Garenta takes car rental, the tour-agency shelf turns entirely Turkish, and one assistant stops searching

Vertical: Travel — tour agencies & car rental — Turkey
Method: A single point-in-time study — five large language models (GPT-5.6-luna-pro, Gemini, Claude, Perplexity, Grok), each asked 30 Turkish-language questions five times
Collection date: 11 September 2026
Edition: 2 of a quarterly series. Prior edition: June 2026, restated 14 September 2026

Key terms: AI visibility, AI Visibility Score, travel brands, Turkish tour agencies, car rental Turkey, Turkish (yerli) travel 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 and measures how often, in what order, and across how many models five large language models name each travel brand. Because tour agencies and car rental are distinct markets, they are ranked separately. It is a single point-in-time study and should be read within the limitations in Chapter 12.

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 travel 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, safer, cheaper or more reliable. 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 service quality, price or trustworthiness. Readers and travellers 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: Car rental changed hands — Garenta displaced Avis at the top — while the tour-agency shelf went the other way and closed entirely to foreign brands, all against a panel in which one assistant largely stopped searching the web.

• Garenta has taken car rental from Avis. Garenta is named in 66.6% of the 299 car-rental responses, up from 39.33% in June, and scores 88.30. Avis follows at 81.80 (54.5%), Enterprise at 74.70. The +27.2 pp move in raw mention rate is the largest of any brand in either market.
• Jolly Tur holds tour agencies, and the gap behind it narrowed. Jolly Tur is named in 74.6% of 449 tour responses (98.80), with ETS Tur (93.80) and Setur (89.60) close behind after rises of +14.8 pp and +13.3 pp respectively.
• No foreign brand qualifies in tour agencies. All 15 ranked tour agencies are Turkish; the foreign column is empty for the first time in the series. By mention, that market is 98.2% Turkish across its 18 questions. Car rental is the opposite shelf: 62.6% Turkish, with 9 of its 32 qualifiers foreign.
• 47 brands qualified for ranking. Of 205 tracked entities, 15 tour agencies and 32 car-rental brands cleared the 5% mention threshold within their own market. Average scores: 54.80 (tour) and 45.20 (car rental).
• One model largely stopped searching. Overall grounding fell from 98.1% to 87.57% — but that is almost entirely one model. Gemini went from 90.7% to 39.3%, and 91 of the 93 ungrounded answers in the entire run are its.
• All three unchanged models narrowed their repertoire. Measured on June’s own brand list so the growing dictionary cannot flatter September, Claude Haiku fell from 4.29 to 2.45 brands per answer, Grok from 9.43 to 7.95, Perplexity from 5.87 to 5.34. These three are identical model strings across editions, so this is behaviour, not a panel change.
• The Turkish share of the open market fell about 5 points, and almost all of it is real. On the 498 responses to the 20 origin-neutral questions, mentions split 75.5% Turkish / 24.5% foreign, against a restated June figure of 80.5%. Origin-label corrections moved the number the other way, by +0.5 pp; newly found brands added +0.1 pp. Chapter 8 decomposes it row by row.
• Sources broadened and thinned. Brand-owned sites remain the largest single type but fell from 48.0% to 41.4% of 5,850 citations, while the long tail grew from 31.5% to 39.4%. 636 distinct domains were cited; 178 of them exactly once.

Why it matters. Travellers are shifting discovery from search engines to AI assistants, and this is the first edition of this series with a prior quarter to compare against. What it shows is that the shelf is not stable: a market leader can change in a quarter, a whole market can close to foreign brands, and the assistants’ own behaviour — how widely they list, whether they search at all — moves as much as the brands do. A reminder: the numbers below describe visibility and information availability — not which brands are best.

2. Methodology

One-sentence takeaway: The same 30 Turkish questions as June were asked five times to five models with no system prompt, producing 748 analysed responses; entities were matched against a 205-row dictionary and ranked with a three-component (45/30/25) score scaled separately within each market.

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, no temperature override, and no region or locale parameter. Because the questions are in Turkish, the models infer the Turkish market from language alone.

Two of five models changed between editions. GPT-4o-mini to GPT-5.6-luna-pro is a tier change, not a version bump. Gemini 2.5 Flash-Lite to Gemini 3.6-flash brought the grounding collapse described in Chapter 5. Claude Haiku 4.5, Perplexity Sonar and Grok 4.3 are identical model strings across both editions. Every comparison with June in this report carries those two confounds, and each is printed beside the figure it affects.

