1,124 responses, 5 models, 45 Turkish questions: the baseline AI-visibility snapshot of Turkey's pet products market

Vertical: Pet products (evcil hayvan ürünleri), Turkey
Method: A single point-in-time study of 1,124 responses across five large language models (Claude, Gemini, ChatGPT, Perplexity, Grok), each asked 45 Turkish-language questions five times
Collection window: 28 August 2026

Key terms: AI visibility, AI Visibility Score, pet products, evcil hayvan ürünleri, kedi maması, köpek maması, kedi kumu, su pınarı, Turkish (yerli) pet brands, LLM brand recommendations, ChatGPT/Gemini/Claude/Perplexity/Grok brand recommendation, Generative Engine Optimization (GEO).

This report reproduces the real Turkish-language questions people ask AI assistants about cat food, dog food, litter, grooming and pet equipment, and measures how often, in what order, and across how many models five large language models name each tracked entity. Because the same questions can be read with or without an origin restriction, and because pet products divides into genuinely different product categories, the entities are ranked in ten separate cuts, defined in Chapter 2. It is a single point-in-time study and should be read within the limitations in Chapter 11.

How to read this report: visibility is not quality

Every number here measures whether an AI assistant names a tracked entity, and how early. It does not measure whether the food is nutritionally sound, whether the litter clumps, whether the price is fair, or whether the recommendation was appropriate for a particular animal. A brand can be highly visible and mediocre, or excellent and invisible. Read the leaderboards as a map of findability in a new discovery channel — nothing more. Nothing in this report is veterinary advice.

1. Executive Summary

One-sentence takeaway: Reflex leads Turkey’s pet shelf overall, but the market splits cleanly in two — the local cut has no foreign qualifier, while the origin-neutral cut is foreign-led, especially by imported premium food and specialist litter/grooming.

• Reflex leads decisively. It tops the general leaderboard at 87.92, appearing in 25.53% of all 1,124 responses — 287 answers — and named by all five models. Royal Canin (82.60, 15.66%) and Felicia (68.52, 15.30%) follow, with Trendyol (68.18) and VanCat (63.98) completing the top five.
• 28 entities qualified for the general ranking. Of 225 tracked entities, 28 cleared the 5% mention threshold across all 45 questions: 18 Turkish and 10 foreign. Their average score is 56.58, and 24 of the 28 are named by all five models.
• The origin clause changes the market, not just the emphasis. On the 22 questions that explicitly ask for yerli brands, not one foreign entity qualifies and Turkish entities take 98.39% of qualifier mentions. On the 23 origin-neutral questions, 18 of the 30 qualifiers are foreign, foreign entities take 66.21% of qualifier mentions, and the top seven — Royal Canin, Farmina, Acana, Science Plan, Ever Clean, Furminator, Orijen — are all foreign before Reflex appears at #8.
• The split follows product categories. Of the 18 foreign qualifiers in the origin-neutral cut, seven are dry food and four are litter or grooming. Turkish brands take the top two places in Litter & Grooming (VanCat 87.05, Proline 85.48), while Turkish entities take the top three in Equipment & Accessories (Xenon Smart 82.02, Trendyol 76.56, Catrevi 75.07) — with Trendyol a marketplace rather than a brand. Premium imported nutrition is where domestic visibility thins.
• Four cuts, four leaders. Royal Canin leads Pet Food at 91.15 despite being outnamed by Reflex on volume (34.59% against 43.06%); Xenon Smart leads Equipment & Accessories; VanCat leads Litter & Grooming at a 41.07% mention rate; and Reflex leads the general, local and Cross-category cuts. Reflex is the only entity to rank in all ten cuts.
• The models differ more in how much they say than in whom they name. Grok produces 10.02 tracked-entity mentions per answer against ChatGPT’s 2.42, and it is the only model that never returned an answer without a tracked entity in it. Reflex is the most-named entity for three of the five models; Claude puts Royal Canin first, and Grok answers with marketplaces — Trendyol in 72.0% of its answers, ahead of any brand.
• One model barely searched. Overall 88.61% of answers were web-grounded, but that average conceals a split: Claude, Perplexity and Grok grounded 100% of their answers and ChatGPT 99.56%, while Gemini grounded only 43.3%. 127 of the 128 ungrounded responses in the entire study are Gemini’s, which makes a within-model search-versus-memory comparison possible (Chapter 6).
• Sources are retailer-led. Retailer sites account for 33.04% of citations against 19.39% “other”, 17.50% editorial, 15.49% brand-owned, 7.02% forum/social and 6.91% marketplace. Petlebi’s main domain is cited by 280 responses, while its community subdomain is cited by 103 responses; those response sets may overlap.

