AI assistants such as ChatGPT, Gemini, Claude and Perplexity are answering shopping questions more and more often. For an e-commerce site to appear in those answers, these systems first have to be able to access the site and understand its content correctly. To measure where Turkish e-commerce sites stand on this, we scanned 1,914 Turkish sites.
No customer data was used in the study. All findings were collected on 14 November 2026 from the sites’ publicly accessible home pages, robots.txt files and llms.txt files. We describe the method and its limitations in detail at the end of the article.
Key findings
- Turkish e-commerce sites hardly block AI crawlers at all. Only 6.1% of the 528 e-commerce sites in the sample block at least one AI bot in their robots.txt file. The share blocking AI search bots (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, PerplexityBot) is 0.6%.
- Half of llms.txt files carry no real content. 39.4% of e-commerce sites have an llms.txt file, but only 19.7% have one written specifically for the site. The rest are templates generated automatically by the platform (10.8%) or empty files (8.9%).
- The platform largely determines AI readiness. Structured data (JSON-LD) is present on the home page of 98.7% of Ticimax sites, 88.2% of Shopify sites and 27.9% of IdeaSoft sites.
- Readiness for selling abroad is weak. Only 28.6% of e-commerce sites point to other language versions with hreflang tags.
- Basic SEO tags are widespread. 93.6% of e-commerce sites have a canonical tag and 86.9% have a meta description.
Scope of the study
The sample consists of all .tr domains among the top one million sites of Tranco, an open popularity list used by researchers (excluding second-level domains for government, education and similar bodies). There were 1,914 sites on this list; 1,607 of them could be reached. The others blocked automated requests or did not respond.
Of the sites reached, 528 were classified as e-commerce sites: sites where a known e-commerce platform was detected, or where the home page has an “add to cart” button or a cart link. Unless stated otherwise, the percentages below refer to these 528 sites.
Which platforms do Turkish e-commerce sites use?
| Platform | Number of sites | Share |
|---|---|---|
| Custom-built or undetected | 188 | 35.6% |
| Ticimax | 79 | 15.0% |
| Shopify | 68 | 12.9% |
| T-Soft | 60 | 11.4% |
| WooCommerce | 53 | 10.0% |
| IdeaSoft | 43 | 8.1% |
| ikas | 14 | 2.7% |
| Other (Magento, OpenCart, PrestaShop, SAP, Salesforce) | 23 | 4.4% |
A third of the e-commerce sites in the sample use their own software, or their platform cannot be determined from the home page code. Among ready-made platforms, Ticimax, Shopify and T-Soft stand out. These shares do not reflect the Turkish market as a whole, but this sample of .tr sites that are relatively popular.
Access for AI crawlers
AI companies use two kinds of crawlers. Training crawlers (such as GPTBot, ClaudeBot and Google-Extended) collect content for model training. Search crawlers and crawlers that act on behalf of a user (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, PerplexityBot) read a page while answering a question and cite the source. For a site to appear as a source in AI answers, it needs to be open to the second group in particular.
| Crawler | E-commerce sites blocking it | All sites blocking it |
|---|---|---|
| GPTBot (OpenAI, training) | 4.7% | 12.1% |
| ClaudeBot (Anthropic, training) | 5.3% | 12.6% |
| Google-Extended (Gemini training) | 4.3% | 11.3% |
| CCBot (Common Crawl) | 5.5% | 9.3% |
| OAI-SearchBot (ChatGPT search) | 0.0% | 5.1% |
| ChatGPT-User | 0.2% | 5.6% |
| Claude-SearchBot | 0.4% | 5.5% |
| PerplexityBot | 0.0% | 5.5% |
E-commerce sites are noticeably more open to AI crawlers than the sample as a whole. While 15.2% of all sites block at least one AI bot, the figure for e-commerce sites is 6.1%. Hardly any e-commerce site blocks AI search bots. The share of e-commerce sites that have written separate rules for AI bots in their robots.txt file is 16.4%. In other words, most sites have not made a deliberate decision on this; the bots are allowed by default.
Being open to access is not enough on its own. For an AI assistant to recommend a product, it has to be able to read from the page accurately what the product is, who it is suitable for, its price and its stock status. Structured data and clear product content matter for this.
llms.txt: The file exists, the content does not
llms.txt is a recommended but non-standard file in which a site summarises itself and its important pages for AI systems in Markdown format. Google states that it does not need such a file for its AI features; even so, some AI tools can read it.
| E-commerce sites | |
|---|---|
| Have an llms.txt file | 39.4% |
| Written specifically for the site | 19.7% |
| Automatic template from the platform (Shopify) | 10.8% |
| Empty (heading only) | 8.9% |
The breakdown by platform makes the picture clearer. 94% of Shopify sites have an llms.txt file, but most of these are Shopify’s automatic file that gives shopping instructions to AI agents; only 10.3% carry site-specific content. 97% of T-Soft sites have the file, but 78% of them consist of nothing more than a heading. None of the IdeaSoft sites have an llms.txt file. On Ticimax, WooCommerce and ikas sites, all of the files that exist carry site-specific content.
In short, the llms.txt rate reflects the platform’s default settings more than a deliberate choice by the stores.
