750 responses, 5 models, 290 entities: İstikbal, Bellona and Doğtaş hold the top three on every cut, and the shelf behind them stays mostly domestic even when the question does not ask for Turkish brands

Vertical: Furniture, sofas, bedroom & home furnishing (mobilya) — Turkey
Method: A single point-in-time study of 750 responses — five large language models (Claude, Gemini, ChatGPT, Perplexity, Grok), each asked 30 Turkish-language questions five times
Collection date: 25 September 2026
Edition: 2 of a quarterly series. Edition 1 (June 2026) asked the same 30 questions; two of its five models were replaced for this edition.

Key terms: AI visibility, AI Visibility Score, furniture brands, Turkish furniture (mobilya) market, Turkish (yerli) furniture 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 furniture, and measures how often, in what order, and across how many models five large language models name each brand, retailer and marketplace. Entities are ranked separately within six cuts of the same questions — all questions, origin-constrained, origin-neutral, and three question types — 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

This report measures one thing: how often, how prominently, and across how many AI assistants a brand is named when people ask about furniture in Turkey. A high score means these systems currently have a lot of information available about a brand and surface it readily — not that the brand is better-built, longer-lasting or better value. 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. Readers and consumers 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: İstikbal, Bellona and Doğtaş hold the top three places on all six published cuts, and the shelf behind them is entirely domestic when the question asks for Turkish brands and mostly domestic when it does not — domestic entities take 100.00% of qualifying mentions on the 13 origin-constrained questions and 86.63% on the 17 origin-neutral ones (foreign 0.00% and 13.37%), with origin on file for every qualifying mention on both.

• İstikbal leads on score; Bellona is named most. İstikbal tops the general board at 96.10, named in 53.47% of all 750 responses (401) and by all five models. Bellona (94.64) is named more often, in 58.53% (439 of 750), and Doğtaş (89.65) in 56.93% (427 of 750). İstikbal leads because it is named earliest: its MRR of 0.5563 is the highest in the cut. Enza Home (79.10) and Kelebek (69.65) complete the top five.
• The same three hold the top three places on every cut. İstikbal, Bellona and Doğtaş occupy the first three places on all questions, on the origin-constrained and origin-neutral questions, and on each of the three question types. Bellona leads the origin-constrained cut (94.03) and the attribute (99.07) and use-case (92.30) cuts; İstikbal leads the other three.
• 23 of 290 tracked entities qualify on the general cut — 22 domestic and 1 foreign — with an average score of 55.59. The foreign qualifier is IKEA, sixth at 67.19. Four marketplaces — Masko (#13), Modoko (#16), Trendyol (#20) and Hepsiburada (#23) — rank alongside the brands, because they are what the assistants name.
• The question changes how domestic the shelf is, not who leads it. On the 13 origin-constrained questions (325 responses), domestic entities take 100.00% of qualifying mentions and foreign entities 0.00%, with origin on file for 1,564 of 1,564. On the 17 origin-neutral questions (425 responses), domestic entities take 86.63% and foreign entities 13.37%, with origin on file for 1,848 of 1,848; 4 of the 20 entities in that share are foreign, and IKEA is fourth on that board. Origin is assigned by brand home this edition and by a different rule in June, so no origin share here is compared with June’s.
• The assistants agree on the top, not on how long an answer is. Bellona is the most-named entity for ChatGPT 5.6 Luna Pro, Perplexity Sonar and Grok 4.3; Doğtaş for Claude Haiku 4.5; Enza Home for Gemini 3.6 Flash. Grok 4.3 names 10.62 entities per answer and Gemini 3.6 Flash 9.79, against 4.35 for Claude Haiku 4.5. Three qualifying ranks lean on one model: Grok 4.3 supplies 60 of Trendyol’s 91 general-cut mentions and 44 of Hepsiburada’s 60, and Gemini 3.6 Flash supplies 30 of Mudo Concept’s 40.
• Grounding was complete for four models and partial for one. 84.13% of all 750 responses were web-grounded. Four models grounded 100.00% of their 150 answers; Gemini 3.6 Flash grounded 20.67%, and the other 119 of its answers returned no grounding metadata at all — which is not evidence that it answered from memory.
• Against June, Enza Home rose most and Sivriler Mobilya fell most — on a panel that changed. On the same 30 questions, with June’s answers counted the way this edition counts and with Gemini 2.5 Flash-Lite and GPT-4o-mini replaced by Gemini 3.6 Flash and ChatGPT 5.6 Luna Pro, Enza Home moved from 22.12% of 746 responses to 41.60% of 750 (+19.48 points) and Sivriler Mobilya from 9.65% to 1.33% (-8.32). Enza Home is Gemini 3.6 Flash’s most-named entity and also sits in the top ten of all three models that answered in both editions (Chapter 8).
• The three leaders’ own sites lead the sources. bellona.com.tr and istikbal.com.tr are each cited in 46 of 150 question-model cells (30.67%), ahead of dogtas.com (41 cells, 27.33%). Claude Haiku 4.5 returned exactly 5 citations on every one of its answers, so its domain counts are capped and do not compare with the other models’.

Why it matters. Shoppers increasingly ask AI assistants which furniture brand to buy and where to buy it, and whether a brand appears in those answers is becoming a discovery channel in its own right. June’s edition set the baseline; this one repeats its 30 questions word for word, which makes a first reading of movement possible — with the confounds stated beside it, because two of the five assistants answering are not the ones that answered in June. The clearest finding is a structure rather than a movement: the same three brands sit at the top of every cut, and the shelf behind them stays mostly domestic whether or not the question asks for Turkish brands. A reminder: the numbers below describe visibility and information availability — not which brands are best.