Why it matters. Shoppers are shifting discovery from search engines and marketplaces to AI assistants. Independent survey work suggests roughly four in ten Turkish internet users were using ChatGPT monthly in mid-2025 (GWI), while TÜİK’s first measurement of the subject put self-reported generative-AI use at 19.2% of people aged 16–74 in 2025 — different definitions, both indicating a channel that now exists at scale. Pet products is a repeat-purchase category with an unusually engaged and research-driven customer base, and this edition asks a narrow question about it: when someone asks an assistant in Turkish which food, litter or water fountain to buy, whose name comes back. This is the first (baseline) snapshot of that shelf. One reminder: the numbers describe visibility and information availability, not which brands are best.

2. Methodology

One-sentence takeaway: Five large language models were asked 45 Turkish questions five times each, with no system prompt, producing 1,125 responses of which 1,124 were analysed. Entities were discovered from the answers themselves, human-reviewed, and ranked with a three-component (45/30/25) score computed separately within each of ten cuts.

2.1 Scope and models

The study queried five large language models directly via API with identical questions. All models received only the user question. No system prompt was used, temperature was not set manually (provider default), and no region or locale parameter was defined. Because the questions are in Turkish, the models infer the Turkish market from language alone.

Scale: 45 questions × 5 repeats × 5 models = 1,125 responses. Every cell of the grid existed exactly once, with no duplicates and no errors. One response — Gemini’s, on a Turkish-brand litter question — returned a completed status with an empty body and was excluded from every denominator, which is why Gemini’s per-model denominator is 224 against 225 for the other four models. That leaves 1,124 responses analysed, of which 1,063 named at least one tracked entity; 61 named none. Single collection window, 28 August 2026.

Denominators are computed per question-and-model cell, never as questions × models × repeats. Every published rate in this report is stated against the denominator of its own cut.

2.2 Web-search configuration

No model was instructed to search or not to search. Each was left to its own default behaviour, and whether it retrieved live web content was recorded per response rather than assumed. Across the study, 996 of 1,124 responses (88.61%) carried citations or search results.

That average is not evenly distributed. Claude Haiku, Perplexity Sonar and Grok grounded every answer, ChatGPT 5.6 Luna Pro grounded all but one, and Gemini 3.6 Flash grounded 97 of its 224 answers (43.3%). Gemini’s remaining 127 answers were produced from what the model already knew. Its visibility figures therefore reflect parametric memory more than current web content, and should not be read as equivalent to the other four. Chapter 6 examines what changes when it searches.

Two providers also returned citations in awkward shapes and were handled separately: Gemini returned redirect URIs for all of its references, so its cited domains were read from the search-result titles, and Grok returned citation titles as bare numerals, so its domains were read from the URL.

2.3 Questions and the ten cuts

The 45 questions are real Turkish-language phrasings of the kind people type into an assistant — “en iyi kedi kumu hangisi”, “kedim kısırlaştı, hangi mama markasına geçeyim?” — rather than survey items. No question was rewritten for this edition. Each carries three tags:

• Scope: 22 questions explicitly ask for Turkish or yerli brands; 23 place no origin restriction.
• Question type: 15 discovery (“which brands are good”), 15 attribute (“low-dust, good-clumping litter”), 15 use case (“my cat has been neutered, which food should I switch to”).
• Segment: Pet Food (17), Equipment & Accessories (16), Litter & Grooming (9), Cross-category (3).

Because a single ranking would hide more than it shows, entities are ranked in ten separate cuts. Every cut has exactly one denominator, and scores are scaled within the cut:

Design balance. Globally the design is even: 22 local against 23 open, and 15 questions of each type. Within segments it is not, and this constrains what can be compared:

Because of that imbalance, the number of distinct entities qualifying in one segment should not be compared with another. Segment leaderboards are read individually, not against each other.

2.4 Entity discovery, resolution and classification

No entity list was supplied in advance. Entities were discovered from the responses themselves: 22,943 raw candidate n-grams were reduced to 1,070 above a three-response floor by a stoplist and a lowercase-usage test, then rooted to collapse fragments and product-line variants. All 1,070 were reviewed by hand — 245 keep, 820 drop, 5 merge — and sub-brands whose parent was also in the study were merged into the parent. The final dictionary holds 225 entities, producing 5,713 tracked-entity mentions across 1,063 responses. No entity in the dictionary finished with zero matches.

Matching is alias-based and word-boundary anchored, applied to URL-stripped text so that a link target cannot inflate an entity’s count, and folded with an explicit Turkish translation table. Turkish suffixes attached with an apostrophe are matched; suffixes attached directly without one are not, which undercounts slightly and never overcounts.