Scale: 30 questions × 5 repeats × 5 models = 750 responses collected, 748 analysed. Two Grok answers returned completed with an empty body; they are excluded from every denominator rather than counted as “brand not mentioned.” Denominators are computed per question-model cell, never as questions × models × repeats. The questions cover two markets: tour agencies (18 questions → 449 responses) and car rental (12 questions → 299 responses).

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 was inferred from whether the provider returned any citation or source. This is a reasonable proxy, not direct proof of the model’s internal process. In this run 87.57% of analysed answers were grounded, against 98.1% in June — see Chapter 5.

2.3 Questions and cuts

The 30 questions are unchanged from June. Of them, 10 are “constrained” (they explicitly ask for Turkish or local providers) and 20 are “open” (no origin restriction). The Turkish-versus-foreign comparison in Chapter 8 rests solely on the 20 open questions — the only fair basis for comparing Turkish and foreign brands, since a constrained question asks for a Turkish answer by name.

The data is cut eight ways. local + open sum to the full set; the two market cuts sum to the full set; the three question-type cuts sum to the full set. All three identities hold at 748.

2.4 Entity resolution

June’s validated dictionary of 175 rows was used as the seed. Candidates were extracted case-insensitively from URL-stripped answer text with Turkish-aware case folding, word-boundary anchored matching, and longest alias first so a short name is never counted twice inside a longer one.

• 205 entities tracked — 163 carried forward from June and 42 new this edition.
• 12 entities absorbed by same-domain merges. A domain-uniqueness check found 11 domains carrying more than one entity. tatiltur.com held both Tatil Tur and Tatiltur; qcar.com.tr held QCAR and QCAR Mobilite; wts.web.tr held three. All were merged into a single entity and the matcher re-run. June’s published count of 173 tracked brands was restated to 162 on the same basis.
• 486 candidates dropped with a recorded reason each. The largest is TÜRSAB, which appears in 132 responses — the Turkish travel agency association, named constantly as a trust signal but not a company.
• Six name merges applied to the alias map before scoring: ETS into ETS Tur, Çizgi into Cizgi Rent a Car, Moov into GetirAraç, Romos Travel into Luxury Tour Turkey, Antonina Turizm into Antonina Tourism, Data Turizm into DataTur.
• 47 entities carried from June were never named in any September response. Only one of them had qualified in June: Berse Rent A Car, from 9.00% to zero. They are listed in the appendix rather than deleted, because the absence is itself a finding — and eight of them still had their websites cited as sources (Chapter 10).

Six entities are tracked but excluded from every leaderboard as out of scope: AJet, Balkan Turları, Pegasus, Tatil Villası, Vizesiz Rotalar, Şikayetvar. They remain in the dataset and in the appendix index.

2.5 The AI Visibility Score

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

Each component is scaled so the highest-scoring qualifying, rankable entity in that cut equals 100, then the three are combined. Brands below the 5% line carry no published score.

Worked example — Garenta, Car rental. It is named in 199 of 299 responses, so its mention rate is 0.665552, which is also the cut maximum: 100 on the mention component. Its MRR is 0.532552 against a cut ceiling of 0.870833, giving 61.1543. It is named by 5 of 5 models, giving 100. Combined: 0.45 × 100 + 0.30 × 61.1543 + 0.25 × 100 = 88.3463, published as 88.30.

2.6 Why a score cannot travel between editions unchanged

A score is a competitive position, not an absolute measurement. It says where a brand stands against the strongest performer in its market this quarter. It does not say that 80 is good and 50 is bad, and it is not a percentage of anything. The consequence is that a score can move because the curve moved, without the brand doing anything at all.

This quarter the curve moved in both markets, in opposite directions.

Jetset was named 25 times in June — 5.56%, barely above the qualifying line — and set the position ceiling for the entire tour-agency market. In September it was named twice. It no longer qualifies and no longer sets the ceiling, so the divisor fell by a fifth and every tour-agency score rose for that reason alone. Car rental moved the other way: Biwatt’s own MRR climbed, raising the divisor and suppressing every car-rental score.