2. Methodology

One-sentence takeaway: Five models answered 30 Turkish questions five times each, giving 750 responses across 150 question-model cells; 290 entities were scored within each cut, and since June two models, the counting rule and the origin rule have changed.

2.1 Scope and models

The study queried five large language models through their APIs with identical Turkish-language questions. Each of the 30 questions was put to each model five times, giving 150 question-model cells and 750 responses analyzed, all collected on 25 September 2026. The questions are in Turkish and models infer the market from them. The deployment’s system prompt and locale parameter were not recorded with this run, so this report makes no statement about either.

2.2 Web-search configuration

Every model was asked to search the web. Whether it did is a separate question, and the table below keeps the two apart: the first column is what the deployment asked of each model; the rest is what each model returned.

Overall, 84.13% of the 750 responses were web-grounded. Four models grounded 100.00% of their answers. Gemini 3.6 Flash grounded 20.67%: 31 answers report a search, and the other 119 carry no grounding metadata of any kind. An “unknown” answer is not a memory answer. The model was asked to search, and it is not known whether it did; the answer may rest on retrieval that was not reported, on the model’s own knowledge, or on both. This report never treats those 119 answers as answers from memory.

June’s edition carried a chapter comparing web-grounded answers with answers from model memory. This edition does not repeat it. Every model was asked to search, so there is no memory segment by design, and the only ungrounded answers are one model’s, of unknown provenance. Splitting the board by grounding would compare Gemini 3.6 Flash with the other four models, not search with memory.

A citation cap, not a measurement. Claude Haiku 4.5 returned exactly 5 citations on every one of its 150 answers. Its domain counts are truncated at that ceiling and do not compare with an uncapped model’s; Chapter 9 reads them that way.

2.3 Questions and cuts

All 30 questions are the ones asked in June, carried verbatim: none was added, none retired and none reworded. Each is typed by what it asks — discovery (a general “which brand?” or “where to buy?”), attribute (a specific property, such as stain-resistant fabric or durability) or use_case (a need, such as a home with a cat or a wedding furniture package) — and scoped by whether it constrains origin. A constrained question asks for Turkish brands (“türk malı mobilya markası arıyorum, hangileri iyi” — I’m looking for a Turkish-made furniture brand, which ones are good?); a neutral one does not (“en iyi mobilya markası hangisi” — which is the best furniture brand?).

Every metric is computed separately within each of six cuts:

June called the first three cuts Main, Local and Open. The all-questions cut is referred to in this report as the general cut.

June’s category chapter read the same three question types; this edition reads them in Chapter 6, each with its own leaderboard. No question in this study records a topical subcategory, and under this series’ rules a subcategory is published as a cut only when the study declares it before collection. This study did not, so no subcategory figure appears in this report. Chapter 7 returns to this.

2.4 Entity resolution

Every name in every answer is matched against Herm.io’s brand registry. A name that matches a registered brand or one of its aliases is counted directly and never becomes a candidate. Names that do not match are reviewed as candidates: 323 this period. Of those, 279 were resolved to a brand, 29 were given a verdict by a language model, and 15 were answered by a stored verdict from earlier work. These figures count the work, not the field, and do not reconcile to the 290 entities tracked.

Fifteen candidates were rejected as not a brand, and the most frequent of them show what the rules do.

A product name is not a brand. “Anka Koltuk Takımı”, “Luna Koltuk Takımı”, “Lacy Koltuk Takımı” and “Tommy Koltuk Takımı” (5 responses each) and “Viola Koltuk Takımı” (4) are sofa-set product names — koltuk takımı means sofa set — not brands, and none is counted.

Not every capitalized word is a name. “Selamlar” (5 responses) is a greeting — “regards” — not a name. It was extracted as a candidate and rejected.

One brand can be registered more than once. Kelebek, Yataş, Saloni and Zebrano had each been registered under more than one spelling, and the duplicates were merged before this edition was measured: “Kelebek” and “Kelebek Mobilya”, for example, are counted as one brand.

Resemblance is not identity. Among entities named in at least three answers, six near-name pairs were flagged — Mudo Concept and Mudo, Lider Vip and Lider, and QHome and Maq Home among them. None of the six shares a website on file, and each pair is tracked as two entities.

The result is 290 entities: 87 brands, 142 brands that are also retailers, 23 retailers, 15 marketplaces, 8 sub-brands, 4 publishers and 2 licensed brands; 9 have no entity type on file. Origin is on file as domestic for 218 and foreign for 55; 17 have no origin on file. One entity, Trafton, is provisional: research never resolved it, and it may duplicate a brand under another spelling. No entity has a product category on file, so scope could not be resolved for any of them, and all 290 are eligible to rank.

2.5 AI Visibility Score (0–100)

Score = 0.45 × Mention + 0.3 × Position + 0.25 × Breadth

Mention is the share of the cut’s responses that name the entity. Position is the mean reciprocal rank (MRR) of the entity’s first appearance in each answer that names it — first is 1, second 0.5, and so on. Breadth is the number of models, of five, that name it. Each component is scaled to the highest value among qualified entities in that cut and is never clipped. An entity qualifies for ranking on a cut if it is named in at least 5% of that cut’s responses; there is no absolute mention floor.

Worked example — İstikbal, general cut. İstikbal is named in 401 of 750 responses, a mention rate of 53.47%. The highest mention rate among qualified entities in the cut is Bellona’s, 58.53%, so İstikbal’s mention component is 91.34. Its MRR of 0.5563 is the cut maximum, so its position component is 100.00. It is named by 5 of 5 models, so breadth is 100.00. The composite is 0.45 × 91.34 + 0.3 × 100.00 + 0.25 × 100.00 = 96.10. The mention component is computed from the unrounded inputs (0.534666667 against 0.585333333); dividing the two-decimal rates printed on this page gives a slightly different number.