Four resolution decisions are worth stating plainly, because each one moved the tables:

• Turkish case folding. Nine entity pairs had split into dotted and dotless variants — PETKIT folds to petkit under Turkish rules and had separated from Petkit with zero overlap between the two response sets. All nine were folded. The same pass recovered Pro Line, which had been dropped on 9 responses while Proline was kept.
• A name collision between two companies. Champion covers a Turkish brand and Canadian Champion Petfoods, the parent of Acana and Orijen. Of the 74 responses using the word, 14 said “Champion Petfoods” and 3 used the bare word alongside Acana or Orijen. The entity was split at the matcher level, so both appear separately in this report and rank on different shelves: Turkish Champion qualifies in the local cut, Champion Petfoods only in the origin-neutral one.
• A merge target that was not yet an entity. Hurley had been merged into “West Paw”, which had never been kept in its own right. West Paw was promoted to a canonical entity inheriting West, Hurley and Zogoflex.
• Split entities detected by identical response sets. Fakir and PetPard appeared in exactly the same 18 responses, a signature of one brand recorded twice. They were merged, along with three lower-count pairs.

As a backstop, every domain cited by 100 or more responses was checked against the dictionary once forum and editorial sources were excluded. No top-tier entity was missing.

Classification. Each entity carries an origin, a type and a scope fit. Origin follows the trademark, not the operator: a foreign mark run under Turkish licence is Foreign. Where the trademark owner could not be verified — typically marketplace-native brands with no corporate registration trail — the entity is marked Unknown, kept in every table, and excluded from origin shares. 18 entities are in that position, and none of them qualifies in the general cut.

Entity types are Brand, Sub-brand, Licensed-brand, Retailer, Brand+Retailer and Marketplace. Scope fit marks whether an entity is core to pet products (201 entities) or adjacent to it (24) — Xiaomi, for instance, sells pet water fountains but is not a pet company.

2.5 AI Visibility Score (0–100)

Three things matter for visibility: how often an entity is named, how early it appears, and how many models name it. The score combines all three.

Each component is scaled within its cut against the highest qualifying value in that cut, so 100 means “the best in this cut”, not “perfect”. An entity qualifies for a leaderboard when its mention rate reaches 5% within that cut.

Worked example — Reflex, general cut. Reflex was named in 287 of 1,124 responses, a mention rate of 25.53%, which is the highest in the cut, so its mention component is 100.0. Its MRR is 0.4814 against the cut maximum of 0.8058 (held by Royal Canin), giving a position component of 59.74. It appears in all five models, so its breadth component is 100.0. The composite is 0.45(100.0) + 0.30(59.74) + 0.25(100.0) = 87.92.

The leader does not reach 100 because no single entity tops all three components — that is the expected behaviour of component scaling, not a defect. Because the maxima come from qualified entities only, an unqualified entity can exceed 100 on an individual component: 61 of the 1,119 unqualified rows do. They are reported, never clipped.

Scores cannot be compared across cuts. A score of 80 in the general cut and a score of 80 in Litter & Grooming are different absolute quantities, because each is scaled against a different leader. Comparisons across cuts must use the raw layer — mention rate, MRR and breadth — which is carried on every table in this report for exactly that reason.

2.6 Reading origin figures

Because 22 of the 45 questions constrain the answer to Turkish brands, the general cut is structurally biased toward domestic entities. Origin is therefore analysed on the open cut only, with the local figure stated alongside it every time. Chapter 8 does this in full; nowhere in this report is a Turkish-versus-foreign claim made on the general cut alone.

2.7 Neutrality and self-exclusion

This is a neutral market-wide study with no focus brand and no client. Herm.io has no commercial relationship with any entity ranked here, and all citations to Herm.io’s own domains were excluded before analysis. Marketplaces are ranked where the models name them as answers, because that is how the answers actually read; every leaderboard also has a marketplace-excluded variant that removes them.

3. Overall Visibility Leaderboards

One-sentence takeaway: Reflex leads the general and local cuts, but the open cut is a different market with a different leader — the three headline cuts do not tell the same story.

As a reminder, these rank visibility, not product quality, nutrition, safety or value.

3.1 General leaderboard: 28 qualified entities

Average score of the 28 qualified entities: 56.58. Origin split: 18 Turkish, 10 foreign.

Reading. Reflex tops the table at 87.92, named in 25.53% of answers — 287 of 1,124 — by all five models, and it leads on volume by a clear margin: Trendyol at 18.59% is the only other entity above 17%. Royal Canin is second on a much lower mention rate (15.66%) but the highest position score in the cut: when it is named, it is named early, with an MRR of 0.8058 against Reflex’s 0.4814. These are two different kinds of visibility, and the score deliberately rewards both.

24 of the 28 qualified entities are seen by all five models. The four exceptions are Mystic and Hepsiburada on four models each, and BonaCibo and Spectrum on three.

