746 responses, 5 models, three market cuts: the first AI-visibility snapshot of Turkey's furniture market

Vertical: Furniture, sofas, bedroom & home furnishing (mobilya) — Turkey
Method: A single point-in-time study of 746 usable responses — five large language models (Claude, ChatGPT, Gemini, Grok, Perplexity), each asked 30 Turkish-language questions five times
Collection date: 27 June 2026 Key terms: AI visibility, AI Visibility Score, furniture brands, Turkish furniture (mobilya) market, Turkish (yerli) furniture brands, LLM brand recommendations, ChatGPT/Gemini/Claude/Grok/Perplexity brand recommendation, Generative Engine Optimization (GEO).

This report reproduces the real Turkish-language questions consumers ask AI assistants and measures how often, in what order, and across how many models five large language models name each furniture brand. 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: AI visibility in Turkey’s furniture market is concentrated in a tight “big three” — İstikbal, Bellona and Doğtaş — that leads every cut within a few points of one another; below them visibility falls off a cliff, and the origin-neutral shelf is strongly domestic.

• It’s a three-way race at the top, not a runaway. İstikbal tops the Main leaderboard with a 96.78 score, but Bellona is actually mentioned most often (50.5% of all 746 responses vs İstikbal’s 46.9%); İstikbal leads because it is named earlier in answers (a higher position score). Doğtaş (90.12) completes the trio. The three sit within 6.7 points of each other and all appear in all five models.
• Then a cliff. A 23.5-point gap separates third-placed Doğtaş (90.12) from fourth-placed Kelebek (66.62). This break — far larger than any other on the list — is the defining feature of the market: a clear big three, then everyone else.
• 28 entities qualified in the Main cut. From a 352-entity dictionary, only 28 cleared the 5% threshold; their average score is 52.5. Of the 28, 22 are brands and 6 are retailers/marketplaces (Vivense, Koçtaş, Trendyol, Modoko, Hepsiburada, Masko), kept in deliberately so the report describes the shelf as shoppers actually meet it.
• One foreign brand cracks the top tier. IKEA is fifth in the Main cut (58.73) and the only foreign entity to qualify there. Every other qualified entity is Turkish.
• The models broadly agree. Unlike some markets, no model flips the leader: all five place İstikbal, Bellona and Doğtaş in their top three. They differ mainly in breadth — Grok names ~10.6 brands per answer and Perplexity ~9.0, while Claude and GPT-4o-mini are leaner (~4.2).
• Answers are almost always grounded. 98.4% of responses carried a citation; only 12 were memory-only — too few to support a reliable web-vs-memory comparison (we report it as an indicative note only).
• The yerli cut concentrates even harder — but leaks. In the 13 Turkish-only questions, the big three lead by even wider margins (İstikbal 99.15). Notably, IKEA still appears at #17 in this “Turkish-brands-only” cut — evidence that models do not perfectly honour an origin instruction.
• The open market is strongly domestic. Across the 17 origin-neutral questions, 88.4% of answers name at least one Turkish brand versus 34.8% that name at least one foreign brand; only two foreign brands qualify (IKEA #4, Roche Bobois #23). Turkey’s furniture shelf, as AI sees it, is domestically owned.
• Sources split between brand sites and a listicle long tail. Brand-owned sites are the single largest citation block (43.3%), nearly tied with a 291-domain editorial long tail (42.7%); retail/marketplace (8.4%) and forum/UGC (5.6%) are minor. The three most-cited domains are the big three’s own sites: bellona.com.tr, istikbal.com.tr, dogtas.com.

Why it matters. Consumers are shifting product discovery from search engines to AI assistants. Whether a brand appears in those assistants’ answers is becoming a discovery channel in its own right. This report captures the first snapshot (baseline) of this new “visibility shelf” for Turkey’s furniture market: which brands stand out, how the picture changes when you ask specifically for yerli brands or drop the origin restriction, and how a brand can read its own position. A reminder: the numbers below describe visibility and information availability — not which brands are best.

