Which sites does AI cite in Türkiye?

In Türkiye, AI engines answering buyers' Turkish questions cite marketplaces, social media and forums most, and brands' own sites rarely. GeoMono is a platform that measures how visible brands are in AI answers. This post gives the sites shown in the 16,703 answers we measured for 15 brands between 16 August and 31 October 2026, by engine and by sector. Every figure carries what it was counted over.

The short answer: Trendyol, Hepsiburada, Instagram, YouTube and Reddit lead the list. But the real finding is not the ranking, it is the spread: 141,498 source links went to 19,844 different sites, and no single site reached even 7% of answers.

What did we measure?

We asked five AI engines (ChatGPT, Gemini, Claude, Perplexity and the AI Overviews in Google's search results) Turkish questions written the way a buyer would ask them, without the brand's name, and recorded the source links in every answer.

Answers we could not measure were not counted: 317 calls returned an error, 24 never ran, and the source list of 16 answers could not be read. For 130 questions Google showed no AI Overview in its search results; there was no answer to read, so those were not counted as "cited nothing" either. Counting an error as "no sources" would report a provider outage as a finding.

Reach is the number of the 16,703 measured answers in which a site was cited at least once. One answer linking the same site ten times counts once.

How did we keep client information out?

A site entered a table only if it met two conditions. First, it was cited in the answers of at least 5 different brands. Second, no more than half of the answers citing it came from one brand's questions. The second rule matters: even if each of five brands saw a bank's site once, if most of its answers came from one brand's questions, the row describes that brand, not the landscape.

Of the 19,844 sites, 263 appeared for at least 5 brands, and 174 of those also passed the second rule. The tables below are built from those 174 sites only. Gemini returns its sources as Google redirect addresses; we read the site's name from the link's title. The 1,877 links from AI Overviews and Gemini that pointed back to Google itself did not resolve to a site; we counted them and kept them apart.

Which sites were cited most?

Read by type, the table falls into four groups:

• Marketplaces and price comparison: Trendyol, Hepsiburada, Akakçe, Cimri. When an engine recommends a product, it often shows the page where it is sold.
• Social media and video: Instagram, YouTube, Facebook, Reddit. Their weight varies a lot by engine (below).
• Forums and user content: Ekşi Sözlük, DonanımHaber, Technopat, Şikayetvar, Onedio. Places where buyers write about their own experience.
• Reference and press: English Wikipedia, Hürriyet, and further down DergiPark (215 answers, 1.3%) and Turkish Wikipedia (171 answers, 1.0%).

Each engine has a signature

Five engines looking at the same questions do not read the same sites. For each engine, the site it cited most, within that engine's measured answers:

Three things stand out. Google's two surfaces, AI Overviews and Gemini, lean most on social media and video: Instagram leads AI Overviews and YouTube leads Gemini by a wide margin. Claude made English Wikipedia its most cited source in answers to Turkish questions; 367 of the 377 answers citing English Wikipedia came from Claude. ChatGPT spreads its sources most widely: none of the sites that could enter the table reached 1.5% of ChatGPT's answers, and its most frequent ones included public sources such as the Turkish Ministry of Trade's consumer pages and the Official Gazette.

Part of this difference comes from how the engines look at the web. According to Google's AI features documentation, AI Overviews shows pages from Google Search's index as supporting links. Perplexity searches the web in real time for every question and adds numbered sources to each answer; every Perplexity answer we measured had at least one source. ChatGPT and Claude decide per question whether to search, so some of their answers cite nothing.

Next step

Is your brand named in AI answers?

The visibility measurement asks five engines Turkish questions that do not name your brand, and records every answer. Answers that could not be measured are shown separately, never as zero.

See the visibility measurement

By sector: consumer brands and brands that sell to businesses

We do not give sector-by-sector tables. With the brands we have, a sector holds a few brands at most, and its table would identify the client in it. The only split that keeps the five-brand floor is whether a brand sells to consumers or to businesses.