İstikbal scores 96.10 rather than 100 because the highest mention rate in the general cut belongs to Bellona, not to the leader. Because the maxima come from qualified entities only, an unqualified entity can exceed 100 on a component: 98 of 1,103 entity-cut rows do (8.90%), all of them unqualified — 10.40% of the 945 unqualified rows. A value above 100 says the entity beats the cut leader on that one component while being named too rarely to rank.

A score belongs to one cut’s scale. The scale moves whenever the leader moves, so an AI Visibility Score is never compared across cuts, across editions or across verticals. This report shows no score change since June. Where it compares editions (Chapter 8), it uses raw mention rate on the general cut only. The maxima behind every table are published in the report’s data file.

2.6 What changed since June, and what is comparable

Three changes matter for any comparison with June. The panel: three models are carried and two were replaced, and nothing in this report corrects for that. The counting rule: June’s page counted any answer whose text contained a name; this edition counts a brand or company only where it is named in the answer, not where it appears only in a web address. Under June’s rule a marketplace was counted wherever its name appeared, including as a place to buy, so marketplaces read higher on June’s page than in June’s answers counted the way this edition counts. The brand registry has also changed since June — duplicate registrations merged, brands renamed — so a June level here can differ from June’s page for any entity, not only marketplaces. So that both sides of every comparison count a brand the same way, June’s levels in this report are June’s answers, counted the way this edition counts — not the figures June’s page printed. The origin rule: see 2.7.

2.7 Origin, neutrality and self-exclusion

Origin is the brand home: the country the brand itself is operated from, which is not always its owner’s. Fendi Casa is Italian though LVMH is French; Cetaphil is American though Galderma is Swiss. Where no separate brand home can be established, the parent’s headquarters stands in. Local-market naming variants are folded into their parent before origin is read, so a brand named after its market subsidiary is not recorded as domestic. Tables show ISO country codes; shares use a domestic/foreign rollup. Entities with no origin on file stay in every table, are excluded from every origin share, and are counted — never defaulted.

June used a different rule. June’s edition counted a brand as Turkish only if it was founded and headquartered in Turkey; this edition assigns every origin by brand home. Within this edition the rule is published here beside every origin figure, which is what makes a domestic-versus-foreign comparison safe; across the change it is not, and no origin share in this report is compared with June’s. June also published a different measure on its origin-neutral questions — the share of answers naming at least one Turkish brand. That is not a share of qualifying mentions, and it is not set beside the shares in this report.

Marketplaces and retailers rank in every leaderboard, because they are what the assistants actually name. The brand-only view in section 3.3 removes the four qualifying marketplaces — Hepsiburada, Masko, Modoko and Trendyol — by name and changes nothing else. Retailers, IKEA and Koçtaş among the qualifiers, are retained everywhere, as are brands that also run their own stores.

Origin shares are read on a brand-only, qualified basis. The headline share is computed over qualified entities with origin on file, with marketplaces and publishers left out — a marketplace is where a brand is bought and a publisher is where it is written about — and own-label retailers kept in. It is always printed with its coverage: the qualifying mentions with origin on file. The same arithmetic over every tracked entity, qualified or not, is published as a sanity check; it describes the long tail, not the shelf a reader sees. This basis applies to shares only; it does not change who ranks.

Neutrality and self-exclusion. This is a market-wide study with no focus brand. Herm.io’s own domain is excluded from the source data before analysis, so the company’s own content neither appears among the cited domains nor influences any figure.

3. Overall Visibility Leaderboard

One-sentence takeaway: Twenty-three of 290 tracked entities qualify on the general cut; İstikbal leads on score because it is named earliest, Bellona is named most often, and IKEA is the only foreign qualifier.

The list below ranks the 23 entities named in at least 5% of all 750 responses, by AI Visibility Score. As a reminder, it ranks visibility, not quality.

Average score of the 23 qualified entities: 55.59. Twenty-two are domestic and one is foreign.

3.1 Tier narrative

The top three (96.10, 94.64, 89.65). İstikbal, Bellona and Doğtaş are each named in more than half of all 750 responses — 53.47% (401), 58.53% (439) and 56.93% (427) — and by all five models. What separates them is position. Bellona is named most often, but İstikbal is named earliest, with the cut’s highest MRR, 0.5563, against Bellona’s 0.4569 and Doğtaş’s 0.3871. “Named often” and “named early” pull in different directions at the very top of this board.

Enza Home and Kelebek (79.10, 69.65). The two are named in almost exactly the same share of responses — 41.60% (312 of 750) and 41.47% (311 of 750) — and both by all five models. Position separates them: Enza Home’s MRR is 0.4100, Kelebek’s 0.2368.

Two owners appear twice on this board. Kelebek and Doğtaş are both Doğanlar Mobilya Grubu brands, and Enza Home is a Yataş Group brand, as is Yataş (#12). This report ranks brands, not companies, and no figure in it adds two brands together.

IKEA, the one foreign qualifier (67.19). IKEA, a Swedish retailer, is sixth, named in 24.13% of responses (181 of 750) with an MRR of 0.4382. It is the only entity on the board whose origin is not Turkish.

The middle of the board (59.08–46.12). Lazzoni (59.08, 21.33%) leads a band that runs to Saloni Mobilya (46.12). Inside it, rank follows position as much as frequency. Gündoğdu Mobilya ranks tenth on an 8.13% mention rate (61 of 750) with an MRR of 0.3877; Vivense, named in 22.40% (168 of 750), ranks eleventh at 52.12 with an MRR of 0.1836. The two sit a few hundredths of a point apart, and gaps of that size here carry no meaning. Saloni Mobilya qualifies on 5.20% (39 of 750), just above the threshold, and reaches 15th on an MRR of 0.3175. Masko, a marketplace, is 13th at 47.80.