Two marketplaces rank in the top fifteen. Trendyol at #4 and Hepsiburada at #14 are not brands; they are where the models send people to buy. They are kept in the headline ranking because that is how the answers read, and §3.3 shows the table without them. Chapter 5 shows how much of their position comes from a single model.

The foreign entries cluster. Six of the ten foreign qualifiers are dry food — Royal Canin, Farmina, Acana, Science Plan, Orijen and Brit Care — and the sub-brands among them (Science Plan under Hill’s, Brit Care under Brit) are named by the models more often than their parents are. Ever Clean at #10 is litter, and KONG, Ferplast and Trixie are equipment. There is no foreign equivalent of Reflex in this data: no foreign brand spans categories the way the domestic leader does.

3.2 The same cut, ranked by mention rate

Identical data, ranked by volume alone rather than by the weighted score:

The reordering is instructive. Hepsiburada climbs from #14 to #3 and Trendyol from #4 to #2, because both are named often but late — their MRRs are 0.2074 and 0.2796 against Royal Canin’s 0.8058. BonaCibo rises from #17 to #8 on the same logic, helped further by being named by only three models. Moving the other way, Royal Canin falls from #2 to #4 and Ever Clean, KONG and Catrevi leave the top ten entirely, because all three are named comparatively early but rarely.

Neither ranking is more correct than the other: the first asks how prominently is this entity named, the second asks how often. The AI Visibility Score blends the two deliberately, and readers should hold both pictures in mind for the rest of the report.

3.3 Marketplace-excluded view

Two of the 28 qualified entities are marketplaces rather than brands. Removing them gives a view of the same cut without marketplaces; retailers remain:

Removing Trendyol and Hepsiburada promotes VanCat to #4 and Farmina to #5 but changes nothing at the top: Reflex, Royal Canin and Felicia hold first through third. Retailers stay in — Petlebi and Pet Garden are retailers, not marketplaces, and they qualify in cuts where the models recommend a shop rather than a brand. In a market where a large share of discovery happens on retailer sites, stripping retailers out would remove part of the shelf being measured.

3.4 Local leaderboard: 25 qualified entities (Turkish-only questions)

When the 22 questions that explicitly ask for Turkish or yerli brands are isolated, the shelf does not merely tilt — it clears out:

Average score of the 25 qualified local entities: 54.10. Origin split: 24 Turkish, 0 foreign, 1 unknown.

Reading. Reflex leads at 96.17 on a 34.43% mention rate — its strongest showing in any headline cut — with Felicia (76.08), VanCat (74.29), Mystic (70.01) and Catrevi (69.09) following. Not a single foreign entity clears the threshold, and Turkish entities take 98.39% of the cut’s qualifier mentions. The only non-Turkish qualifier is Pisi Pisi at #25, whose trademark owner could not be verified and which is therefore marked Unknown rather than foreign.

That is a stricter result than it may appear. Foreign brands that a Turkish shopper might plausibly think of as domestic — long-established importers with Turkish operations — do not slip through. Where an origin filter is stated, the models respect it.

Two entries need a caveat. Hepsiburada ranks #18 on 102 responses but is named by only two of the five models, and BonaCibo (#9), Anatolian (#15), Spectrum (#19), Micho (#20), Oligatto (#22), Enjoy (#23), Pet Garden (#24) and Pisi Pisi (#25) are each named by three. Breadth is doing visible work at the bottom of this table, and small differences there should not be over-read. Every low-breadth entry in the report is collected in one place in Chapter 11.

4. Segment Leaderboards

One-sentence takeaway: The four product segments produce four different leaders and four different origin balances — pet products is not one shelf but several.

Segment cuts are scaled independently, so a score in Pet Food cannot be compared with a score in Litter & Grooming. Compare mention rates instead.

4.1 Pet Food (17 questions, 425 responses)

30 qualified entities. Origin split: 22 Turkish, 8 foreign. Average score 50.03.

The largest segment is also the most contested. Royal Canin leads at 91.15 while being named less often than Reflex — 34.59% against 43.06% — because its MRR of 0.8092 is the highest in the segment by a wide margin. When a model recommends food, Royal Canin is very often the first name in the answer; Reflex appears in more answers but further down them.

Six of the top nine are foreign: Royal Canin, Farmina, Acana, Science Plan, Orijen and Brit Care. The Turkish entries that hold their own — Reflex (#2), Felicia (#3), BonaCibo (#8), Mystic (#10), Enjoy (#11) — are mainstream everyday brands rather than premium ones. This is the segment where the origin story in Chapter 8 originates.

4.2 Equipment & Accessories (16 questions, 400 responses)

20 qualified entities. Origin split: 8 Turkish, 10 foreign, 2 unknown. Average score 56.29.

