# Turkish question, English source: in which language does AI cite sources for Turkish buyers?

> In answers to Turkish questions, 10.0% of sources whose language we could read were not Turkish; in ChatGPT, 31.0%. What it means for Turkish brands.

Published: 2026-11-07
<https://geomono.com/en/blog/turkce-soru-ingilizce-kaynak>

When AI engines answer a Turkish question, they take most of their sources from Turkish pages, but not every engine to the same degree: in the answers we measured, 10.0% of the sources whose language we could read were not Turkish, and in ChatGPT the figure rose to 31.0%. GeoMono is a platform that measures how visible brands are in AI answers. This post gives the language of the sources shown in answers to the Turkish questions of 15 brands, by engine and by sector, and explains what it means for a Turkish brand.

The short answer: Turkish content is the majority, but a quarter of answers rest on at least one page in another language. The answer to a buyer's Turkish question can come from a page written in a language you never wrote in.

## What did we measure?

Between 16 August and 31 October 2026, for 15 brands, we asked five engines (ChatGPT, Gemini, Claude, Perplexity and Google's AI Overviews) Turkish questions written without the brand's name. 11,985 of the 16,703 measured answers cited at least one source; those answers held 141,498 source links.

We read a page's language from its address, without opening the page:

- If the address carried an explicit language mark, we took it: a path such as `/en/` or `/tr/`, a subdomain such as `en.` or `tr.`, a language parameter such as `hl=tr`.
- If not, we looked at the domain: sites on `.tr` counted as Turkish, domains belonging to a country other than Türkiye (`.uk`, `.de`, the `.gov` of US public bodies) as not Turkish.
- If not, we looked at the words in the address: Turkish words such as "nasil", "fiyatlari", "rehberi", or English words such as "best", "guide", "how".
- If the address said nothing, and at least 90% of at least 20 readable addresses of the same site were in one language, we took that language. `trendyol.com/p/123` says no language, but Trendyol's readable addresses are Turkish.

If none of these said a language, the page stayed "unreadable". 44,000 of the 141,498 links (31.1%) stayed that way; sites hosting content in every language, such as YouTube, Instagram and Reddit, lead them. We did not guess unreadable links as "Turkish" or "not Turkish"; we kept them out of the shares. Gemini returns its sources as Google redirect addresses and only the site name can be read; since the language of 60.5% of Gemini's links could not be read, we give Gemini no share in the engine table.

> [!UYARI]
> The brands are our clients and group companies; GeoMono's own measurements are not in this count. Every figure is given over the questions of at least 5 brands. Reading a language from an address is an estimate: a page can carry Turkish words in its address and be written in English. Read the figures as magnitudes, not exact shares.

## How much of the sourcing is not Turkish?

| Measure | Value |
|---|---|
| Links with a readable language | 97,498 |
| Not Turkish | 9,713 (10.0%) |
| Answers citing at least one non-Turkish page | 3,073 of 11,985 (25.6%) |
| Answers resting only on non-Turkish pages | 838 of 11,985 (7.0%) |

One link in ten is not Turkish, but one answer in four rests on at least one non-Turkish page. The difference comes from this: an answer usually shows several sources, and one of them can be in another language. In one answer in fourteen, none of the sources shown was Turkish.

This share needs context. According to W3Techs' [content language statistics](https://w3techs.com/technologies/overview/content_language) of 7 November 2026, 49.5% of websites whose content language is known are in English and 1.6% in Turkish. With half the web in English, 90% of sources for Turkish questions being Turkish shows that the engines largely respect the question's language. The issue is not the average but the gap between engines and between topics.

### A .tr domain does not mean Turkish

Only 37,772 of the 139,621 links that resolved to a site (27.1%) went to `.tr` sites. That does not mean three quarters of links went to foreign sites: most of the Turkish sites cited most, such as Trendyol, Hepsiburada, Ekşi Sözlük and Şikayetvar, are on `.com`. A domain is not enough to measure language; that is why we used the language rules above.

## Which engine cites how many foreign sources?

| Engine | Links with a readable language | Not Turkish | Answers citing at least one non-Turkish page | Answers resting only on non-Turkish pages |
|---|---|---|---|---|
| ChatGPT | 5,690 | 1,762 (31.0%) | 835 of 2,242 (37.2%) | 677 of 2,242 (30.2%) |
| Claude | 37,005 | 4,304 (11.6%) | 762 of 1,522 (50.1%) | 63 of 1,522 (4.1%) |
| Perplexity | 33,537 | 3,017 (9.0%) | 1,124 of 3,371 (33.3%) | 68 of 3,371 (2.0%) |
| Google AI Overviews | 12,991 | 567 (4.4%) | 298 of 2,422 (12.3%) | 17 of 2,422 (0.7%) |

There are three different behaviours:

- **ChatGPT** shows the fewest links (8,057), but a third of them are not Turkish. One in three of its source-citing answers rested only on non-Turkish pages. Prominent among them were the English pages of foreign public bodies (in areas such as health and consumer safety) and of software makers. ChatGPT's [help page](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt) says it turns a question into one or more search queries sent to its search partners; it does not say in which language a Turkish question is searched.
- **Claude**'s sources are mostly Turkish, but half of its answers include at least one English page. English Wikipedia leads them: 367 of the 377 answers citing English Wikipedia came from Claude. Of the 171 answers citing Turkish Wikipedia, 55 came from Claude, 80 from Perplexity and 36 from AI Overviews.
- **Google AI Overviews** is the most Turkish surface: only 4.4% of its readable sources were not Turkish. According to the [Google Türkiye Blog](https://blog.google/intl/tr-tr/urun-duyurulari/turkiyede-ai-modu-ve-ai-bakisi-donemi-basliyor/), AI Overviews rolled out gradually in Türkiye from 18 February 2026. For a surface fed by Google Search results, a Turkish search bringing Turkish sources is the expected outcome.