2. Methodology

One-sentence takeaway: Five large language models were asked 30 Turkish questions five times each — with no system prompt — producing 746 usable responses; brands and shopping entities were identified by automated extraction plus human validation and ranked with a three-component (45/30/25) score, computed separately for the Main, Local and Open cuts.

2.1 Scope and models

The study queried five large language models directly via API with identical questions. Each model received only the user question — no system prompt was used, temperature was left at the provider default, and no region/locale parameter was defined. Because the questions are in Turkish, the models infer the Turkish market from language alone; an explicit locale was deliberately not set.

Scale: 30 questions × 5 repeats × 5 models = 750 responses; 4 were excluded (empty Gemini generations), leaving 746 usable. Single run, 27 June 2026. The final dictionary holds 352 entities (338 Turkish, 14 foreign), comprising 336 brands, 11 retailers and 5 marketplaces. Across the Main cut, about 5,260 brand mentions were captured.

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 Sonar is search-grounded by design). Search was neither forced nor disabled, so differences in usage reflect each model’s own behaviour, not an external constraint. Whether a response drew on web search or the model’s own knowledge was inferred from whether the provider returned any citation/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 98.4% of usable answers were grounded.

2.3 Questions and the three cuts

The 30 questions are evenly split across three behavioural types (10 each) — discovery (“the best / which brand should I buy” recommendation questions), attribute (durability, MDF quality, stain-resistant fabric, value-for-money, etc.) and use_case (need-driven: small living rooms, homes with cats or children, wedding furniture packages, young-people’s rooms, balcony sets). Unlike some verticals, this furniture edition has no topical sub-categories, so the category breakdown uses these three behavioural types.

The same questions are read in three market cuts, each ranked independently:

The Turkish-vs-foreign comparison (Chapter 8) rests solely on the 17 Open questions — the only fair basis for comparing Turkish and foreign brands.

2.4 Brand extraction and validation

Brands were extracted in two stages. (1) Automated extraction: emphasised brand names and citation domains were collected, and spelling/format variants merged into a single canonical name. (2) Human validation: the candidate list was reviewed by hand — real brands confirmed, non-brands (category words, generic phrases, fabric/material types, places) filtered out, aliases merged, each brand labelled Turkish/foreign, and a further research pass run on the long tail.

Origin rule. A brand is Turkish if founded and headquartered in Turkey; foreign otherwise, even when widely sold in Turkey (e.g. IKEA, Roche Bobois). Entities kept. Per study scope, retailers and marketplaces are kept as entities (e.g. Trendyol, Hepsiburada, Koçtaş, Vivense, Modoko, Masko) so the shelf is described as shoppers actually meet it; a brand-only view can be derived by filtering for brands. İnegöl note. İnegöl is a furniture-manufacturing hub near Bursa, not a brand; generic “İnegöl …” phrases are treated as a place, while specific İnegöl-based companies with their own domains are kept as brands.

Definitions.

• Mention: a validated brand (or alias) appears in the answer text — counted once per response (presence, not frequency), matched word-boundary aware, case- and diacritic-insensitive, longest-alias-first so a shorter name never double-counts inside a longer one.
• Position (MRR): mean reciprocal of the brand’s first-appearance rank within each answer (first = 1.0; second = 0.5; …), averaged only over the responses where the brand appears.
• Breadth: number of the 5 models (0–5) that mention the brand within that cut.

2.5 AI Visibility Score (0–100)

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

Each component is scaled so the leading qualified brand in each cut = 100 (breadth scaled across the five models), then combined with the weights. Mention frequency carries the most weight because being named at all is the primary signal of visibility; position is second because a brand named early is more likely to reach the reader; breadth is third because appearing across multiple models indicates a more robust, less model-specific presence. The weights are a deliberate editorial choice: at the top of the leaderboard the gaps are large enough to be robust to reasonable reweighting, while in the tightly-packed middle small differences of a point or two should not be over-interpreted.