Xenon Smart leads at 82.02 on a 22.50% mention rate, ahead of Trendyol (76.56) and Catrevi (75.07). This is the narrowest field in the study — 20 qualifiers on 400 responses — and the one where the models most often answer with a shop rather than a brand: Trendyol and Hepsiburada take second and sixth, and Amazon qualifies at #20.

One scaling artefact matters here. KONG ranks #5 on 26 responses — a 6.50% mention rate, the lowest in the segment’s top ten — carried by an MRR of 0.9167, the highest position figure anywhere in this study. Because position components are scaled against the cut maximum, KONG is the yardstick: every other entity’s position score in this segment is expressed as a fraction of an entity named in one answer in fifteen. KONG is a real and well-known toy brand, but its rank here reflects the score’s design as much as its share of the conversation.

Two entities in this segment carry an unverifiable trademark owner and are marked Unknown: Haustier at #9 and Petyco at #17. Both are treated as neither Turkish nor foreign in Chapter 8.

4.3 Litter & Grooming (9 questions, 224 responses)

28 qualified entities. Origin split: 17 Turkish, 7 foreign, 4 unknown. Average score 45.69.

The most domestically-owned segment, and the one with the highest mention rates in the study. VanCat leads at 87.05 on a 41.07% mention rate, with Proline named in 49.11% of the segment’s responses — nearly half — though a weaker MRR of 0.4184 puts it second on score. Sandy (#7) and Oligatto (#9) add two more Turkish litter brands to the top ten.

Foreign presence here is specialist rather than general: Ever Clean (#3) and Sanicat (#5) in clumping litter, Furminator (#4) in grooming tools with an MRR of 0.8110. These are category specialists that Turkish manufacturers do not directly replace.

This is also the segment with the smallest denominator outside Cross-category, and the one whose question mix is most uneven — nine questions, of which only one is a use-case question. Its distinct-entity count should not be read against the other segments.

4.4 Cross-category (3 questions, 75 responses)

Read this section as indicative only

Cross-category contains three questions and 75 responses. At that denominator an entity needs just four mentions to clear the 5% threshold, which is why 51 entities “qualify” here — more than in Pet Food, a segment with nearly six times the responses. This is a small-denominator effect, not a finding about the market. Thirty-two of the 51 qualifiers are named by fewer than three models and 27 appear in fewer than ten responses. The top ten is shown for completeness; no interpretation should be drawn from it, and this cut is excluded from the segment comparison below.

4.5 Segment comparison

Three segments, three leaders, and three different origin balances. Equipment & Accessories is the only segment where foreign qualifiers outnumber Turkish ones — and its leader is nonetheless Turkish. Pet Food has the most Turkish qualifiers in absolute terms but hands the top place, and six of the top nine, to imported brands. Litter & Grooming is the segment domestic manufacturers most clearly hold.

Only one entity leads more than one cut anywhere in this study, and only Reflex ranks in all ten.

5. Differences Between Models

One-sentence takeaway: The models broadly agree on the leading names and disagree sharply about how many entities to give and whether a marketplace counts as an answer.

5.1 Per-model behaviour summary

The clearest split is how much the models say. Grok produces 10.02 tracked-entity mentions per answer against ChatGPT’s 2.42 — a four-fold difference — with a median of nine tracked entities per answer, and it is the only model that never returned an answer containing no tracked entity at all. ChatGPT gives short, committed answers: 544 mentions in total, 19 answers with no tracked entity in them, and the narrowest median. Perplexity and Gemini sit between.

Breadth of answer and breadth of vocabulary come apart. Grok also has the widest vocabulary at 169 distinct entities, but Gemini names 1,279 mentions drawn from only 130 distinct entities — long answers from a comparatively short list — while Claude gives short answers (714 mentions) from 107.

Grounding depth differs too, and not in the direction the grounding rate suggests. Perplexity cites 11.87 domains per grounded answer against ChatGPT’s 2.81. Both grounded essentially every answer; they read very different amounts of the web before replying.

5.2 Each model’s most-named entities

Reflex is the most-named entity for three of the five models — Gemini (32.59%), ChatGPT (16.89%) and Perplexity (25.33%). Claude puts Royal Canin first, on a modest 12.89%, which reflects Claude’s generally sparse answers rather than a different view of the market. Grok is the outlier, and §5.3 explains why.

Underneath the leaders the models diverge more than the headline table suggests. Claude names Anatolian and Econature in 8.00% of its answers apiece; neither qualifies in the general cut. Perplexity is the only model to put a retailer, Petlebi, in its own top twelve. Gemini alone names Temizmama in 12.50% of its answers.

5.3 The Grok marketplace effect

Grok answers Turkish pet questions by naming shops. Trendyol appears in 72.0% of Grok’s answers and Hepsiburada in 69.33% — both ahead of any brand — and marketplaces account for 15.21% of all Grok’s tracked-entity mentions against 1.26% of Claude’s.