The tail of the board (45.15–33.22). Modoko, Koçtaş, Koleksiyon, Alfemo, Trendyol, Loda, Mudo Concept and Hepsiburada. The marketplaces show the board’s sharpest split between frequency and position: Trendyol is named in 12.13% of responses (91 of 750) and Hepsiburada in 8.00% (60 of 750), with MRRs of 0.1892 and 0.1311 — when they appear, they appear late. Every qualifier except Trendyol, Mudo Concept and Hepsiburada is named by all five models; those three are named by four, and Chapter 5 shows that each leans on a single model. The remaining tracked entities sit below the threshold, and the appendix lists them all.

3.2 The same board by mention rate

Scores reward prominence as well as frequency, so the ranking by raw mention rate is a different list, and the difference is informative.

The first six are the same entities on both lists, in a different order at the top: Bellona and Doğtaş are first and second by volume, İstikbal third. Enza Home, Kelebek and IKEA hold the same place either way. Vivense and Mondi lose most to position — Vivense is 7th by volume and 11th by score, Mondi 10th and 14th — and Yataş is 9th and 12th. Modalife and Çilek go the other way, 11th and 12th by volume but 8th and 9th by score, because they are named earlier than their frequency would place them.

3.3 Brand-only view

Removing the four qualifying marketplaces — Hepsiburada, Masko, Modoko and Trendyol, named explicitly, with nothing else changed — leaves 19 entities. Scores are unchanged, because the scale is still the general cut’s; entities ranked below a removed marketplace move up. IKEA and Koçtaş stay: they are retailers, not marketplaces.

4. Two Shelves: Origin-Constrained and Origin-Neutral Questions

One-sentence takeaway: On the 13 origin-constrained questions domestic entities take 100.00% of qualifying mentions and foreign entities 0.00%, and on the 17 origin-neutral questions they take 86.63% and 13.37% — with origin on file for all 1,564 and all 1,848 qualifying mentions respectively.

The 30 questions divide into 13 that ask for Turkish brands and 17 that do not. Each set has its own leaderboard, its own denominator and its own scale. Neither is the market on its own, so the two figures are always quoted together, each with the number of questions behind it.

4.1 Origin share on both cuts

Origin-constrained (13 questions, 325 responses): domestic entities take 100.00% of qualifying mentions and foreign entities 0.00%, with origin on file for 1,564 of 1,564 qualifying mentions. All 24 qualified entities in the share are domestic.

Origin-neutral (17 questions, 425 responses): domestic entities take 86.63% and foreign entities 13.37%, with origin on file for 1,848 of 1,848. Of the 20 qualified entities in the share, 16 are domestic and 4 foreign.

The assistants read “yerli” as a constraint and answer it with domestic names only. Without it, foreign entities enter the shelf, but domestic entities still take most of the qualifying mentions, because the three most-named entities on the neutral shelf — Doğtaş, Bellona and İstikbal — are domestic.

The sanity check points the same way. Across every tracked entity, qualified or not, domestic entities take 99.52% and foreign entities 0.48% of mentions on the constrained questions (origin on file for 2,098 of 2,128), and 81.34% and 18.66% on the neutral questions (2,840 of 2,870). On both, the long tail is more foreign than the shelf: a constrained question narrows what qualifies, not everything that is named.

Left out of every share, and still ranked. On the constrained cut, five qualifying marketplaces are left out of the share: Trendyol (46 mentions), Masko (35), Hepsiburada (32), Modoko (32) and Mobiliyum AVM (22). On the neutral cut, four are: Trendyol (45), Modoko (35), Masko (30) and Hepsiburada (28). Across all tracked entities, 17 marketplaces and publishers with 213 mentions are left out on the constrained cut, and 14 with 161 on the neutral cut. Every qualifying marketplace is domestic, so including them would raise the domestic share for a reason that has nothing to do with brands. They still rank in every leaderboard.

4.2 The origin-constrained shelf

29 entities qualify on the 13 origin-constrained questions (325 responses). Twenty-four enter the origin share, all domestic, and the five marketplaces are left out of it.

Bellona leads at 94.03, named in 67.08% of these responses (218 of 325). İstikbal is second at 94.01 on 58.15% (189 of 325), named less often but earlier: its MRR is 0.5412 against Bellona’s 0.4335. Doğtaş is third at 85.28 on 59.38% (193 of 325).

Below them, this shelf brings smaller domestic makers forward on position. Kilim Mobilya ranks eighth on a 7.69% mention rate (25 of 325) with an MRR of 0.5047; Adore Mobilya tenth on 5.23% (17 of 325, MRR 0.5165); Weltew Home eleventh on 6.77% (22 of 325, MRR 0.4941). All three are named by 4 of 5 models. Gündoğdu Mobilya, tenth overall, is sixth here on 13.54% (44 of 325). Ranks built mainly on position and thin mentions should be read cautiously.

No foreign entity qualifies on this cut. Foreign entities are named on the constrained questions — they take 0.48% of all-entity mentions with origin on file — but none often enough to reach the 5% threshold.

Masko is the highest-placed marketplace on this shelf, seventh at 56.80 on 10.77% (35 of 325). Mobiliyum AVM, a marketplace that does not qualify on the general cut, is 17th here on 6.77% (22 of 325).

4.3 The origin-neutral shelf

İstikbal leads this shelf at 95.77, named in 49.88% of these responses (212 of 425), with Bellona (92.78, 52.00%) and Doğtaş (91.18, 55.06%) behind it. Doğtaş is the most-named entity on this shelf but the latest-named of the three, with an MRR of 0.4022 against İstikbal’s 0.5697.