Qualification (≥5% rule): only brands mentioned in at least 5% of a cut’s responses are ranked (28 Main, 31 Local, 26 Open). This threshold prevents a rare brand that happens to appear first in a handful of answers from inflating the position component. Unfiltered metrics for all entities are available on request (see Chapter 12).

2.6 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 domain 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. Herm.io is the publisher of this report; excluding its own footprint ensures the company does not measure, cite, or benefit from itself.

3. Overall Visibility Leaderboard (Main cut)

One-sentence takeaway: A tight big three (İstikbal, Bellona, Doğtaş) leads within 6.7 points; a 23.5-point cliff then separates them from the rest, and the 28-entity list — brands plus six retailers/marketplaces — tapers quickly into a long tail.

The list below ranks all 28 entities that cleared the 5% threshold in the Main cut, by AI Visibility Score. As a reminder, this ranks visibility, not brand quality. Tap a brand name to open its page.

Average score of the 28 qualified entities: 52.5.

3.1 Tier narrative

The big three (96.78 / 94.36 / 90.12). İstikbal, Bellona and Doğtaş are in a class of their own — each appears in roughly half of all 746 answers and in all five models. The interesting nuance is how they differ: Bellona is mentioned most often (50.5%), but İstikbal is named earliest (the highest position score, MRR 0.64), which is why İstikbal leads overall despite the lower mention rate. Two different questions sit behind a brand’s visibility — “Is it named often?” and “Is it named early?” — and here they pull in different directions.

The cliff. Below Doğtaş, the score drops 23.5 points to Kelebek (66.62, 33.4%) — the single largest break on the list. Everything past this point is materially less visible than the leaders.

The upper-middle (ranks 5–11). Here brands and shopping entities mix: IKEA (58.73) is the lone foreign name; Gündoğdu (56.24) and Enza (55.62) are domestic makers; and three entities — the retailer Vivense (55.87) and the marketplaces Koçtaş (55.22) and Trendyol (54.64) — sit among them, because shoppers (and the models) treat “where to buy” and “which brand” as one shelf. Medusa (#11, 54.46) is a telling case: despite a low 7.0% mention rate, it ranks here because when it is named it appears very early (MRR 0.4954) — niche but high-prominence.

The long tail (ranks 12–28). Scores fall from ~49 to 28.07 (Alfemo). Most entities still appear in all five models, but their mention and position components decline; at the very bottom, breadth also drops (Sivriler, Masko and Divanev at 4 models, Alfemo at 3). A lower rank here means lower current visibility to AI systems — not lower quality — and many capable makers sit below the 5% threshold entirely.

A brand-only reading. Of the 28 qualified entities, 22 are brands; filtering the six retailers/marketplaces out leaves the same leaders (İstikbal, Bellona, Doğtaş, Kelebek, IKEA, Gündoğdu, Enza, Medusa …) in the same order. Keeping the entities in does not distort the brand picture; it simply adds the shopping destinations that share the shelf.

4. Differences Between Models

One-sentence takeaway: The five models agree on the big three but differ sharply in breadth — Grok and Perplexity name many brands per answer, while Claude and GPT-4o-mini stay lean — and Perplexity surfaces by far the widest long tail.

4.1 Per-model behaviour summary

Two patterns stand out. Breadth varies enormously. Grok names ~10.6 brands per answer and Perplexity ~9.0, while Claude (4.23) and GPT-4o-mini (4.21) give short, selective lists. Perplexity surfaces the widest long tail of all — 245 distinct brands across its answers, roughly double the others — yet its top three is still the big three. So a brand’s chance of being named at all depends heavily on which model is asked, even though the leaders are stable.

4.2 The agreement story

Where some markets see the models disagree on the leader, furniture shows strong consensus: every model places İstikbal, Bellona and Doğtaş in its top three. The disagreement is about how much else gets named. Lean models (Claude, GPT-4o-mini) effectively reproduce the big three plus a handful of others; broad models (Grok, Perplexity, Gemini) reach deeper into the domestic long tail, lifting mid-tier makers such as Enza, Gündoğdu and Vivense into view. For a brand outside the big three, being named is largely a question of whether a broad model is doing the answering.