That single behaviour largely explains two entries in the general leaderboard. Of Trendyol’s 209 mentions across the whole study, 162 are Grok’s (77.5%); of Hepsiburada’s 187, 156 are Grok’s (83.4%). Both would still qualify without Grok, but neither would sit at #4 and #14.

This is a real observation about how one widely-used assistant answers, not an error to be corrected — but it should be held in mind wherever a marketplace appears in this report, and it is the main reason every leaderboard has a marketplace-excluded variant.

6. Search or Memory

One-sentence takeaway: Gemini answered 43.3% of its questions with a live web search and the rest from memory, and the two modes produce almost entirely different brand lists — but Gemini also chose when to search in a way that is tangled up with the question type.

Four of the five models grounded essentially every answer, so an all-model grounded-versus-ungrounded comparison would be a comparison between Gemini and everyone else. 127 of the 128 ungrounded responses in the study are Gemini’s. Only a within-Gemini comparison is defensible, and that is what follows.

6.1 What changes when Gemini searches

The contrast is stark. Of the twelve entities that gained most from searching, eleven are Turkish and one is of unverified origin — Xenon Smart, Catrevi, Tailpetz, Doggie, ATA HOME and Bilicra go from being named in 9–14% of Gemini’s searched answers to being named in none at all when it answers from memory. Of the twelve that gained most from answering off memory, all twelve are foreign, and eight are dry food: Farmina rises from 3.09% to 31.50%, Royal Canin from 2.06% to 26.77%, and Science Plan, Orijen and Brit Care go from zero to over 17%.

6.2 Why this cannot be read as cause and effect

The confound. Gemini did not decide at random when to search. It grounded 69.72% of the questions that asked for Turkish brands and only 18.26% of the origin-neutral ones. So its “searched” answers are disproportionately answers to questions that asked for Turkish brands in the first place, and its “memory” answers are disproportionately answers to origin-neutral questions — which, as Chapter 8 shows, are foreign-led regardless of grounding. The origin pattern in the tables above cannot be separated from the question mix with this data, and no causal claim is made here.

6.3 The uneven searching is itself the finding

Set the brand tables aside and the grounding pattern stands on its own. Gemini searched most on Equipment & Accessories (63.75%) and least on Pet Food (30.59%); most on attribute questions (56.00%) and least on use-case questions (31.08%); and far more when asked for Turkish brands than when not.

That is a coherent pattern rather than noise. The questions Gemini answered from memory are the ones where a general-purpose model plausibly believes it already knows the answer — global food brands, open-ended situational advice. The questions it looked up are the ones where a model has reason to think its memory is thin: local brands, and specific product attributes like dust and clumping. A model’s confidence in its own knowledge is what determines whether a brand gets a live-web chance to appear, and Turkish brands here are visible mainly on the questions where the model doubted itself.

7. Question-Type Ownership

One-sentence takeaway: Open-ended discovery questions spray names widely, attribute questions concentrate hard on a short list, and marketplaces do best exactly where the question is most specific.

Discovery — 44 qualifiers. “Which cat food would you recommend?” produces the widest field in the study outside Cross-category. Reflex leads at 88.52 on a 31.47% mention rate, Royal Canin follows at 85.28, and the origin split among qualifiers is almost exactly even: 20 Turkish, 21 foreign. Asked an open question, the models reach for a long list and reach abroad as readily as at home.

Attribute — 23 qualifiers, the narrowest field. “Low-dust litter that clumps well, which brand?” halves the field. And it produces the single strangest row in this study: Trendyol leads the attribute cut at 83.70, ahead of Reflex, with Hepsiburada sixth. Asked for a specific product property, the models answer more often with a place to compare products than with a brand that has the property. Turkish entities take 15 of the 23 qualifying places here.

Use case — 28 qualifiers. “My cat has been neutered, which food should I switch to?” produces the most decisive answers of the three. Royal Canin leads at 94.64 with an MRR of 0.8522, the highest of any leader in any cut: when the question describes a situation, a specialist food is named first and named confidently. KONG (#5, MRR 0.9133) shows the same pattern for chew toys.

The pattern across the three is consistent: the more specific the question, the shorter the list of entities that come back, and the earlier the leading name appears in the answer. For a brand, discovery questions are the widest door and use-case questions the narrowest — but a brand that gets through a use-case door arrives at the top of the answer.

8. Open Market: Turkish vs Foreign

One-sentence takeaway: Turkish pet brands are highly visible when the question asks for them and much less visible when it does not — the two shelves have different leaders, different composition and different product centres of gravity.

This chapter is read on the open cut — the 23 questions that place no origin restriction — with the local figure stated alongside every time, for the reason given in §2.6.