IKEA is fourth at 81.26, named in 40.94% of these responses (174 of 425) with an MRR of 0.4329 — ahead of Enza Home (80.42, 39.76%) and Kelebek (68.50, 38.12%). On the questions that do not ask for Turkish brands, the Swedish retailer is the one foreign name among the leading group.

The other three foreign qualifiers sit near the threshold. Minotti, Italian, is 13th on 5.18% (22 of 425) with an MRR of 0.4092, named by 4 of 5 models; B&B Italia, also Italian, is 15th on 5.41% (23 of 425). Roche Bobois, French, is lower on the board. Trendyol (10.59%) and Modoko (8.24%) are the highest-placed marketplaces, 16th and 18th; Mudo Concept is 20th on 8.00% (34 of 425).

4.4 What this chapter does not compare

No origin share here is compared with June’s, because June assigned origin by a different rule (section 2.7), and June’s origin-neutral figure measured something different. The constrained figure is never read on its own: 100.00% domestic and 0.00% foreign on 13 questions that ask for Turkish brands describes the question as much as the market, and it stands beside 86.63% domestic and 13.37% foreign on the 17 that do not. And the question-type cuts in Chapter 6 each blend constrained and neutral questions, so their counts of domestic qualifiers are not origin measurements.

5. Differences Between Models

One-sentence takeaway: İstikbal, Bellona and Doğtaş are in the top five of every model, but Grok 4.3 names 10.62 entities per answer to Claude Haiku 4.5’s 4.35, and three qualifying ranks — Trendyol, Hepsiburada and Mudo Concept — rest mainly on one model.

5.1 Per-model behavior

The leaders are shared; their order is not. Bellona is the most-named entity for Grok 4.3 (84.67% of its 150 answers), ChatGPT 5.6 Luna Pro (72.00%) and Perplexity Sonar (46.00%). Claude Haiku 4.5 names Doğtaş most (33.33%) and Gemini 3.6 Flash names Enza Home most (74.00%). İstikbal, Bellona and Doğtaş are the top three for four of the five models and sit in the top five of all five.

Answer length is the largest difference. Grok 4.3 names 10.62 entities per answer and reaches the widest repertoire, 170 distinct entities. Gemini 3.6 Flash names 9.79 per answer (114 distinct) and Perplexity Sonar 5.75 (132). ChatGPT 5.6 Luna Pro (5.30, 99 distinct) and Claude Haiku 4.5 (4.35, 126 distinct) give short, selective answers; Claude names fewer entities per answer than ChatGPT but more distinct ones across its 150 answers. For an entity at the edge of visibility, being named at all depends heavily on which assistant answers.

5.2 Each model’s most-named entities

The models agree on the core: Doğtaş, Bellona, İstikbal, Kelebek and Enza Home are in every model’s top ten. They differ beneath it. Grok 4.3 has two marketplaces in its top ten — Trendyol seventh (40.00%) and Hepsiburada ninth (29.33%) — and Modalife sixth (42.67%). Gemini 3.6 Flash has Vivense sixth (51.33%) and Lazzoni seventh (48.00%). Claude Haiku 4.5 places IKEA fourth (19.33%), higher than any other model does, and has the marketplace Modoko eighth (8.00%). IKEA is in the top ten of four models; Grok 4.3’s is the one that leaves it out.

5.3 One model’s habit

Three qualifying entities on the general cut draw at least three in five of their mentions from a single model:

Two marketplace ranks are largely Grok’s. Grok 4.3 accounts for 60 of Trendyol’s 91 general-cut mentions (65.93%) and 44 of Hepsiburada’s 60 (73.33%), naming them in 40.00% and 29.33% of its 150 answers. The two ranks — Trendyol 20th and Hepsiburada 23rd — describe one assistant’s habit of pointing to where to buy more than a consensus across the panel. Masko and Modoko, the other two marketplaces on the general board, draw less than 60% of their mentions from any one model.

Mudo Concept’s rank is largely Gemini’s. Gemini 3.6 Flash supplies 30 of Mudo Concept’s 40 general-cut mentions (75.00%), naming it in 20.00% of its 150 answers. Gemini 3.6 Flash joined the panel this edition, so this is also a reminder of how a panel change can move a single entity.

5.4 What only one model names

Every model names entities no other model names at all: 41 for Grok 4.3, 26 for Claude Haiku 4.5, 20 for Gemini 3.6 Flash, 16 for Perplexity Sonar and 15 for ChatGPT 5.6 Luna Pro.

6. Question-Type Ownership

One-sentence takeaway: İstikbal, Bellona and Doğtaş take the top three places on all three question types; discovery questions qualify 39 entities against 20 for attribute and 23 for use-case questions, and the specialists below the leaders change with the type.

Discovery — “which brand?” or “where to buy?” — spreads visibility widest: 151 distinct entities, 39 qualifiers and the lowest average score of the three, 45.81. İstikbal leads at 95.11 with the cut’s highest MRR, 0.6279. Doğtaş is named most, in 70.00% of these responses (175 of 250), the highest rate any entity reaches on any question type. The top five are all domestic: İstikbal, Doğtaş, Bellona, Enza Home and Kelebek. Hepsiburada qualifies on this cut on mentions from two models only (19 mentions).

Attribute — a specific property — narrows the field to 20 qualifiers from 118 distinct entities, with the highest average score, 55.12. Bellona leads at 99.07, named in 56.80% of these responses (142 of 250) and holding the cut’s highest MRR, 0.5541; Doğtaş is named most, at 58.00% (145 of 250). IKEA is fifth at 71.81 on 23.60% (59 of 250), with an MRR of 0.5264. Eymense qualifies here on two models only (17 mentions).