5. Web Search or Model Memory?

One-sentence takeaway: Almost every answer was web-grounded (98.4%); the memory-only segment is too small (12 responses) to support a leader comparison, so this is reported as an indicative note only.

The 746 responses split into 734 grounded answers (carrying a citation) and just 12 own-knowledge answers (no citation). Because the own-knowledge segment is so small, the figures below are indicative only and should not be read as a finding.

The grounded segment simply mirrors the overall picture — the big three on top. The only thing worth flagging in the memory segment is that Enza appears unusually high (8 of 12 answers), level with Bellona, which might hint that an established name carries further without search. With just 12 responses behind it, that is a hypothesis to test in future editions, not a conclusion.

6. Category Ownership

One-sentence takeaway: The big three lead all three behavioural types; discovery questions elicit the widest brand lists, while use-case questions pull retailers (Vivense) and value makers (Gündoğdu) forward.

Each behavioural type covers 10 questions. The table shows responses, the average number of brands named per answer, and the most-mentioned brands in each.

Reading. The big three top every type, but the texture differs. Discovery questions produce the widest lists (≈9.0 brands per answer) and lift mid-tier domestic makers such as Enza into the top five. Attribute questions — durability, MDF quality, stain-resistant fabric — are where IKEA surfaces most strongly (it makes the attribute top five but not the others), reflecting how often it is named as a quality/value reference. Use-case questions — wedding packages, homes with children or pets, small rooms — bring Gündoğdu and the retailer Vivense forward, because need-driven questions often resolve into “where do I buy a full set,” not just “which brand.”

7. The Yerli Cut: Asking Specifically for Turkish Brands

One-sentence takeaway: When questions explicitly ask for yerli furniture, the big three lead by even wider margins and a set of smaller domestic makers surfaces — yet IKEA still leaks into this Turkish-only cut at #17.

This cut uses only the 13 Turkish-only questions (323 responses); 31 entities qualified. The leaders tighten their grip: İstikbal 99.15, Bellona 94.13, Doğtaş 90.37, then Kelebek (68.45) and Gündoğdu (64.62).

Two things stand out. First, the yerli framing surfaces smaller domestic makers that do not qualify in the Main cut at all — Ergül (#11), CaddeYıldız (#14), Savenis (#25), Minar Mobilya Online (#29) and Kargılı (#31) — exactly the kind of regional manufacturers a “find me a Turkish brand” question is meant to reach. Second, and more revealing about the models themselves: IKEA still appears at #17 (5.6%) in a cut whose questions explicitly ask for Turkish brands. The models do not perfectly honour an origin constraint — a small but consistent leakage worth tracking, since it shapes what a shopper who asked for “yerli” actually sees.

8. Open Market: Turkish vs. Foreign

One-sentence takeaway: In origin-neutral questions the furniture shelf is overwhelmingly domestic — 88.4% of answers name a Turkish brand versus 34.8% a foreign one — and only IKEA and Roche Bobois qualify as foreign.

This chapter relies on the 17 open questions with no origin restriction (423 responses) — the only group where Turkish and foreign brands can be compared fairly. With 423 responses behind it, this is a more robust base than a small aside; still, treat the exact figures as a baseline to track across editions.

8.1 Origin reach

The bars below show the share of open-market answers that name at least one brand of each origin (an answer can name both, so these do not sum to 100).

Domestic dominance here is striking, and it is the clearest contrast with other verticals: where some markets split close to evenly between Turkish and foreign once the origin restriction is dropped, Turkey’s furniture shelf stays domestic even when nothing forces it to. Of the 26 qualified open-market entities, 24 are Turkish and only 2 are foreign.