8.1 Origin share

The two cuts describe two different markets. Ask for Turkish brands and no foreign entity clears the threshold: 24 of the 25 qualifiers are verified Turkish and one is of unknown origin. Ask without saying anything about origin and roughly two-thirds of qualifier mentions are foreign — and the top seven places are foreign without interruption.

8.2 Open-market top 15 entities (any origin)

Royal Canin scores 100.0 here — the only entity anywhere in this study to top all three components of its cut simultaneously. It is named in 27.48% of origin-neutral responses with an MRR of 0.8298 across all five models. Reflex, the overall study leader, is eighth.

8.3 The pattern: a market split by product, not just by origin

The foreign share is not spread evenly across the shelf. Of the 18 foreign qualifiers in the open cut, seven are dry food (Royal Canin, Farmina, Acana, Science Plan, Orijen, Brit Care, Champion Petfoods), four are litter or grooming specialists (Ever Clean, Furminator, Sanicat, Lindocat), and the rest are equipment brands (Xiaomi, KONG, Ferplast, Trixie, Catit, Petkit, Petlibro).

The twelve Turkish qualifiers cluster elsewhere. Six are cross-category brands — Reflex, Proline, VanCat and others that appear in food, litter and accessories alike — and only four are food brands. The domestic industry’s visibility in this data rests on breadth across everyday categories; the foreign presence rests on depth in premium nutrition and specialist grooming.

That distinction matters for how the origin gap should be read. It is not that models decline to name Turkish brands when unprompted; Reflex, Trendyol, Proline and Xenon Smart all qualify in the open cut. It is that in the highest-volume conversation in this vertical — what to feed a cat or dog — the names the models reach for first are imported. When the same models are asked about litter or a water fountain, the answer is domestic.

9. The Discovery Ecosystem: Where AI Learns About Brands

One-sentence takeaway: This is a retailer-taught market — pet shops, not pet brands, are the most-cited sources, and one retailer group appears in roughly a third of the study’s answers.

585 distinct domains were cited across the study.

9.1 The 12 most-cited domains

petlebi.com is cited by 280 responses — 24.91% of the entire study, while its community subdomain sosyal.petlebi.com is cited by 103 responses. Those response sets may overlap. Six of the top twelve domains are pet retailers; two are marketplaces; one is a general forum (Ekşi Sözlük, cited by 112 responses); and exactly one — catrevi.com, at #8 — is a brand’s own site.

9.2 Source-type mix

9.3 The retailer finding

Retailer sites take 33.04% of citations across 170 domains — more than double the brand-owned share of 15.49%. Add marketplaces and the commerce share reaches 39.95%. Editorial sources take 17.50% from just 55 domains, a concentrated set of Turkish consumer-advice sites. Forum and social sources take 7.02%, of which Ekşi Sözlük and Petlebi’s community subdomain are the bulk.

“Other” sits at 19.39% and is the most fragmented type in the study: 262 domains, none individually significant, together contributing less than the retailer category does from 170.

The practical reading is that in this vertical the models are learning about brands largely from the shops that sell them rather than from the brands themselves. Only eleven brand-owned domains appear in the top fifty at all, and the highest of them, catrevi.com, sits eighth. A brand’s visibility in these answers is mediated by how it is described on retailer category pages and in comparison articles — copy that the brand mostly does not control.

10. What the Patterns Suggest

These are readings of the data, not recommendations, and each is a hypothesis the next edition can test.

1. Two shelves need two strategies. A Turkish brand that is highly visible on yerli questions and absent from origin-neutral ones is winning only the smaller conversation. The open cut is where the discovery volume sits for anyone not already searching for a domestic brand, and it currently belongs to imported food.

2. Category breadth is what carries a domestic brand. The Turkish entities that survive the origin-neutral cut are the ones the models associate with several categories at once. Depth in one category, at domestic price positions, does not appear to travel.

3. Retailer copy is doing the describing. With brand-owned sites at 15.49% of citations and retailers at 33.04%, how a product is written up on a shop’s category page is likely to matter more for AI visibility than how it is written up on the brand’s own site.

4. Specific questions reward specialists — and marketplaces. Attribute questions concentrate on a short list, and that list is topped by two marketplaces. A brand with a genuinely distinctive product attribute has a narrow but unusually early door to walk through here; a brand without one is likely to be replaced by a link to a shop.

5. Whether a model searches determines who it can name. Within Gemini, the brands that appear only in searched answers are almost all domestic, and the ones that appear only in unsearched answers are entirely foreign. Established international brands are already inside the model’s memory; smaller and local brands have to be found on the live web. That places a premium on being findable and clearly described at the moment of retrieval.