Use case — a need — qualifies 23 from 154 distinct entities, with an average score of 51.37. Bellona leads at 92.30 and is also named most, at 50.80% (127 of 250). Çilek is fifth at 67.40 on a 14.00% mention rate (35 of 250), carried by an MRR of 0.6531, the highest of any question type. IKEA is fourth at 67.98 on 22.00% (55 of 250), which makes it the only foreign entity in any of the three top fives. Modoko qualifies here on two models only (13 mentions).

Each type rests on the same number of responses, 250. The differences between them are not origin findings: discovery splits 5 origin-neutral and 5 origin-constrained questions, attribute and use case 6 and 4, and Chapter 4 is where origin is measured.

7. Attribute Axes

One-sentence takeaway: No attribute axis can be published this edition: no entity has a product category or audience on file, and no subcategory was declared before collection, which this series requires before publishing one as a cut.

An attribute chapter ranks entities by something they are — the product categories they make, or the audience they serve — rather than by the questions asked. This edition’s data carries no such attribute: none of the 290 tracked entities has a product category on file, and no audience attribute is recorded. A product-shelf chapter cannot be written from this export, and this report does not substitute an estimate for one.

No question in this study records a subcategory either, and under this series’ rules a subcategory is published as a cut only when the study declares it before collection. This study did not. June’s “Category Ownership” chapter read the three question types rather than product categories; that reading continues, with full leaderboards, in Chapter 6.

8. What Changed Since June

One-sentence takeaway: On the same 30 questions, against June’s answers counted the way this edition counts, 90 of the 199 entities named in both editions rose and 102 fell — Enza Home gained 19.48 points and Sivriler Mobilya lost 8.32 — with two of the five models replaced between editions, which no figure here corrects for.

8.1 How this comparison is built

This chapter compares raw mention rate on the general cut and nothing else; AI Visibility Scores do not travel between editions. All 30 questions were asked in both editions, so the change on the shared questions is the plain difference between the two levels: June’s 746 responses against September’s 750.

June’s levels are recounts. Every June figure in this chapter is June’s answers, counted the way this edition counts — re-read with this edition’s brand registry so both sides count a brand the same way. They are not the figures June’s page printed. June’s page counted a name wherever it appeared in the answer text, including as a place to buy, so marketplaces read higher on June’s page than in this recount; and because the registry has changed since June, other entities can differ too. No figure from June’s page is quoted in this chapter.

The panel is the confound, and it is declared, not corrected. Claude Haiku 4.5, Perplexity Sonar and Grok 4.3 answered in both editions. Gemini 2.5 Flash-Lite and GPT-4o-mini answered in June only; Gemini 3.6 Flash and ChatGPT 5.6 Luna Pro answered in September only. Rebasing adjusts for the question set, which did not change; nothing adjusts for the panel. Every number below carries that confound: the direction of a large move is better supported than its exact size.

Origin in this chapter is this edition’s coding, applied to both editions. June’s own origin rule is not re-read, so the domestic and foreign counts below are one rule applied twice, not an origin share compared across the rule change.

Of the 199 entities named in both editions, 90 rose, 102 fell and 7 were unchanged. Domestic entities account for 153 of the 199, of which 66 rose; foreign entities for 36, of which 21 rose; 10 have no origin on file and are neither. A further 91 entities were named only in September’s answers and 65 only in June’s. Those entities are not in the tables below, which rank only entities named in both editions.

8.2 The movers

Gains. Enza Home moved from 22.12% of June’s recounted answers (165 of 746) to 41.60% (312 of 750), +19.48 points. Kelebek gained 10.10 points, from 31.37% to 41.47%, and Lazzoni 8.87, from 12.47% to 21.33%. All three leaders rose as well: Doğtaş +8.54, Bellona +8.00 and İstikbal +6.68. Koleksiyon moved from 2.55% (19 of 746) to 8.80% (66 of 750), +6.25, and qualifies on the general cut this edition. IKEA (+5.23, from 18.90% to 24.13%) is the only foreign entity among the ten largest gains.

Falls. Sivriler Mobilya fell 8.32 points, from 9.65% (72 of 746) to 1.33% (10 of 750), and sits below the qualification threshold this edition. Trendyol fell 7.17 points, from 19.30% to 12.13%, measured against June’s answers counted the way this edition counts rather than against June’s page. Gündoğdu Mobilya (-5.81, from 13.94% to 8.13%) and Koçtaş (-5.67, from 12.20% to 6.53%) fell and still qualify. İnegöl Mobilya, a registered retailer of that name, is one of three entities — with Almila Mobilya and Konfor Mobilya — that rest on too few mentions this edition (7, 7 and 3) to read as more than direction.

8.3 Where the gains sit

This report publishes no shared-model comparison and carries no per-model June figures, so it cannot say how much of any gain the panel change accounts for. What it can show is where each mover is named now.

Enza Home’s current visibility does not sit only in the models that joined. It is Gemini 3.6 Flash’s most-named entity (74.00%) and fifth for ChatGPT 5.6 Luna Pro (38.00%), but it is also fifth for each of the three carried models — Claude Haiku 4.5 (17.33%), Perplexity Sonar (20.67%) and Grok 4.3 (58.00%). Kelebek’s profile is similar: third for Gemini 3.6 Flash (70.67%) and fourth for ChatGPT 5.6 Luna Pro (40.00%), and in the top ten of all three carried models. Whether either rose within the carried models cannot be read from this report.

Lazzoni’s profile leans on the joining models. It is seventh in Gemini 3.6 Flash’s answers (48.00%) and ninth in ChatGPT 5.6 Luna Pro’s (10.00%); of the three carried models, only Perplexity Sonar has it in its top ten (sixth, 16.67%). Yataş (+7.41) is in the top ten of both joining models and of one carried model, Grok 4.3 (28.00%). Koleksiyon is in no model’s top ten. For these entities the panel change cannot be ruled out as a large part of the gain.