8.2 Open-market (any origin)

8.3 The one foreign exception

IKEA is the only foreign brand with real open-market visibility — fourth overall (68.62), ahead of every domestic name except the big three, and the foreign brand that consumers and models alike reach for as the international reference point. After IKEA, the next foreign name is Roche Bobois, which barely qualifies at #23 (6.6%) as a luxury reference. There is no broad foreign mid-tier here: no cluster of global brands sitting between the Turkish leaders the way foreign names cluster in some other categories. For Turkish furniture brands, the open-market reading is encouraging on its face — but the question to watch each quarter is whether that single foreign exception broadens.

9. The Discovery Ecosystem: Where AI Learns About Brands

One-sentence takeaway: AI’s picture of furniture brands is built from two roughly equal blocks — the brands’ own websites and a long tail of nearly 300 small editorial/listicle sites — with retail and forums playing only minor roles.

When a model answers with web search, it draws on the content available to it at that moment. The cited source addresses were reduced to domains and ranked by how many answers cited them — a map of where these systems read about furniture brands. Note: the machine-readability of source data varies by provider, so this analysis is best-effort (see Chapter 11).

9.1 The most-cited domains

The three most-cited domains are the big three’s own websites — bellona.com.tr, istikbal.com.tr, dogtas.com — each cited in roughly a quarter to a third of answers and by all five models. Right behind them sit “best furniture brands” listicle sites (eniyimobilyamarkalari.com, eniyilerden.com, umutspot.com) and the retail/marketplace names (trendyol.com, koctas.com.tr, vivense.com). The picture is owned brand content + editorial round-ups, with retail listings a clear third.

9.2 Source-type mix

9.3 The concentration-plus-fragmentation finding

The mix tells two stories at once. Brand-owned sites are concentrated and powerful: just 94 domains account for 43.3% of citations, because a handful of large brands’ own websites are cited again and again. Editorial/Other is fragmented and large: 291 domains add up to almost the same share (42.7%), none individually dominant — the furniture equivalent of a long tail of small “top 10 brands” articles and local sites. Together these two blocks are ~86% of all citations; retail/marketplace (8.4%) and forum/UGC (5.6%, mostly YouTube and Ekşi Sözlük) are comparatively minor. In short: to AI, a furniture brand’s visibility rests on its own current website plus how often the listicle ecosystem names it.

10. What the Patterns Suggest

One-sentence takeaway: The most visible furniture names share broad model coverage, a strong owned-website presence that is itself the top-cited source, and — for the leaders — frequent early mention; entities (retailers/marketplaces) are part of the shelf, and the domestic shelf dominates origin-neutral questions.

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

1) Breadth is a baseline condition. Every top entity appears in all five models (breadth component 100). The brands that slip down the list are the ones that lose model coverage (Sivriler, Masko, Divanev at 4 models; Alfemo at 3). Appearing in only some models caps a brand’s ceiling.

2) Mention and position are different levers. Bellona is named most often; İstikbal is named earliest; Medusa is rarely named but, when it is, appears near the top. A brand can be visible by frequency, by prominence, or both — and they move independently.

3) Owned web presence and visibility move together. The three most-cited domains in the entire study are the big three’s own websites. In a near-fully-grounded run (98.4%), a current, well-structured brand site is the factor most visibly associated with the top of the leaderboard — an observed association, not a proven cause.

4) The shelf includes shops, not just brands. Retailers and marketplaces (Vivense, Koçtaş, Trendyol) qualify alongside manufacturers, and rise specifically in use-case questions. For furniture, “which brand” and “where to buy” are answered together; a brand competing for visibility is competing on a shelf that also contains its sales channels.

5) Domestic brands own the open shelf. Even with no origin restriction, Turkish brands dominate (88.4% vs 34.8%), with IKEA the lone strong foreign exception. This is the headline number to track: whether the domestic shelf stays this dominant, or foreign names broaden beyond IKEA, across future quarters.