9 brands selling to consumers (11,742 measured answers):

Trendyol is first in the overall table but absent here: among consumer brands, more than half of the answers citing Trendyol came from one brand's questions, so the rule left it out.

6 brands selling to businesses (4,961 measured answers):

The difference between the two tables is the size of the reach figures. In consumer questions, a few large sites are shared across many brands' answers. In business questions, each sector has its own sites: the widest-reaching shared site appeared in only 1.5% of answers. For a business brand there is almost no "site everyone reads"; which pages the engines read is found by measuring that brand's own questions.

Where is the brand's own site?

The measured brand's own site was cited in 575 of 16,703 answers (3.4%); 849 of 141,498 links (0.6%) went to it. Since the questions do not name the brand, this is the expected picture: when a buyer asks "which X is best", an engine looks at marketplaces, forums and comparison sites more than at a brand's own account of itself.

The engines differ here too. Perplexity cited the measured brand's site most often (224 of 3,371 answers, 6.6%), ChatGPT least often (47 of 3,433 answers, 1.4%).

What should a brand take from this?

1. Measure your own source list. These tables average 15 brands. The sites read for your questions may be very different, especially if you sell to businesses.
2. Read the engines separately. Instagram leads in AI Overviews, English Wikipedia in Claude, marketplaces in Perplexity. A single "AI source list" describes every engine wrongly.
3. Do not neglect your marketplace and forum pages. Your product page description, its reviews and the answers to questions in forums may be the first text an engine reads about you.
4. Publish clear answers on your own site. Your own site is cited in 3.4% of answers; small, but not zero. Dated, sourced pages that answer the buyer's question directly are the most direct way to grow that share. We explain how in how to write content AI engines cite.

Remember too that being cited is not the same as being mentioned: an engine can read your site and not say your name. We cover the difference in brand mentions and source citations. Our earlier source measurement, with different brands, is in who AI reads about Turkish brands; the two were made with different brands and questions, so their percentages should not be compared.

Frequently asked questions

Which site does AI cite most for Turkish questions?

Across the 15 brands we measured, Trendyol was cited in 1,053 of 16,703 measured answers (6.3%), followed by Hepsiburada (4.7%), Instagram (4.5%) and YouTube (4.3%). This ranking depends on the brands and questions measured; it changes with other brands. Most of these sites are marketplaces, social media or forums. Even so, no single site reached 7% of answers: the sources spread over 19,844 different sites, not a few large ones.

Which engine looks at which sources?

The engines have different signatures. Google AI Overviews cited Instagram in 17.6% of its answers, Gemini YouTube in 13.5%, Claude English Wikipedia in 10.8%, Perplexity Trendyol in 14.0%. ChatGPT spread its sources most widely. So a brand needs a source list for each engine separately; a site that leads in one engine can be almost absent in another.

Is my brand's own site cited in AI answers?

In this measurement, the brand's own site was cited in only 3.4% of answers to questions that did not name the brand. To know your own position you need to measure with your own questions; an average does not describe a single brand. The engines differ too: Perplexity cited the brand's own site most often (6.6%), ChatGPT least often (1.4%). Pages that answer the buyer's question directly are the most direct way to grow that share.

Why don't you give results by sector?

With the brands we have, a sector holds a few brands at most. Publishing that cell would identify our client in that sector. So every figure is given over at least 5 brands, and sites dominated by one brand are not listed. Instead of sectors, we split the brands into those selling to consumers (9 brands) and those selling to businesses (6 brands); the article gives the two groups' tables separately.

What happened to the answers that could not be measured?

We did not count the 357 answers that returned an error, never ran or whose source list could not be read, nor the 130 questions for which Google showed no AI Overview. Counting them as "cited nothing" would report a provider error as a finding. 16,703 of the 17,190 planned answers were measured; every share in the article is calculated over those measured answers, and the number not measured is given separately.