9. The Discovery Ecosystem: Where AI Learns About Brands

One-sentence takeaway: The three leaders’ own sites are the most widely cited of 226 domains — bellona.com.tr and istikbal.com.tr each in 46 of 150 question-model cells — and one model’s citations are capped while another’s grounding is partial, so the map is uneven by model.

When an assistant answers with web search, it draws on whatever is available to it at that moment. The cited addresses were reduced to domains and counted by question-model cell — one cell is one question put to one model, holding that question’s repeats — never by response. Herm.io’s own domain was removed before any of these counts. Three units appear in this chapter, and they are not interchangeable:

• Citations recorded — 6,269. The citation entries the models returned.
• Link rows — 2,288. Attribution rows, one per model, entity, domain and attribution rule; one mention can be attributed to several domains.
• Mentions attributed — 6,409. Those rows’ mention counts, summed. Because one mention can sit on several rows, this is not a count of distinct mentions and does not compare with any mention rate’s numerator.

Across all answers, 226 domains were cited, 126 of them in a single cell only.

9.1 The most-cited domains

bellona.com.tr and istikbal.com.tr are each cited in 46 of the 150 cells (30.67%), followed by dogtas.com (41 cells, 27.33%), kelebek.com (25, 16.67%) and enzahome.com.tr (23, 15.33%) — the own sites of the general board’s top five. Fourteen of the fifteen domains have an owner among the tracked entities; trendyol.com is the one marketplace among them. mobilyahaber.com.tr is the outlier: no tracked entity owns it, and it is cited in only 14 cells (9.33%), yet it carries 629 attributed mentions, the most of any domain in the table — a reminder that attribution rows count every entity a cited page is credited with.

9.2 Entities’ own sites

142 of the 226 domains are the own site of some tracked entity. Tracked is not ranked: marketplaces, publishers and entities that qualify in no cut have sites too, so this count does not reconcile with the leaderboard, and it should not be read as brands citing themselves. Rows on those domains account for 2,278 of the 6,409 attributed mentions — attribution-row units, not distinct mentions.

That figure is understated. Three cited domains look like a tracked entity’s own site but are not attributed to it, because the registry holds a different website: cezbemobilya.com (Cezbe Mobilya; registry cezbemobilya.com.tr), cilek.live (Çilek; registry cilek.com) and ikea.com.tr (IKEA; registry ikea.com). They are therefore not counted among own sites.

In 33 rows, the rule that assigns a mention to the entity’s own domain and the rule that assigns it by position in the answer disagreed about which citation the mention belongs to. Those conflicts are surfaced in the data, not resolved.

9.3 Why the map is uneven

The models do not contribute equally. Claude Haiku 4.5 returned exactly 5 citations on every one of its 150 answers — 750 in all — so its domain counts stop at a ceiling and do not compare with an uncapped model’s. Perplexity Sonar alone returned 3,730 of the 6,269 citations. Gemini 3.6 Flash returned 256 and grounded 20.67% of its answers. A domain’s place in this table therefore reflects which models cite at length as much as what any one model reads.

This edition carries no source-type classification of domains and no per-response citation counts, so June’s source-type mix has no counterpart here. Nothing in this chapter tests whether being cited causes visibility or merely accompanies it.

10. What the Patterns Suggest

One-sentence takeaway: Six neutral readings, not recommendations: the same three lead on every cut, the shelf stays mostly domestic on both question scopes, frequency and position are separate routes onto the board, where-to-buy entities share the shelf partly through one model’s habit, answer format decides who is seen in the long tail, and the leaders’ own sites are the most cited.

This chapter describes patterns in the data. It is not advice, and visibility remains separate from quality.

The same three lead on every cut. İstikbal leads the general cut (96.10), the origin-neutral cut (95.77) and the discovery questions (95.11); Bellona leads the origin-constrained cut (94.03) and the attribute (99.07) and use-case (92.30) questions; and Doğtaş is in the top three of all six. All three are in the top five of every model. No other finding in this report is as stable across the ways the data can be cut.

The shelf stays mostly domestic on both question scopes. On the 13 origin-constrained questions domestic entities take 100.00% of qualifying mentions and foreign entities 0.00%; on the 17 origin-neutral questions, 86.63% and 13.37% — with origin on file for every qualifying mention on both. The assistants read “yerli” as a constraint; without it, the foreign presence on the shelf is IKEA in fourth place and three other foreign qualifiers lower on the board.

Frequency and position are separate routes onto the board. Bellona is named most often on the general cut (58.53%) and ranks second; İstikbal is named less often (53.47%) and ranks first because it is named earliest. Gündoğdu Mobilya ranks tenth on 8.13% with an MRR of 0.3877, one place above Vivense on 22.40% with an MRR of 0.1836. On the use-case questions, Çilek reaches fifth on 14.00% with an MRR of 0.6531.

Where-to-buy entities share the shelf, partly through one model. Four marketplaces and two retailers, IKEA and Koçtaş, qualify on the general cut alongside the brands. Grok 4.3 supplies 60 of Trendyol’s 91 general-cut mentions and 44 of Hepsiburada’s 60. A marketplace in an answer usually means where to buy, and for these two it mostly means one assistant’s answering style.

Answer format decides who is seen in the long tail. Grok 4.3 names 10.62 entities per answer, reaches 170 distinct entities and names 41 that no other model does; Claude Haiku 4.5 names 4.35 per answer, 126 distinct and 26 alone. For an entity at the edge of the board, whether it is seen depends on the assistant’s format as much as on the entity.

