Statistics about Czech e-commerce are often borrowed from international studies. On 6 August 2026, I therefore checked 22 selected Czech e-shop domains using one consistent method. One page timed out, 21 loaded, and one of those served a parked domain instead of an active store. The market results below therefore use 20 active e-shops.
That distinction matters. This is not a representative survey of every Czech store and it is not a league table of individual brands. It is a transparent technical snapshot of a deliberately varied sample that can be repeated later with the same method.
In the sample of 20 active Czech e-shops, 7 websites (35%) exposed llms.txt, 13 (65%) had some structured data, and 8 (40%) used an Organization or Store type. Exactly one H1 appeared on 12 websites (60%), complete OG title and image on 9 (45%), and the median share of images without alt text was 27.7%. Median lab desktop LCP was 0.93 s, while four sites exceeded CLS 0.1. The presence of llms.txt or schema alone does not prove better rankings or AI citations.
Methodology and public data
- Selection: 22 known e-shop domains of different sizes and sectors. This is a purposive sample, not a random or representative one.
- Final sample: 20 active e-shops. One domain was parked at the time of testing and one page timed out; both are excluded from the active-store aggregates.
- When and how: 6 August 2026, headless Chromium, desktop 1440 × 900, no network throttling, each website's homepage.
- Content checks: H1, meta description, alt attributes, Open Graph and JSON-LD from the DOM after load.
- Website files: robots.txt, sitemap.xml and llms.txt requested from their standard paths with a plain HTTP client.
- Performance: LCP and CLS through PerformanceObserver. These are lab desktop measurements, not mobile CrUX field data.
Download the anonymised aggregate dataset as CSV. The file contains no store names or domains; it records the published metrics, sample sizes, units and methodological notes.
Finding 1: llms.txt is appearing, but it is not a ticket into AI results
7 of 20 active e-shops (35%) exposed llms.txt. That is an observation about adoption of an experimental format — nothing more.
Google's official guide to generative AI in Search says that foundational SEO still applies and that useful, original and indexable content matters. A special AI text file is not a requirement. OpenAI describes access for OAI-SearchBot for discovery in ChatGPT Search rather than requiring llms.txt.
The measurement therefore cannot show that websites with llms.txt are cited more often. It only shows that seven websites exposed the file at its standard path.
Separately, 11 of 20 active e-shops exposed a sitemap at /sitemap.xml, and 14 of 20 served robots.txt to a plain HTTP client. A failed request does not automatically mean Googlebot is blocked; bot protection can treat clients differently and should be verified in server logs or Search Console.
Finding 2: basic website representation is inconsistent
| Check | Active-sample result | What the data supports |
|---|---|---|
| Exactly one H1 | 12 of 20 (60%) | 8 sites had zero, two or three H1 elements |
| Meta description | 19 of 20 (95%) | nearly standard in this sample |
| Images without alt text | median 27.7% | 12 sites left more than one fifth of images without alt text |
| Complete OG title + image | 9 of 20 (45%) | 11 sites lacked the complete sharing pair |
| Any JSON-LD data | 13 of 20 (65%) | presence alone does not prove completeness or correctness |
| Organization / Store | 8 of 20 (40%) | fewer than half described this basic identity in schema |
| FAQPage | 0 of 20 | the parked domain outside the active sample was the only page with FAQPage |
Structured data can help a search engine understand visible content and make supported types eligible for richer results. It does not guarantee a ranking or an AI citation. Google also limits FAQ rich results mainly to authoritative government and health websites, so the absence of FAQPage is not automatically a commercial problem for a regular e-shop.
Finding 3: desktop lab results are fast, but the spread is wide
- Median LCP was 0.93 s, the average 1.01 s and the maximum 3.34 s.
- 4 of 20 websites exceeded CLS 0.1; one exceeded 0.25.
- The median homepage made 136.5 HTTP requests, with a maximum of 403.
- Median DOM size was 3,115 elements, with a maximum of 8,380.
With no network or CPU throttling, this is a best-case lab scenario. The figures are not a substitute for mobile CrUX or PageSpeed Insights data and they do not reveal the conversion rate of individual stores.
What e-shop operators can actually take from the study
- Verify indexation and visible content first. Important products, categories, prices, delivery and contact information must be available to users and crawlers in normal HTML.
- Use schema as an accurate description, not a trick. Organization, WebSite, Product and offer data must match what visitors can see.
- Check the technical basics. A clear main heading, useful alternative text where an image carries information, reliable sharing metadata and a stable mobile layout help people and machine processing.
- Publish something original for AI visibility. First-party data, a clear method, real experience and stated limitations are more valuable than another generic article or llms.txt on its own.
- Measure AI traffic separately. I explain how to handle referrals from ChatGPT, Perplexity and other sources in How to measure traffic from AI search.
If you need the same kind of diagnostic for your store, I handle it through SEO for companies and e-shops, Shoptet support and e-shop redesign.
Summary
Of 22 selected domains, 21 websites loaded and 20 remained in the active-store comparison. In the final sample, 35% exposed llms.txt, 65% had some JSON-LD data, 60% had exactly one H1 and 45% had a complete OG title and image. Median lab desktop LCP was 0.93 s, while four websites exceeded CLS 0.1.
The main conclusion is not "install llms.txt". It is the gap between the mere presence of a technical element and a website's verified usefulness: AI and classic search both need accessible, accurate and original content that can be discovered and used with confidence.
Measured 6 August 2026; methodology and aggregation updated 8 August 2026. Request an audit of your e-shop — you will receive concrete findings and priorities rather than an automatic score without context.