Therefore every trend figure in Chapter 9 is published beside the change in raw mention rate and model coverage. Read the score for rank within this quarter; read the mention rate for movement between quarters. The leaderboards in Chapter 3 carry a Δ column showing each brand’s score change since June; because the ceiling movement above produces part of that change for reasons unrelated to the brand, read Δ as direction, not exact size. The layer that compares like for like is the raw one: mention rate, MRR, breadth.

2.7 Origin, neutrality and self-exclusion

A brand’s origin is the country of the company that owns the brand, not the company that operates it in Türkiye. Avis, Budget, Enterprise, Sixt, Goldcar, Alamo and National are all run in Türkiye by Turkish operators and all remain foreign. Where the global proprietor and a regional licence-holder differ, the global proprietor governs: Budget’s Turkish site names a UK company as holder of the mark for Europe, the Middle East and Africa, but the mark is owned globally in the United States, so Budget is recorded as US.

Evidence is recorded per entity and audited against the source it rests on rather than assigned by list.

Only 11 of 205 origins rest on a trademark register entry. The great majority — 73 — rest on corporate evidence: a KVKK notice, a terms page, a contact page, a TÜRSAB licence record, and in one case a newspaper interview in which Central Rent A Car’s owner describes transferring the brand. That is real evidence and it is published with its URL, but it is weaker than a registry record, and the tier column says so. A further 119 entities are carried from June without re-examination; none of them qualifies in either market. One entity, Palm Rent, remains unresolved and is excluded from every origin share.

10 origins changed from June, and one further change was considered and rejected.

The largest of these is Coral Travel, which June recorded as Russian and which is Turkish — Coral Holding A.Ş., WIPO mark 1738482. It is the single change that empties the foreign column of the tour-agency leaderboard, and its effect on the origin share is isolated in Chapter 8 rather than folded into the market movement.

This is a market-wide, neutral study with no focus brand. 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: Tour agencies keep a clear leader and a purely Turkish board; car rental changed hands, with Garenta now ahead of Avis by 6.5 points.

As a reminder, these rank visibility, not service quality, price or reliability. The two markets are scored on separate scales and their numbers are not comparable with each other. The Δ column compares a brand only against its own June score in its own market, and gives direction rather than exact size — see 2.6.

3.1 Tour agencies — 15 qualified brands

Average score of the 15 qualified tour agencies: 54.80. All 15 are Turkish.

Reading. Jolly Tur tops the table on the maximum mention rate in the market and full breadth, but its lead is narrower than June’s: ETS Tur and Setur both gained more than 13 points of raw mention rate this quarter. Four brands are named in more than half of all tour responses. Below them the board thins quickly — two of the bottom four are named by two models or fewer, and their ranks should be read with Chapter 4 in hand.

3.2 Car rental — 32 qualified brands

Average score of the 32 qualified car-rental brands: 45.20. 23 are Turkish and 9 foreign.

Reading. Garenta leads on the strength of frequency — it is named in two thirds of car-rental answers, the highest mention rate anywhere in the study — while Avis retains a slightly better position score. The international chains have not left: Avis, Enterprise, Budget and Sixt hold four of the top six. Biwatt at #7 is the clearest small-sample artefact in the study: 20 mentions, a 6.7% mention rate, and the highest MRR in the market at 0.8708. It ranks where it does because of where it is named, not how often, and the same caution applies to Avis Filo, Goldcar and Metafleet lower down.

3.3 The brand-only view

Aggregators and marketplaces are ranked as entities throughout this report, because they are what the assistants actually name. For readers who want the direct-provider picture, the same two boards with marketplaces stripped by name:

Stripping marketplaces removes two brands from the tour board and seven from car rental, and it does not change who leads either market.

4. Differences Between Models

One-sentence takeaway: The five assistants agree on who leads and on almost nothing else — and the three models that did not change between editions all named fewer brands per answer than they did in June.

4.1 Per-model behaviour

All three unchanged models narrowed. Claude Haiku 4.5, Perplexity Sonar and Grok 4.3 are identical model strings across the two editions, and each named fewer of June’s brands per answer than it did in June — Claude most sharply, from 4.29 to 2.45. Because the comparison is held to June’s brand list on both sides, this is not an artefact of the dictionary growing. Gemini is the one model that widened, from 6.08 to 7.45, and it is also the model that changed version and largely stopped searching; those two facts are not separable here.

