karaca, Emsan or Jumbo.

The second group is foreign and appears only when the question allows it. Groupe SEB — Tefal and WMF — covers 22.8% of the corpus overall, but that splits into 2.9% on local questions and 41.9% on open ones. Zwilling Group shows the same shape more sharply: 1.6% local against 27.8% open. Foreign group visibility in this market is not weak; it is conditional on the question not asking for Turkish brands.

Şişecam is the exception among the foreign-facing pattern — a Turkish group whose visibility is roughly even across local (12.4%) and open (14.8%) questions, because Paşabahçe is recognised as a Turkish glassware name in both contexts.

How to read this. Group concentration is not a claim about market power, pricing or competition — those are commercial questions this study does not measure. It is a caution about the leaderboard: three names in the top six sharing an owner means the visible shelf is narrower than a rank-ordered list makes it look. parentGroup is carried on every entity in the published data for anyone who wants to collapse them.

10. Where the Category Boundary Leaks

One-sentence takeaway: A quarter of all answers name a brand from outside kitchenware entirely — appliances, small appliances, cabinetry — and while whole-kitchen questions concentrate it, this is a property of the whole corpus, not six bad questions.

The Turkish word mutfak means both “kitchen” and, colloquially, “kitchenware.” Assistants resolve that ambiguity inconsistently, and the study measured how often.

25.8% of all responses name a brand from outside the category. On the six whole-kitchen questions it reaches 51.3%; on the 39 product-specific questions it is still 21.8%. This is not six ambiguous questions producing noise — it is a standing feature of how these systems answer Turkish kitchen queries, concentrated about 2.4× on the broadest ones.

Retrieval does not explain it. The least-grounded model, Gemini at 19.6%, has the lowest whole-kitchen drift (36.7%); the heavily-grounded Grok has the highest (66.7%). On product-specific questions the ordering reverses again, with Gemini highest at 33.3% and Perplexity lowest at 12.3%. Whatever produces category drift, it is not simply a matter of whether the model looked something up.

10.1 One word changes the answer

The spread across six near-identical questions runs from 80.0% to 12.0%, and the lowest one is the only question that names the category twice: “mutfak ve sofra ürünlerinde iyi yerli markalar neler.” Adding sofra — table, tableware — appears to anchor the assistants to kitchenware and cut drift by a factor of nearly seven against the bare-mutfak phrasing.

That observation rests on 25 responses per question and should be treated as a hypothesis rather than a result. But it points at something practical: the ambiguity lives in the prompt, not only in the model.

10.2 What the models name instead

The list is appliance makers: Arçelik, Bosch, Beko, Siemens, Vestel on major appliances; Arzum, Philips, Braun on small ones; Scavolini on cabinetry. Read charitably, this is not a failure of comprehension — a shopper furnishing a kitchen may well want a kettle and a dishwasher alongside their pans, and the assistant is answering the question a person might have meant. It is a scope boundary problem, and it is why the study keeps a separate Out-of-scope classification rather than silently dropping the names.

11. The Discovery Ecosystem: Where AI Learns About These Brands

One-sentence takeaway: 44.2% of citations point to brands’ own websites, karaca.com alone is cited by a third of all responses, and — unusually for this series — recipe sites and forums together supply nearly a fifth of the evidence.

When an assistant answers with web search, it draws on the content available to it at that moment. We collected the source addresses returned by the models, reduced them to domains, and ranked them by how many distinct responses cited each. Across the run, 937 responses carried at least one citation, referencing 479 distinct domains across 5,539 domain-response pairs.

11.1 The 12 most-cited domains

karaca.com is cited by 385 responses — 34.2% of everything collected — and by all five models. The gap to second place is large: trendyol.com at 213. Below them the table is a mix of brand sites (kutahyaporselen.com, schafer.com.tr, aryildiz.com, zwilling.com, tefal.com.tr, porland.com), two marketplaces, and — notably — three editorial and listicle sites, including a recipe site, nefisyemektarifleri.com.

The Models column is worth reading alongside the count. A domain cited by all five models is part of the shared evidence base; one cited by two is a single assistant’s habit. Among the top 12, six are cited by all five models — and all six are brand or marketplace domains, not editorial ones.

11.2 Source-type mix

11.3 Two things this mix shows

Brand-owned content is the largest single input, at 44.2%. Add retailer (8.1%) and marketplace (9.4%) domains and commercial sources account for 61.7% of citations. Read as a description rather than advice: in this market, a current, machine-readable brand site is closely associated with AI visibility.

Editorial and community content is a real presence here — 17.9% combined. Recipe sites, listicles and forums (nefisyemektarifleri.com, onedio.com, listelist.com, kadinlarkulubu.com, donanımhaber.com, ekşisözlük, YouTube, Reddit) supply nearly a fifth of the evidence base. Kitchenware is a category people write about publicly — reviews of pans that warp, dinner sets that survive the dishwasher — and the assistants read it.

A caveat on classification. Source types are derived from the entity dictionary for the 128 domains whose owner is known. The remaining 351 are typed by pattern, and the consistency audit can only assert correctness for the resolved subset. Treat the mix as accurate at the top of the distribution and indicative in its tail.

12. What the Patterns Suggest

One-sentence takeaway: Visibility in this market is associated with breadth across models, a strong own-site presence, and a clear category association — but the leaderboard’s shape is also determined by corporate structure and by one assistant’s answering style.

This chapter is a neutral reading of the data. It describes patterns associated with visibility; it is not advice, a service, or a recommendation, and visibility remains separate from quality.

