We ask first.

This website works without cookies. If you agree, we will also use cookies for measuring visits and our advertising. You can change your mind at any time from “Cookie settings” at the bottom of every page. Cookie policy

Greek online shops let ChatGPT in. They tell it the price, not the barcode.

Noria, data study. Published:

The aggregate data (CSV)

AI assistants, ChatGPT, Gemini, Perplexity, now recommend products and stores, and are starting to buy for their users. To recommend an online shop they must be able to read it and understand what it sells. On 20 October 2026 Noria automatically checked the same 3,745 Greek online shops as its previous study, with the code of its free check «Can ChatGPT see your site?».

The key numbers

  1. Only 0.5% of Greek online shops block ChatGPT search in robots.txt, and 1.8% block any AI search crawler (Noria, October 2026).

    Sample and confidence interval

    19 of 3,676 stores. 95% confidence interval (Wilson): 0.3% to 0.8%.

    66 of 3,676 stores. 95% confidence interval (Wilson): 1.4% to 2.3%.

  2. 6.3% block the crawlers that collect text for model training but leave ChatGPT search open.

    Sample and confidence interval

    233 of 3,676 stores. 95% confidence interval (Wilson): 5.6% to 7.2%.

  3. Only 0.4% show an almost empty page without JavaScript. The rest send their text in the HTML, where AI crawlers read it.

    Sample and confidence interval

    13 of 3,641 stores. 95% confidence interval (Wilson): 0.2% to 0.6%.

  4. 88.8% of product pages carry product data, and 84.8% carry availability.

    Sample and confidence interval

    2,028 of 2,284 stores. 95% confidence interval (Wilson): 87.4% to 90%.

    1,936 of 2,284 stores. 95% confidence interval (Wilson): 83.2% to 86.2%.

  5. Only 14.7% of product pages give a GTIN (barcode) and 48.6% a brand: the details machines use to recognise the same product from store to store.

    Sample and confidence interval

    336 of 2,284 stores. 95% confidence interval (Wilson): 13.3% to 16.2%.

    1,109 of 2,284 stores. 95% confidence interval (Wilson): 46.5% to 50.6%.

  6. Just 1.4% state their return policy as data and 1.5% their shipping, two details Google recommends in product data.

    Sample and confidence interval

    32 of 2,284 stores. 95% confidence interval (Wilson): 1% to 2%.

    35 of 2,284 stores. 95% confidence interval (Wilson): 1.1% to 2.1%.

  7. 33.2% have on the page all 9 details OpenAI's product feed requires.

    Sample and confidence interval

    759 of 2,284 stores. 95% confidence interval (Wilson): 31.3% to 35.2%.

  8. 60.8% say, as data, which business they are, and only 47.5% with a name and a logo or address.

    Sample and confidence interval

    2,213 of 3,641 stores. 95% confidence interval (Wilson): 59.2% to 62.4%.

    1,729 of 3,641 stores. 95% confidence interval (Wilson): 45.9% to 49.1%.

Assistants can get in.

Which crawlers robots.txt blocksShare of stores blocking each crawler, on the whole site or part of it. In 3,676 stores whose robots.txt answered.
OAI-SearchBot (ChatGPT search)0.5%19/3,676, 95% confidence interval: 0.3–0.8%
ChatGPT-User1.3%47/3,676, 95% confidence interval: 1–1.7%
Claude-SearchBot0.4%14/3,676, 95% confidence interval: 0.2–0.6%
PerplexityBot1.3%46/3,676, 95% confidence interval: 0.9–1.7%
Bingbot0.3%12/3,676, 95% confidence interval: 0.2–0.6%
Googlebot0%0/3,676, 95% confidence interval: 0–0.1%
GPTBot (OpenAI training)6.3%233/3,676, 95% confidence interval: 5.6–7.2%
ClaudeBot (Anthropic training)6.1%226/3,676, 95% confidence interval: 5.4–7%
Google-Extended (Gemini training)4.1%149/3,676, 95% confidence interval: 3.5–4.7%
No robots.txt (everything allowed)5.8%212/3,676, 95% confidence interval: 5.1–6.6%
robots.txt managed by Cloudflare4.4%163/3,676, 95% confidence interval: 3.8–5.1%

OpenAI, Anthropic and Google run separate crawlers for search and for model training. Greek online shops use this well: those that block anything almost always block training and leave search open.

robots.txt says what a store declares, not what its firewall does. Settings such as Cloudflare's «Block AI bots» block without showing there, and those we cannot measure from outside.

The text gets through. The identity, halfway.