Answer length moves independently of breadth. Grok writes the longest answers (3,116 characters on average) and names the most brands; Perplexity writes the shortest (1,483) but still names more than twice as many brands as Claude, whose answers are longer.

4.2 Each model’s most-named brands

Each model’s six most-named entities across its own responses. Rate is the share of that model’s answers naming the brand.

Jolly Tur appears in every model’s top six, which is the only genuine consensus in the table. Claude’s list is otherwise unlike anyone else’s: its most-named entity is Tatiltur, at 25.3% of its answers, against 10.7% across the whole study.

4.3 When a rank is really one model’s habit

Among entities named at least 20 times, 12 take more than 60% of their mentions from a single assistant. Entities below the 20-mention floor are not assessed, because the share is unstable at low counts.

Two of these are ranked car-rental brands named by effectively one model. Kiralasür takes 100% of its 20 mentions from Grok — it is the only qualifying brand in the study with a breadth of 1, and its rank of #32 exists because one assistant knows it. DiscoverCars takes 95.0% of its 20 mentions from Grok. In tour agencies, Fest Travel takes 96.7% from Gemini and Tatiltur 79.2% from Claude. None of these is an error; all four cleared the threshold honestly. But a reader looking for a market position should treat them differently from Garenta or Jolly Tur, which every model names.

5. Web Search vs. Model Memory — Why This Edition Still Omits the Split

One-sentence takeaway: There are now enough ungrounded answers to compare, but 91 of the 93 come from a single model, so any comparison would measure Gemini rather than the effect of searching.

June omitted this comparison because almost every answer was grounded. September omits it for the opposite reason, and the reason is itself the finding.

Gemini’s grounding rate fell from 90.7% to 39.3%. That single change accounts for essentially the entire drop in the panel average. Three models searched on every answer; GPT-5.6-luna-pro missed two. Splitting the run into grounded and ungrounded segments and comparing the leaders would produce a table that looks like a finding about web search and is actually a table about Gemini — so it is not published.

Two things are worth saying that do not require the split. First, Gemini changed version between editions, so this is not a clean before-and-after of one system’s behaviour. Second, Gemini did not become less visible in the results as a consequence: it names more brands per answer than it did in June, and it is the dominant model behind four of the twelve concentrated entities in Chapter 4. Whatever it lost in live sources, it did not lose in willingness to name brands. The grounding rate will be measured again next edition; if the ungrounded segment spreads across models, the comparison returns.

6. Question-Type Ownership

One-sentence takeaway: Tour agencies reshuffle by question type — ETS Tur outranks Jolly Tur on attribute questions — while Garenta leads all three types in car rental.

Each market’s questions are split across three behavioural types: discovery (general “best” or “recommended”), attribute (specific qualities — affordable, cancellation-friendly, electric fleet) and use_case (need-driven — first trip abroad, honeymoon, family car, long-term rental). Rates below are relative to that type’s responses within that market, so the denominators differ from the leaderboards.

Tour agencies

Car rental

Reading. The tour-agency order is not as stable as June’s. Jolly Tur leads discovery and use-case questions but ETS Tur leads attribute questions at 72.0%, ahead of Jolly Tur’s 67.2% — the only cut in the study where Jolly Tur is not first. Setur is second on use-case questions and third on the other two. Car rental is the reverse of June’s picture: Garenta leads all three types, most decisively on discovery at 78.7%, where June’s board was led by international chains. The chains hold second and third almost everywhere, and Rent Go breaks into the use-case top three.

7. Entity Type as an Attribute

One-sentence takeaway: Marketplaces are a bigger share of the car-rental shelf than of the tour-agency shelf, but no entity type is a route to the top of either.

This chapter reports an attribute of the entities, not a cut of the data: every entity carries a type, and the tables below group the existing leaderboards by it. It is not a separate measurement and the denominators are the same as Chapter 3.

The tour-agency board is dominated by Brand+Retailer entities — operators that both package and sell their own product — which take nine of fifteen places including the top four. Car rental is more mixed: it is mostly pure Brand entities, with seven aggregators and one licensed brand, Avis Filo. No type wins a market: the leaders in both are conventional operating brands, and the aggregators cluster in the middle of the car-rental board rather than at either end.