{{cta}}

## By sector: more foreign sources in business questions

| | 9 brands selling to consumers | 6 brands selling to businesses |
|---|---|---|
| Links with a readable language | 68,309 | 29,189 |
| Not Turkish | 5,254 (7.7%) | 4,459 (15.3%) |
| Answers citing at least one non-Turkish page | 1,995 of 8,427 (23.7%) | 1,078 of 3,558 (30.3%) |
| Answers resting only on non-Turkish pages | 585 of 8,427 (6.9%) | 253 of 3,558 (7.1%) |

In business questions, the non-Turkish share of sources doubles. One possible explanation: in topics such as software, industrial equipment and corporate services, comparison and guide pages written in Turkish are scarce and English ones plentiful. When an engine looks for an answer to a Turkish question and finds no Turkish page, it shows the English one.

## What does it mean for Turkish brands?

If the answer to a buyer's Turkish question comes from an English page, the chance that your name is on that page is low. English "best tools" lists usually describe global brands; they do not know a brand working in Türkiye. In practice there are three consequences:

1. **The Turkish content gap is an opportunity.** If no good comparison, guide or definition page exists in Turkish for your topic, the engine shows the English one. A Turkish page that fills that gap becomes one of the few Turkish sources the engine can choose. If you sell to businesses, the gap is larger.
2. **Consider your English presence too.** Claude leaning on English Wikipedia and ChatGPT on English sources makes how your brand is described in English sources matter as well. If your site has an English version, make sure it says the same as the Turkish one. What to watch out for with Wikipedia is in [Turkish Wikipedia or English Wikipedia](/en/blog/vikipedi-mi-ingilizce-wikipedia-mi).
3. **Read the engines separately.** In AI Overviews Turkish sources stand almost alone; in ChatGPT a third are foreign. The same content strategy gives different results in the two.

Which sites are cited, regardless of language, is in [which sites does AI cite in Türkiye](/en/blog/yapay-zeka-turkiyede-hangi-siteleri-kaynak-gosteriyor), and how to write content that gets cited is in [how to write content AI engines cite](/en/blog/yapay-zekanin-alintiladigi-icerik-nasil-yazilir).

## Frequently asked questions

### Does AI cite English sources for Turkish questions?

Yes, but as a minority. In the answers we measured, 10.0% of the sources whose language we could read were not Turkish; 25.6% of source-citing answers rested on at least one non-Turkish page and 7.0% only on non-Turkish pages. The share varies a lot by engine: of the sources whose language we could read, 31.0% were not Turkish in ChatGPT and only 4.4% in Google AI Overviews.

### Which engine cites the most English sources?

ChatGPT. 31.0% of its sources with a readable language were not Turkish, and 30.2% of its source-citing answers rested only on non-Turkish pages. In Google AI Overviews the share was lowest, at 4.4%. Claude's sources are mostly Turkish, but half of its source-citing answers included at least one English page, led by English Wikipedia.

### How did you determine the pages' language?

From the page's address: a language path, a subdomain, a language parameter, the domain ending and the words in the address. Where the address said nothing, we took the language of the same site's readable addresses. The language of 31.1% of links could not be read even so, and they were left out of the shares.

### Should my brand produce English content too?

Turkish first. Most sources shown for Turkish questions are Turkish, and the gap in Turkish content, especially on business topics, is why engines go to English pages. If you have an English version, making sure it gives the same, current information as the Turkish one is a sound start. If you sell to businesses especially, writing Turkish comparison and guide pages on your topic is the most direct way to keep the engine from reaching for an English source.

### Is a .tr site cited more often?

This measurement does not show that. Only 27.1% of links went to `.tr` sites, and most of the Turkish sites cited most are on `.com`. The domain works as a sign of language, but it does not decide whether a site gets cited. The real sign of language is the language path and the words in the address, so the measurement never used the domain ending on its own, only together with the rest of the address.

## Sources

- The language of sources shown in answers to Turkish questions, 2026-11-07: 33 measurements between 16 August and 31 October 2026, 15 brands (9 sell to consumers, 6 to businesses), Turkish questions that do not name the brand; 11,985 of the 16,703 measured answers cited at least one source, 141,498 links; language read from the page's address, or where the address says nothing, from the language of at least 90% of at least 20 readable addresses of the same site; the language of 44,000 links (31.1%) could not be read and was left out of the shares; GeoMono's own measurements excluded; the brands are our clients and group companies, not a market sample; counted on 7 November 2026
- Share of sites on the .tr domain, 2026-11-07: 139,621 of the same 141,498 links that resolved to a site; for Gemini the site name was read from the link's title
- [Usage statistics of content languages for websites](https://w3techs.com/technologies/overview/content_language), W3Techs, data of 7 November 2026
- [Searching the web with ChatGPT](https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt), OpenAI Help Center, accessed 7 November 2026
- [Türkiye'de AI Modu ve AI Bakışı Dönemi Başlıyor](https://blog.google/intl/tr-tr/urun-duyurulari/turkiyede-ai-modu-ve-ai-bakisi-donemi-basliyor/), Google Türkiye Blog, 18 February 2026