Who does AI read when it describes Turkish brands? Şikayetvar, Ekşi Sözlük, marketplaces

When an AI engine answers a question about a Turkish brand, it draws most of the material for its answer not from the brand's own site but from what other people have written about that brand. In our own measurements, the sources with the widest reach were complaint platforms, forums, video sites and marketplaces. This post explains what we measured, why it may be so and what a brand can do about it; it also states which measurement each figure comes from, its date and its scope.

What did we measure?

We record the source links that AI engines show in their answers. We weigh a source not by its raw link count but by its reach: in how many separate analyses, or in how many separate responses, it was shown. The reason is simple: a single long response that links to the same site ten times is the habit of one answer, not a market trend.

The figures in this post come from two separate measurements, and we never combine them:

We recounted Measurement A's figures from the same data on 7 October 2026. The first count had found 16 brands, 11,203 responses and 75,417 source links; corrections to the stored data in the meantime changed the totals slightly. The reach figures below came out the same in both counts.

Measurement A: complaint and review sites came out on top

Across the 60 analyses we ran up to 3 August 2026, the sources with the widest reach were:

Şikayetvar, a Turkish complaints platform, appeared in 15 of the 19 brand records and with all five engines that cite sources; that was the widest engine spread in the table. Trendyol and Hepsiburada, the Turkish marketplaces, were also seen with all five engines. All four of the top four sources were sites written by users: one complaints platform, one video site and two forums. The press, official records and industry bodies came nowhere near this reach.

In the same analyses, 71,027 of the 75,914 source links the engines showed, that is 93.6%, pointed somewhere other than the brand's own site. In 6 of the 60 analyses the brand's own site was not cited even once; in 3, Şikayetvar was cited more often than the brand's own site.

The sources were also spread differently across engines. Şikayetvar, Trendyol and Hepsiburada appeared with all five engines; YouTube and Reddit with four, and Ekşi Sözlük with three (Google AI Overviews, Claude and Perplexity). A source that is not specific to a single engine stops being the habit of one model and becomes a reflection of what the web holds about these brands.

Having your own site cited does not mean your name is mentioned either. In the same analyses, of the 2,748 responses that cited the brand's own site, 341 (12.4%) did not mention the brand's name in the answer at all; this rate ranged from 3.8% to 22.9% depending on the engine. The engine may have read your site but not told the customer your name. We explain this gap in detail in our post on brand mentions versus source citations.

Measurement B: the picture changes for corporate brands

In Measurement B we look only at questions that do not name the brand, and we list a source only if it was shown for all three brands. A source that appears for only one or two brands tells you about those brands, not about the market.

In this measurement Şikayetvar appeared in only 6 of the 1,646 responses and for two of the three brands, so it did not make the list. The sources that appeared for all three brands draw a different picture:

Our reading is this: complaint platforms seem to stand out in questions about consumer products. Measurement A's questions were mostly about products and brands; Measurement B's brands were corporate and its questions did not name the brand. Ekşi Sözlük was among the top five sources in both measurements: opinions written in forum language may travel across sectors, while complaints may follow consumer products. This is an interpretation; confirming it needs a wider measurement, sector by sector.

Wikipedia: two measurements, two different results

Wikipedia is a good example of how two measurements can fail to agree. In Measurement A, English Wikipedia was cited in 45% of analyses and Turkish Wikipedia (Vikipedi) in 32%; English came out ahead. In Measurement B the picture reversed: Turkish Wikipedia appeared in 72 of the 1,646 responses and English Wikipedia in 26. Both results are as measured; because they were measured with different brands and different questions they do not refute each other, they only show that it is too early to set a general rule here. The details are in our post comparing Turkish and English Wikipedia.

Why these sites?

It is not surprising once you look at how an answer is built. Perplexity says on its help page that it searches the web the moment you ask a question and gives every answer with numbered sources. Google explains that AI Overviews and AI Mode can break a question into subtopics and run several related searches. In other words, the engine reads the pages a person would come across when searching.

Someone in Turkey searching for a brand often lands on that brand's complaints page, a forum thread or a marketplace product page. These sites are also large: according to the counters on Şikayetvar's own "Hakkımızda" (About us) page, as of 7 October 2026 the platform has 13.8 million individual members and about 249 thousand registered brands, and shows 22.5 million visitors in the last 30 days. These are figures the platform publishes itself; we have not verified them independently.