Seven entities are named in answers to both markets’ questions. Each appears once in the appendix under its own market, so this is the only place their cross-market visibility is visible:

Tatilsepeti is a tour-agency brand named twice on the car-rental shelf; Enuygun is the reverse and much larger, a car-rental aggregator named in 14 tour-agency answers. Şikayetvar, a complaints platform, is named in both and ranked in neither.

8. Origin: Two Shelves, Two Answers

One-sentence takeaway: On origin-neutral questions 75.5% of mentions go to Turkish brands, down about five points from June — but the headline hides two markets that now point in opposite directions, with tour agencies at 98.2% and car rental at 62.6%.

8.1 The constrained shelf and the origin-neutral shelf

The two figures must be read together, because they answer different questions.

On the 10 constrained questions — those that explicitly ask for a Turkish or local provider — 94.9% of mentions go to Turkish brands. That number is close to a measure of instruction-following, not of market position, and it is reported here only so the origin-neutral figure has something to sit against. Every origin claim that follows uses the 20 open questions and their 498 responses.

8.2 Why the origin share moved

June published 79.2% on the open cut. That figure was restated to 80.5% on 14 September when the same-domain merges and the corrected origin labels were applied to June as well as to September, so that both sides of the comparison sit on the same basis. September is 75.5%. Quoting those two numbers alone would still be misleading, so the change is decomposed:

Read the third row first. The market moved -5.1 points; the origin corrections moved the figure the other way, by +0.5 points. That matters because ten origin labels changed this quarter, and a reader is entitled to suspect that a falling domestic share is really the analyst changing his mind. It is not: had June’s labels been kept, September would read 74.9% rather than 75.4%. The corrections — chiefly Coral Travel moving from Russian to Turkish — make the domestic share higher, not lower. Newly found brands contributed a further +0.1 points, which is close to nothing.

June’s anchor is bounded, not exact, at 80.4–80.5. June’s stored data was collapsed over the five repeats, so where two names of a now-merged pair appeared in the same answer the union cannot be recovered. The bound is 0.1 points wide and does not affect the direction of anything above.

8.3 The two markets have separated

The tour-agency shelf has closed. All 15 qualifying tour agencies are Turkish, and 98.2% of the 2358 mentions in that market go to Turkish brands. In June the sole foreign qualifier was Coral Travel; it is now recorded as Turkish, which is why the foreign column reads zero. That is a labelling change, not a brand leaving the board, and it should be read as such — the honest statement is that the one brand that made the tour-agency board look internationally contested was never foreign.

Car rental remains genuinely split. 9 of its 32 qualifiers are foreign — Avis, Enterprise, Budget, Sixt, Europcar, Hertz, Goldcar, DiscoverCars and Avis Filo — and they hold four of the top six places. But the leader is Turkish for the first time in the series, and the Turkish share of mentions in that market is 62.6%.

8.4 The open market, brand by brand

The individual ranking shows the structure the aggregate hides. The four most visible brands on origin-neutral questions are Turkish tour agencies; the first foreign brand is Avis at #5, and every foreign brand in the top fifteen is a car-rental chain. A traveller asking a question with no origin restriction gets an answer dominated by Turkish tour operators, with foreign visibility almost entirely confined to the car-rental shelf — the same structure as June, on a smaller foreign share.

9. Trend Against June 2026

One-sentence takeaway: Garenta’s +27.2 points of mention rate is the largest move in either market, but the scoring curve moved too — in opposite directions in the two markets — so the leaderboard Δ gives direction and the raw layer is the one that compares like for like.

Read mention rate before score in every row below. Chapter 2.6 explains why: the tour-agency position ceiling fell -20.9% when Jetset stopped qualifying, which lifted every tour-agency score for reasons that have nothing to do with the brands; the car-rental ceiling rose 15.2%, which suppressed every car-rental score. Two of the five models also changed between editions. No figure in this chapter is strictly like-for-like.

The trend set covers every entity that qualified in either edition, in either market: 36 appear in both, 11 are new this edition, and 14 dropped out.