Structured data: A big difference by platform
Structured data (JSON-LD) lets search engines and AI systems understand the information on a page in a machine-readable form. The markup most often expected on a home page is Organization, which identifies the brand, and WebSite with SearchAction, which describes the on-site search.
| Platform | Sites | JSON-LD on home page | Organization | Meta description |
|---|---|---|---|---|
| Ticimax | 79 | 98.7% | 98.7% | 97.5% |
| T-Soft | 60 | 91.7% | 88.3% | 90.0% |
| Shopify | 68 | 88.2% | 86.8% | 94.1% |
| Custom-built / undetected | 188 | 68.1% | 61.7% | 87.8% |
| WooCommerce | 53 | 64.2% | 56.6% | 64.2% |
| IdeaSoft | 43 | 27.9% | 25.6% | 79.1% |
Across all e-commerce sites, any JSON-LD is found on the home page in 75.4% of cases and Organization markup in 71.4%. SearchAction, which describes the on-site search, appears in 39.8%, BreadcrumbList, which shows the page hierarchy, in 7.4%, and FAQPage for frequently asked questions in 4.2%. Although the rates look high for ikas sites, the 14 ikas sites in the sample are too few for a sound comparison.
This measurement looks only at the HTML that the server delivers for the home page. Sites that add structured data with JavaScript after the page has loaded may appear lower in this table than they actually are; Product markup on product pages is also outside the scope of this study.
Basic SEO and readiness for international markets
| Criterion | E-commerce sites |
|---|---|
| Viewport tag for mobile compatibility | 99.3% |
| Canonical tag | 93.6% |
| Meta description | 86.9% |
| Sitemap in robots.txt | 81.4% |
| Sharing image (og:image) | 52.7% |
| hreflang for language versions | 28.6% |
Basic technical SEO has largely been solved by the platforms. What is really missing is readiness for selling abroad: 71% of e-commerce sites do not tell search engines about their different language versions with hreflang. About half of the sites have no image defined that would appear when shared on social media and messaging apps.
What this means for e-commerce sites
- Stay open to AI search bots. Blocking OAI-SearchBot, ChatGPT-User, Claude-SearchBot and PerplexityBot removes any chance of your site being cited as a source in AI answers. You can make a separate decision for training bots; it is possible to manage these two groups separately in robots.txt.
- Do not leave llms.txt to the platform. If you have the file, check its content. If it is empty or only a template, prepare a short file summarising your brand, your product categories, your delivery and returns terms, and your most important pages.
- Complete your structured data. Organization and WebSite on the home page, BreadcrumbList on category pages, and Product markup with price, stock and reviews on product pages are the basic requirements. If your platform does not provide them, they can be added through an app or a theme edit.
- Use hreflang if you sell abroad. Declaring each language version correctly to search engines prevents the page in the wrong language from being shown.
- Write product content so that AI can understand it. What the product is, who it is suitable for, its material, its dimensions and its conditions of use should be written in clear sentences. We cover this topic in detail in our GEO guide for e-commerce.
Method and limitations
- Sample: All .tr domains in Tranco list Y897G (14 November 2026); second-level domains such as gov.tr, edu.tr and org.tr are excluded. Turkish brands on .com domains (e.g. large marketplaces) are not in the sample; this is the study’s most important limitation.
- Data collection: Three requests were sent to each site: the home page, /robots.txt and /llms.txt. No pages requiring a login, no forms and no personal data were accessed. 307 sites were left out of the analysis because they blocked automated requests or did not respond.
- Platform detection: This was done using known traces in the home page code (e.g. cdn.shopify.com for Shopify). Sites that hide their platform or use it headless may appear in the “undetected” group.
- E-commerce classification: Sites with a known e-commerce platform, an “add to cart” button or a cart link were counted as e-commerce. Some service sites that carry out online transactions may have been included in this group.
- Robots.txt interpretation: For each bot, we checked whether crawling of the site’s root page is permitted, using Python’s standard robots.txt parser. Server-side bot blocking (such as by a firewall) is outside this measurement.
- llms.txt classification: Files with fewer than three lines and fewer than 150 characters of content were counted as “empty”, Shopify’s automatic agent-instructions file as a “template”, and the rest as “site-specific”.
- Structured data: Only the HTML delivered by the server for the home page was examined; markup added later with JavaScript and product pages are outside the scope.
Frequently asked questions
Does blocking AI bots harm an e-commerce site?
It depends on the type of bot being blocked. Blocking training bots limits the use of your content in model training. Blocking search bots and bots that act on behalf of a user (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, PerplexityBot), on the other hand, prevents your site from being cited as a source in AI answers. In our study, 0.6% of e-commerce sites block this second group.
Is an llms.txt file mandatory?
No. llms.txt is a recommended format; Google states that it is not required for its AI features. If you are going to publish a file, a complete and up-to-date one is more meaningful than an empty or template-only file.
Can I use this study with attribution?
Yes. You may use the findings under the title “Roas Digitall, Are Turkish E-Commerce Sites Ready for AI Search? A 2026 Study”, with a link to this page. The aggregate results can be downloaded as a JSON file.
Let’s assess your site together
We can review your site’s access for AI crawlers, its llms.txt file and its structured data, and identify the gaps together. You can find our approach on our GEO and SEO pages.