How a brand can locate itself in the data. Using the dataset (available on request), a brand can read four things in order: whether it appears in all five models; how frequently it is mentioned; how early it ranks when mentioned; and which model, cut or category it is weakest in. Together these show where a brand’s visibility gap sits — without saying anything about whether its products are good.

11. Limitations and Notes

One-sentence takeaway: This report is a single, one-market, point-in-time snapshot; the figures should be read within the limitations below.

• Single, point-in-time measurement. Data were collected in one run on 27 June 2026 — a snapshot, not a trend. This is the first (baseline) edition; later quarterly editions will enable trend analysis.
• Web-vs-memory is not analysable here. Only 12 of 746 answers were memory-only; the segment is too small for any leader comparison and is reported as an indicative note only.
• Entities are mixed into the leaderboard. Retailers and marketplaces are ranked alongside brands by design, to describe the real shopping shelf. A brand-only view is easily derived; readers wanting a pure manufacturer ranking should filter the entities out (the order of the leaders is unchanged).
• Origin instructions leak. Models do not perfectly honour a “Turkish-only” request — IKEA appears in the Local cut at #17. Origin cuts should be read as strong tendencies, not hard filters.
• Small score gaps are not meaningful. In the packed middle of each leaderboard, differences of a point or two should not be over-interpreted; only larger gaps (such as the 23.5-point cliff below the big three) are robust.
• Visibility is not quality. The study measures only mention and position — not whether a brand was described positively, whether a recommendation was accurate, or whether a product is well made. A score is not an endorsement.
• Models are probabilistic. The same question can produce different answers; five repeats reduce but do not remove run-to-run variation.
• Citation-as-proxy. “Web search vs own knowledge” is inferred from whether a citation accompanied the answer, not from internal traces.
• Precision over recall. Brand matching was cautious to avoid false positives; some long-tail or implicit mentions may be undercounted. Common-word collisions (e.g. koltuk = sofa, tarz = style) were detected and removed during validation.
• Partial source coverage. Citation structures differ by provider; domain extraction is best-effort and does not reflect the entirety of every answer.
• Origin classification involves judgement for licensed or multinational brands.
• Token caps differ by provider, which can affect answer length and how many brands are named.
• Model versions date quickly. Findings are specific to the model versions in Chapter 2 as of late June 2026; provider updates may shift results in later editions.

12. Appendix & Data

12.1 All questions (30) and their cuts

Distribution: 10 discovery, 10 attribute, 10 use_case; 13 Local (explicitly yerli/Turkish) and 17 Open (origin-neutral).

12.2 Data availability

The study locked a 352-entity dictionary (338 Turkish, 14 foreign; 336 brands, 11 retailers, 5 marketplaces); of these, 28 / 31 / 26 entities cleared the 5% threshold in the Main / Local / Open cuts respectively. To support scrutiny and reproduction, the underlying data behind this report is available on request: the response-level brand mentions, the qualified-brand leaderboards for all three cuts, the full unfiltered brand metrics, the per-model and per-category breakdowns, the open-market and web-vs-memory breakdowns, the top source domains, and the methodology and data dictionary.

All figures reflect a single point-in-time run (27 June 2026) and are subject to the limitations in Chapter 11. This is the first (baseline) measurement; the study will be repeated quarterly.

12.3 About Herm.io & disclosure

Herm.io is a consumer-behaviour and marketing-data company. We study how people discover and choose brands so that brands can reach the right customers. This report is part of our public research and is published as a recurring quarterly study.

Disclosure & neutrality. Herm.io does not sell SEO or GEO (search/AI-ranking) services, and this report does not recommend any. No brand paid to be included, ranked, or described, and a brand’s score is not an endorsement or a judgment of its quality — it is a measure of visibility only. To keep the analysis objective, all citations to Herm.io’s own domain were excluded from the source data, so the company does not appear in, measure, or benefit from its own study (see Section 2.6). Brands that want to understand their position in the data are welcome to book a call for a neutral walkthrough of the findings; this is advisory and free.

Report period: June 2026 · Edition 1 (baseline) of a quarterly series