9.1 The largest moves in raw mention rate

Garenta is the headline: from 39.33% to 66.6% of car-rental responses, a rise of +27.2 points, and from #3 in June to #1. MNG Turizm is the mirror image, falling 22.0 points from 40.44% to 18.49% and dropping from a top-five tour position to #8. Sixt fell 7.2 points and Goldcar 6.3. Note how badly score change would mislead on its own: ETS Tur’s score rose 17.0 points on a mention-rate rise of 14.8, while Enterprise’s score rose only 2.5 on a rise of 13.2 — the difference is the curve, not the brands.

9.2 Entering and leaving the board

Eight of the eleven new qualifiers are in car rental, and three of those — DRD, QCAR and RosCar — qualify partly because the same-domain merges in Chapter 2.4 consolidated mentions that were previously split across two names. Berse Rent A Car is the sharpest exit: 9.00% in June, named in zero responses in September, and the only June qualifier to reach zero. Its website was still cited as a source in nine answers (Chapter 10).

9.3 Two rows whose June baseline is a lower bound

Where a June entity was absorbed into another by the same-domain merges, June’s surviving row keeps only its own published mentions, because June’s stored data cannot tell whether both names appeared in the same answer. For two entities this matters enough to state:

QCAR’s published fall of 3.0 points is a floor: the true fall is somewhere between 3.0 and 8.3 points. Coral Travel’s published rise of 2.3 points is a ceiling: the true rise is between 0.5 and 2.3. Every other absorbed pair’s survivor either does not appear in the trend set or was comfortably below the threshold in June, so no other row is affected.

10. The Source Ecosystem

One-sentence takeaway: Brand-owned sites are still the largest single source type but fell from 48.0% to 41.4% of citations, and the long tail grew — the assistants are reading more places, and more shallowly.

Every source address returned by a model was reduced to a domain and counted once per response. Across the run, 636 distinct domains were cited, in 5,850 domain-citations.

Providers do not return sources the same way, and a single extractor would silently lose one of them. OpenAI, Anthropic and Perplexity return citation objects with real URLs. Grok returns citation titles as bare numerals, so the domain is read from the URL. Gemini returns vertexaisearch redirect wrappers for its references and stores the grounded text, not the URL, in its citations field; 577 of its references arrived wrapped and the real domain was read from the title field. Without that handling, Gemini’s sources would be absent from this chapter entirely.

10.1 The most-cited domains

Eight of the top ten are brand-owned, and the two that are not — the editorial site onedio.com at #3 and the aggregator enuygun.com at #5 — sit among them. The concentration is shallow: 485 of the 636 domains were cited in fewer than ten responses, and 178 exactly once.

10.2 Source-type mix, against June

The picture has shifted away from official brand sites and towards everything else. Brand-owned fell 6.6 points, editorial fell 1.9 and forum content fell 2.7, while the unclassified long tail rose 7.9. Some of that is the new Official / Association category absorbing citations that were previously scattered. The rest is a genuine broadening — with 636 domains cited against 484 in June, the assistants are drawing on half again as many places.

10.3 Read but not named

Eight entities were searched for in all 748 responses, named in none of them, and yet had their own websites cited as sources.

The sharpest case is Berse Rent A Car, which was named in 9.00% of June’s car-rental answers, fell to zero mentions in September, and still had berserent.com cited in nine responses. Being read is not the same as being named: an assistant can open a brand’s site, use what it finds, and recommend someone else.

Two constraints on this finding, in the same breath as the finding. The citation counts come from domain matching, so they say a model cited a page on that domain — not that the page was about that brand. And six of the 47 entities tracked but never named have no domain recorded at all, so the absence of a citation for those six is evidence of nothing.

11. What the Patterns Suggest

One-sentence takeaway: Breadth still separates the top from the middle, but this quarter’s clearest lesson is that the assistants’ own behaviour moved as much as the brands did.

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 still the dividing line. Every brand in the top six of either market is named by all five assistants. Of the six brands across both boards named by two models or fewer, five sit in the bottom three places of their board; the exception is Tatiltur at #10 of 15, which one assistant names in a quarter of its answers. Nothing in the data suggests a brand can rank highly on frequency alone while most assistants do not know it.

2) A market leader can change in a quarter. Garenta moved from #3 to #1 in car rental on a 27-point rise in raw mention rate, and it did so against a curve that was suppressing car-rental scores, not flattering them. Whatever produced that movement, it was large enough to survive a scale that moved against it.

3) The assistants moved too, and not in one direction. Three unchanged models narrowed their repertoires while the one model that widened is also the one that largely stopped searching. A brand’s visibility depends on a shelf whose shape is set by systems that are themselves being revised between measurements.