11. Limitations and Notes

1. Single collection window. All data was collected on 28 August 2026. This is a snapshot, not a trend, and AI answers change as the underlying web changes.
2. One response was excluded. Of 1,125 collected, one Gemini response returned a completed status with an empty body and was removed from every denominator. Gemini’s denominator is 224; the other four models’ is 225.
3. Gemini’s 43.3% grounding rate means 127 of its 224 answers reflect model memory rather than retrieval — 11.3% of the whole corpus. Its figures are not equivalent to the other four models’, and the Chapter 6 comparison is confounded by which questions it chose to search on.
4. Cross-category is indicative only. Three questions and 75 responses produce 51 qualifiers, purely a small-denominator effect. It is excluded from the segment comparison and should not be read as a finding.
5. Within-segment question types are unbalanced. Litter & Grooming has one use-case question; Cross-category has no attribute question. Distinct-entity counts must not be compared across segments.
6. 42 qualifying rows across all ten cuts are named by two models or fewer, and 32 of those are in Cross-category. Outside it the cases are: Hepsiburada in the local cut (#18, two models) and in Litter & Grooming (#22); Amazon in Equipment & Accessories (#20) and in the attribute cut (#23); Temizmama in Pet Food (#30); TofuCat in Litter & Grooming (#28); Mystic (#29) and Spectrum (#43) in the discovery cut; Anatolian (#20) and Trendyol (#22) in the use-case cut. Each is flagged in its own section; all are listed here so the pattern is visible in one place.
7. 27 qualifying rows appear in fewer than ten responses, all of them in Cross-category, where four mentions clear the threshold.
8. Position components in Equipment & Accessories are scaled against KONG, which holds the cut’s highest MRR (0.9167) on 26 responses. A scale anchored to a low-volume entity compresses everyone else’s position score in that segment.
9. 18 entities have an unverifiable trademark owner despite a documented Turkish operator — typically marketplace-native brands with no corporate registration trail. They are kept in the tables, marked Unknown, and excluded from origin shares. None qualifies in the general cut, so the origin picture is barely affected, though four appear in segment leaderboards.
10. Sub-brands under-count structurally where a model names the parent instead. Science Plan and Brit Care are ranked in their own right because the models name them directly; Hill’s and Brit are not.
11. Marketplaces are not brands. Trendyol, Hepsiburada and Amazon rank on the main tables because the models name them as answers. The marketplace-excluded variants strip exactly these three and retain all retailers.
12. Turkish suffix matching is conservative. Suffixes attached with an apostrophe are matched; those attached directly are not, which undercounts slightly and never overcounts.
13. The question set is cat-weighted. 29 of the 45 questions concern cats, 9 concern dogs and 7 are neutral, which is one reason the dictionary holds 51 cat-specific entities against 15 dog-specific ones, and why cat-specific brands take 945 mentions to dog-specific brands’ 193. Four cat-specific brands qualify in the general cut and no dog-specific brand does — a gap that reflects the questions asked as well as the market, and one a differently-balanced question set would need to re-test.
14. ai_score measures AI visibility only. Not quality, not nutrition, not safety, not popularity, not sales.
15. Never compare a scaled score across cuts or editions. A score of 100 in one cut is a different absolute quantity from 100 in another, and the scale moves whenever the leader moves. Use mention rate, MRR and breadth for any comparison.

Frequently Asked Questions

12. Appendix & Data

12.1 All questions (45) and their categories

Distribution: 15 discovery, 15 attribute, 15 use case; 22 local and 23 open; across four product segments (17 / 16 / 9 / 3).

12.2 All tracked entities (225)

12.3 Data availability

The study tracked 225 entities; of these, 28 cleared the 5% threshold in the general cut, 25 in the local cut, 30 in the open cut, and 82 in at least one of the ten cuts. To support scrutiny and reproduction, the underlying data is available on request: response-level entity mentions with grounding flags, all ten qualified leaderboards and their marketplace-excluded variants, full unfiltered entity metrics, per-model and per-question-type breakdowns, the open-market breakdown, the full 585-domain source list, the entity dictionary with its decisions log, and the methodology and data dictionary.

All figures reflect a single point-in-time run (28 August 2026) and are subject to the limitations in Chapter 11. This is the first (baseline) measurement; the study will be repeated quarterly.

About Herm.io & disclosure

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

This edition is neutral and market-wide: there is no focus brand, no client, and no commercial relationship with any entity ranked here. All citations to Herm.io’s own domains were excluded before analysis. Entities are ranked purely on how the five models answered, and no entity was added to or removed from the dictionary on any basis other than the documented resolution rules in §2.4.

Nothing in this report is veterinary, nutritional or purchasing advice. Visibility in AI answers is not evidence that a product is suitable for a particular animal.

Report author: Mert Can Elkaya
Data: Herm.io AI Visibility Database (direct LLM querying)
Edition: 1 (baseline), August 2026
Collected: 28 August 2026

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 →