The leaders’ own sites are the most cited. bellona.com.tr, istikbal.com.tr and dogtas.com are the three most widely cited domains, in 46, 46 and 41 of 150 cells. This is an observed association between being cited and being named, not evidence that one causes the other.

11. Limitations and Notes

One-sentence takeaway: This is one market on one collection day, measured on a panel, a counting rule and an origin rule that all changed since June, with scope unresolved for every entity; read every figure with the notes below.

• The panel changed. Gemini 2.5 Flash-Lite and GPT-4o-mini were replaced by Gemini 3.6 Flash and ChatGPT 5.6 Luna Pro. Chapter 8 declares this confound beside every figure; it does not correct for it, and some movers, Lazzoni among them, are named now mainly by the models that joined.
• The counting rule changed. June’s page counted a name anywhere in the answer text; this edition counts a brand or company only where it is named in the answer, not where it appears only in a web address. June’s levels in this report are therefore recounts of June’s answers, not the figures June printed, and marketplaces read higher on June’s page than in the recount.
• The registry changed. Duplicate registrations were merged before this edition was measured — Kelebek, Yataş, Saloni and Zebrano had each been registered under more than one spelling — and some brands were renamed. The recount of June’s answers uses this edition’s registry.
• The origin rule changed. June required a brand to be founded and headquartered in Turkey; this edition uses brand home. No origin share is compared across the change.
• Origin classification involves judgment. Brand home is a researched decision, and brands run from one country under an owner in another are genuinely arguable cases. 17 entities have no origin on file; they appear in every table and in no share.
• Scope was not resolved. No entity has a product category on file, so the study could not separate core furniture entities from adjacent ones, and all 290 are eligible to rank. An entity whose main business lies outside furniture can rank if the assistants name it often enough.
• Untyped and provisional entities. Nine entities have no entity type on file, and one, Trafton, is provisional and may duplicate a brand under another spelling. None of the ten qualifies in any cut.
• One model’s answers have unknown grounding. 119 of Gemini 3.6 Flash’s 150 answers returned no grounding metadata. They are not memory answers, and nothing in this report treats them as such.
• One model’s citations are capped. Claude Haiku 4.5 returned exactly 5 citations per answer, so its source counts are truncated (Chapter 9).
• Own-site figures are understated. Three cited domains — cezbemobilya.com, cilek.live and ikea.com.tr — look like a tracked entity’s own site but are not attributed to it, because the registry holds a different website.
• Three qualifying positions rest on two models. Eymense on the attribute cut (17 mentions), Hepsiburada on the discovery cut (19) and Modoko on the use-case cut (13). Their visibility is model-specific, not panel-wide.
• Rank built on position. Kilim Mobilya, Adore Mobilya and Weltew Home on the origin-constrained cut, and Çilek on the use-case cut, rank high mainly because they are named early when named at all. That is the score working as designed, but it is a different kind of visibility from being named in half of all answers.
• Small score gaps are not meaningful. In the packed middle of each board — and at the top of the origin-constrained board, where Bellona (94.03) and İstikbal (94.01) sit together — differences of a point or two should not be over-interpreted.
• Components can exceed 100. Because each component is scaled to the qualified leader and never clipped, 98 of 1,103 entity-cut rows (8.90%) exceed 100 on a component — all of them unqualified.
• Near-name pairs. Six pairs of similar names among entities named in at least three answers were flagged: Doğrucan Mobilya / N Mobilya, Erduran Mobilya / N Mobilya, Lider Vip / Lider, Mudo Concept / Mudo, QHome / Maq Home and Özsev Mobilya / V Mobilya. None shares a website on file, and each pair is tracked as two entities. The outcome is in the data; the reasoning behind each merge or non-merge is not.
• Brands, not groups. Some ranked brands share an owner — Kelebek and Doğtaş, or Enza Home and Yataş. The report ranks each brand separately and never adds their figures together.
• System prompt and locale. Neither was recorded with this run.
• Single, point-in-time measurement. All responses were collected on 25 September 2026. Five repeats per question-model cell capture variation within a model, not drift from day to day.
• Models are probabilistic. The same question can produce different answers; repeats reduce this variability but do not remove it.
• Web search depends on momentary conditions. Grounded answers depend on what was indexed and served at the time of collection.
• Precision over recall. A brand or company is counted only where it is named in the answer, not where it appears only in a web address, and a shop named only as the place to buy is often not counted. Brands referred to indirectly, or only by a product name such as a sofa-set model, are not counted either.
• Visibility is not quality. This report measures mention, position and breadth only. It does not assess whether a brand was described positively, whether a recommendation was accurate, or whether a product is well made. A score is not an endorsement.
• Model versions date quickly. These findings are specific to the model versions in section 2.1, as of September 2026.

12. Appendix & Data

One-sentence takeaway: All 30 questions and all 290 tracked entities behind this report are listed below, and the underlying data is available on request.

12.1 All 30 questions

12.2 All tracked entities (290)

The table lists all 290 entities tracked in this study, ordered by general-cut mentions, and marks the 23 that qualify on the general cut. 50 entities qualify in at least one cut. Entities with no origin on file are shown as Unknown. Brand pages are not linked in this edition.

12.3 Data availability

The data behind this report is available on request: per-entity metrics across all six cuts, the qualified leaderboards and their brand-only variants, the entity registry with origins, the per-model breakdowns, the origin-share tables on both bases, the source attribution tables, the raw-layer trend file, and the published scale registry.

All figures reflect a single collection day (25 September 2026) and are subject to the limitations in Chapter 11. This is the second edition; the study is repeated quarterly.

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.

Report period: September 2026 · Edition 2 of a quarterly series

Written by

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 →