1) Breadth is close to a precondition. 19 of the 23 qualified general-cut entities are seen by all five models. The four that are not — Fissler, Trendyol, Hepsiburada and Le Creuset — sit at #13, #20, #21 and #23. In a five-model breadth scale, missing one model costs 5 points of final score outright, and in practice it also signals that a brand’s presence in the underlying evidence is uneven.

2) Mention and position are independent signals. Emsan is mentioned in 35.6% of responses and ranks fifth; Porland is mentioned in 24.7% and ranks fourth, because it is named far earlier when named at all (MRR 0.497 against 0.329). At the extreme, Pirge would rank third in the general cut on score alone but does not qualify, and IKOO — mentioned in 7 responses with an MRR of 0.893 — would place near the top and is not ranked either. “Named often” and “named first” are different achievements with different causes.

3) Category association concentrates visibility. The entities that dominate a segment are the ones with a single, legible specialism: the porcelain houses in Tableware, the knife makers in Cutlery, Korkmaz in cookware and whole-kitchen questions. Generalists lead the general cut; specialists lead segments and own the position component.

4) The shelf is narrower than it looks. Three of the top six entities share an owner, and the two marketplaces in the table owe 96%+ of their visibility to one model. A reader counting independent options in the top ten would overcount.

5) Origin is contested only in cookware and cutlery. Turkish brands hold tableware almost completely — one foreign qualifier in 21 — while foreign brands reach #2 in Cutlery & Storage and second overall on the open cut. The local/open gap shows the assistants understand yerli as a constraint, so the open-cut balance is a genuine market signal rather than an artefact of question wording.

How a brand can locate itself in the data. Using the published dataset, a brand can read four things in order: whether it appears in all five models; how often it is mentioned within the cut that matches its category; how early it is named when mentioned; and which model or question type it is weakest in. These readings show where a visibility gap sits, without saying anything about whether its product is good.

13. Limitations and Notes

One-sentence takeaway: This is a single, point-in-time snapshot of one Turkish vertical, read across ten cuts whose scores are not comparable with each other.

• Single collection window. Everything here is a snapshot of 1 August 2026, 05:57–10:54 UTC. AI answers move; a re-run on another day will differ. This is the baseline edition; trend analysis becomes possible from the second.
• Never compare a scaled score across cuts or editions. 100 in one cut is a different absolute number from 100 in another, and the scale moves whenever the leader moves. Cross-cut comparison uses the raw layer — mention rate, MRR and breadth — which is published in every table.
• Turkish vs. foreign is only comparable on the open cut. 22 of the 45 questions explicitly constrain the answer to Turkish brands. The general cut’s 17-to-6 origin split describes the question design, not the market.
• Qualification is a threshold, and thresholds are arbitrary at the margin. Villeroy & Boch misses the general cut by one response. Pirge (4.00%) and IKOO (0.62%, MRR 0.893) would both rank highly on score and are excluded because ranking is restricted to qualified entities.
• Components above 100 are real, not errors. Unqualified entities are scored on the same cut maxima so the appendix stays on one scale; 21 rows (2.3%) exceed 100 on a component. They are flagged and never clipped.
• Narrow cuts qualify on very few responses. Cutlery & Storage qualifies at 12.5 responses and Whole-kitchen at 7.5, against 56.25 in the general cut. A 5% rate means something different in each.
• Whole-kitchen & Trousseau is indicative only — six questions, no attribute questions, and the heaviest off-market drift in the study.
• The search-vs-memory comparison is not identifiable. 96.3% of ungrounded responses are Gemini’s, and Gemini’s grounded and ungrounded subsets answer different question mixes (68.2% attribute versus 24.9%). Chapter 6 reports the difference and declines the causal claim. Separating retrieval from question type would require forcing search on and off within one model — a design change for a future edition.
• Marketplace visibility is model-specific. Trendyol and Hepsiburada owe 96.1% and 97.4% of their mentions to Grok. Their leaderboard positions describe one assistant more than the market.
• Sub-brands under-count structurally where a model names the parent instead.
• Source-type classification is partly heuristic. 128 of 479 domains resolve to an owner via the dictionary; the rest are typed by pattern, and the classification audit covers only the resolved subset.
• Precision over recall. Entity matching was cautious to avoid false positives, so brands mentioned indirectly may be undercounted. Two near-identical names, LAV (glassware) and Lava (cast iron), were confirmed as separate entities with zero response overlap and are deliberately not merged.
• Models are probabilistic. The same question can produce different answers; five repeats reduce but do not eliminate this.
• Visibility is not quality. This report measures mention and position only. It does not assess whether a brand was described positively, whether a recommendation was accurate, or whether a product is good. A score is not an endorsement.

Frequently Asked Questions

14. Appendix & Data

14.1 All questions (45), their segments and types

Distribution: 15 discovery, 15 attribute, 15 use_case; 22 local and 23 open. The 15 raw subcategories underneath the four segments are carried in the published dataset but are too granular to report on (median size: 2 questions).

14.2 All matched entities (139)

14.3 Data availability

The study decided 399 candidate entities, kept 144, and matched 139 in at least one response; 23 cleared the 5% threshold in the general cut, with 21 qualifying in the local cut and 27 in the open cut. To support scrutiny and reproduction, the underlying data is available on request: response-level entity mentions, all ten qualified leaderboards and their brand-only variants, full unfiltered metrics for every entity in every cut, the scale registry, per-model and per-question-type breakdowns, the corporate-group roll-up, the off-market drift analysis, the complete 479-domain citation list, and the methodology and data dictionary.

All figures reflect a single point-in-time run (1 August 2026) and are subject to the limitations in Chapter 13. This is the first (baseline) measurement; the study will be repeated quarterly.

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 3 August 2026 · Collection window 1 August 2026