4) Some ranks belong to one assistant, not to a market. Twelve entities with at least 20 mentions take more than 60% of them from a single model, and two of those are ranked car-rental brands with a breadth of two or one. Reading such a rank as a market position would be a mistake the raw columns are published to prevent.

5) The two markets have separated further. Tour agencies are now 98.2% Turkish with no foreign qualifier at all; car rental is 62.6% Turkish with nine. A brand’s context — which market, which question type, which assistant — carries as much information as its overall rank.

6) Being cited is not being named. Eight brands had their own websites read and their names left out. Whatever the relationship between web presence and visibility, it is not a simple one, and this edition can measure the gap without explaining it.

12. Limitations

One-sentence takeaway: This is a single point-in-time snapshot, measured on a panel that changed between editions, with scores that are relative to each quarter’s leader — read every figure inside those three constraints.

• Snapshot, not an average. One collection window, 11 September 2026. Models are probabilistic; five repeats reduce but do not eliminate variation.
• The panel changed. Two of five models differ from June, and one of those changed grounding behaviour sharply. No trend figure in Chapter 9 is strictly like-for-like, and each is printed with its confounds.
• Scores are curve-relative. A score says where a brand stands against this quarter’s leader in its market. The leaderboards’ Δ column is cross-edition score arithmetic, which is not a measurement of the market on its own and should be read as direction; the raw layer is the measurable comparison. See Chapter 2.6.
• Small-sample position inflation persists. Biwatt ranks 7th in car rental on 20 mentions, holding the highest MRR in the market. Its rank is a statement about where it is named, not how often. Anything below roughly 20 mentions is flagged wherever it is used.
• Some ranks are one model’s habit. Twelve entities take more than 60% of their mentions from a single assistant; Kiralasür takes 100%. See Chapter 4.3.
• Two trend baselines are bounded, not exact. QCAR and Coral Travel have a June figure that is a lower bound, because June’s stored data was collapsed over repeats. Both are stated with their range in Chapter 9.3.
• Origin evidence is uneven. Only 11 of 205 origins rest on a trademark register entry; 73 rest on corporate evidence of varying strength, and 119 are carried from June without re-examination. One entity is unresolved and excluded from every origin share. Where the global proprietor and a regional licence-holder differ, the global proprietor governs — a defensible rule, not the only one.
• The web-search split is not published. 91 of 93 ungrounded responses come from one model, so a grounded-versus-ungrounded comparison would measure that model. See Chapter 5.
• Source-type mix is heuristic. Domain types were auto-classified; treat the mix as directional. Source coverage also varies by provider, and Gemini’s required separate extraction.
• Aggregators and marketplaces are ranked as entities, because they are what the assistants actually name. A brand-only view is published alongside each board in Chapter 3.3.
• Precision over recall. Matching is deliberately conservative; brands mentioned indirectly are undercounted.
• Visibility is not quality. The score measures how often, how early and across how many assistants a brand is named. It says nothing about price, service, safety or accuracy, and it is not an endorsement.

13. Appendix & Data

One-sentence takeaway: The 30 Turkish questions, every tracked entity in both markets with its raw metrics, and the 47 entities that were searched for and never named.

13.1 All questions (30)

Distribution: 10 discovery, 10 attribute, 10 use_case; 10 constrained (Turkish-only) and 20 open (origin-unrestricted).

13.2 All tracked tour-agency entities

13.3 All tracked car-rental entities

13.4 Data availability

The study tracked 205 entities; of these, 47 cleared the 5% threshold within their market (15 tour agencies, 32 car-rental brands). To support scrutiny and reproduction, the underlying data is available on request: response-level entity mentions, both leaderboards with their brand-only companions, full unfiltered metrics for every entity in all eight cuts, the per-model and per-question-type breakdowns, the origin research trail with a source URL per entity, the full domain-level source list, the trend file against June, and the methodology and data dictionary.

All figures reflect a single point-in-time run (11 September 2026) and are subject to the limitations in Chapter 12. The prior edition, June 2026, was restated on 14 September 2026 — 12 same-domain duplicates merged and 10 origins corrected — so any comparison against a copy of the June page saved before that date will not reconcile.

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.

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 →