Another clue comes from Ahrefs' analysis of 75 thousand brands: the strongest correlation with a brand being mentioned in Google AI Overviews was the number of web pages mentioning the brand's name (0.664); the correlation with backlinks was much weaker (0.218). The study states clearly that this is a correlation, not cause and effect. But the direction matches our measurements: the engines' material is what others write about you.

Do these sources change the tone of the answer?

We do not answer this question with a figure in this post. We previously ran an internal analysis comparing the tone of Measurement A's answers by source type; the tone classification that analysis relied on was later corrected, and the same figures cannot be reproduced today. Publishing a figure that cannot be reproduced would mean writing up something we did not measure as if we had.

What we can say is this much: an answer usually shows many sources, and tone forms across the whole answer. Showing which source alone determines the tone would require changing sources one at a time and asking the same questions again. Without doing that, saying "Şikayetvar makes the answer negative" would be a guess.

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

What can brands do?

The steps below aim to keep the material the engines read accurate and current. We have not measured that any of them changes AI answers by a particular amount, and we make no such promise. What we can say is that the next measurement, run with the same questions, will show the change.

1. First measure which sources are being read

Measure for your own brand rather than relying on a general list. The gap between Measurement A and Measurement B shows that while Şikayetvar may be decisive for a consumer brand, academic publications or Wikipedia may stand out for a corporate brand. Measure each engine separately; the engines read different sources. Our Sources and competitors page explains how this measurement appears in the report.

2. Answer complaints where they are read

According to Şikayetvar's own description, complaints are reviewed by editors before publication and then forwarded to the brand; the brand can write a reply that is published on the site or talk to the person who complained. The platform also offers a corporate membership programme for brands. An answered and resolved complaint is not the same page as an unanswered one; the engine passes on whichever one it reads.

3. Keep your marketplace pages complete

In Measurement A, Trendyol and Hepsiburada appeared as sources with all five engines. If the product title, specifications, dimensions, warranty and returns information are missing or inconsistent, so is the information the engine passes on. Answers to shoppers' questions to the seller are also part of these pages. If you do not have your own site, our post for marketplace sellers covers this step in detail.

4. Publish clear answer pages on your own site

Publish information such as frequently asked questions and warranty and returns terms in paragraphs that each make sense when read on their own. Google says there is no additional requirement for appearing in AI Overviews; it is enough for the page to be indexed and eligible to be shown with a preview in search results. If you use structured data, mark up only information that is visible on the page, as Google's rules require.

Put this information in the first HTML the server sends for the page. According to Vercel's December 2024 analysis, OpenAI's crawlers, Anthropic's ClaudeBot and PerplexityBot do not run JavaScript. Warranty terms loaded by code that runs only in the browser do not exist on the page at all for these crawlers; the engine either cannot give that information or takes it from other pages it can read (a forum, a complaint record or a marketplace listing). You can see where your site stands on this with our free scan.

5. Check whether what the engine says about you is accurate

Engines can pass on outdated or wrong information from the pages they read. Compare what the engines say with the current information on your site. If there is nothing to compare against, the result is not "clean" but "could not be measured": wrong information is also hardest to spot for a brand whose site cannot be read. Our guide to detecting and correcting false information covers this step.

6. Measure again with the same questions

To see the effect of a change you made, run the next measurement with the same questions and the same engines. A measurement with a different question set draws a different picture, as the Measurement A and Measurement B example shows, and does not show the change.

Frequently asked questions

Do complaints on Şikayetvar determine the AI answer?

We have no measurement that shows this. What we show is this: across the 60 analyses we ran up to 3 August 2026, Şikayetvar appeared as a source in 65% of analyses and with all five engines. A source being cited does not mean it determines the answer; causation has to be measured separately.

Isn't my own site enough?

Your own site is necessary but not sufficient on its own. In the same analyses, 93.6% of the source links the engines showed pointed somewhere other than the brands' own sites. What is written about you outside your site is most of the material for the answer.

Do these figures apply to my sector too?

Not directly. Neither of our measurements is a sample of the Turkish market, and they give results that differ from each other. Without measuring for your own sector and your own questions, it is not possible to say which source is decisive.

I don't have my own site; I only sell on marketplaces. What can I do?

In Measurement A, Trendyol and Hepsiburada appeared as sources with all five engines; that is, the engines read marketplace pages. If you do not have your own site, you need to run the measurement by your brand name rather than by a domain. Keeping your product titles, specification tables, question answers and store information complete and consistent is the most direct tool you have; the next measurement with the same questions will show how much this changes the answers.