What a crawler finds on the home pageIn 3,641 home pages read, without running JavaScript.
Almost no text (under 80 words)0.4%13/3,641, 95% confidence interval: 0.2–0.6%
Little text (80 to 249 words)1.4%52/3,641, 95% confidence interval: 1.1–1.9%
Asks not to be indexed (noindex)0.3%11/3,641, 95% confidence interval: 0.2–0.5%
Business data (Organization)60.8%2,213/3,641, 95% confidence interval: 59.2–62.4%
…with a name and a logo or address47.5%1,729/3,641, 95% confidence interval: 45.9–49.1%
Sitemap78.5%2,858/3,641, 95% confidence interval: 77.1–79.8%
llms.txt16.6%606/3,641, 95% confidence interval: 15.5–17.9%

The median home page sends 1,369 words of text in its HTML. The crawlers of OpenAI, Anthropic and Perplexity do not run JavaScript, as a Vercel study showed, so this is what they read.

About one store in six has an llms.txt, almost always automatically: 195 of 200 Shopify stores have one, and on WooCommerce SEO plugins write it, as the files themselves say. No major platform has said it uses the file.

The product shows. The comparison doesn't.

What the product page's data saysIn 2,284 product pages, one per store, confirmed by their platform's template.
Product data (JSON-LD or microdata)88.8%2,028/2,284, 95% confidence interval: 87.4–90%
What Google requires (name, image, price, currency)84.2%1,924/2,284, 95% confidence interval: 82.7–85.7%
Availability84.8%1,936/2,284, 95% confidence interval: 83.2–86.2%
Description80.3%1,833/2,284, 95% confidence interval: 78.6–81.8%
ID (SKU)78.9%1,801/2,284, 95% confidence interval: 77.1–80.5%
Brand48.6%1,109/2,284, 95% confidence interval: 46.5–50.6%
GTIN (barcode)14.7%336/2,284, 95% confidence interval: 13.3–16.2%
Ratings2.7%61/2,284, 95% confidence interval: 2.1–3.4%
Shipping1.5%35/2,284, 95% confidence interval: 1.1–2.1%
Return policy1.4%32/2,284, 95% confidence interval: 1–2%
All 9 fields of OpenAI's feed33.2%759/2,284, 95% confidence interval: 31.3–35.2%

Almost every store tells machines what the product is, what it costs and whether it is in stock. Few give what a comparison needs: the barcode, the brand, shipping cost and time, the return policy, ratings.

These are exactly what the new sales channels ask for. Google recommends them in merchant-listing data, and OpenAI's product feed asks for brand, seller, availability and price for every product.

The platform makes the difference.

Product data, by platformPlatforms with at least 100 confirmed product pages.
Shopify94.7%179/189, 95% confidence interval: 90.5–97.1%
PrestaShop96.9%156/161, 95% confidence interval: 92.9–98.7%
WooCommerce91.9%1,293/1,407, 95% confidence interval: 90.4–93.2%
OpenCart82.2%291/354, 95% confidence interval: 77.9–85.8%
GTIN in product data, by platformThe same product pages.
Shopify35.4%67/189, 95% confidence interval: 29–42.5%
PrestaShop21.7%35/161, 95% confidence interval: 16.1–28.7%
WooCommerce15.4%216/1,407, 95% confidence interval: 13.6–17.3%
OpenCart2%7/354, 95% confidence interval: 1–4%

On Shopify, PrestaShop and WooCommerce, product data almost always comes from the template itself. The GTIN, though, depends on whether the store filled it in, and there the differences are large.

For Magento, CS-Cart, VirtueMart and stores on their own systems we confirmed fewer than 100 product pages, so we do not show them separately.

What to do, in order.

Access is solved for most stores. What makes the difference now is the data: the more complete it is, the more easily an assistant recommends your product rather than someone else's.

  1. Fill in the GTIN and brand of every product, and make sure they reach the page's data.
  2. State your return and shipping policies as data, once for the whole store, as Google recommends.
  3. Add business data (Organization) with your name, logo, contact details and profiles.
  4. Keep the robots.txt decisions separate: search open, training as you decide.
  5. Check the bot settings of your hosting or Cloudflare, which can block without robots.txt saying so.

See where your own site stands, check by check, against Greek online shops.

Run the free check

How we measured it.

Sample

The 3,745 stores of the 7 October 2026 study: names from Skroutz's public merchant sitemap and from Noria's company list, verified as Greek online shops. It is not a random sample of all Greek online shops.

What we read

robots.txt, the home page, one product page, /llms.txt, and /sitemap.xml when robots.txt declared no sitemap. Up to 8 seconds and 3 MB per page, without running JavaScript.

Politeness

We obeyed robots.txt for our own crawler (NoriaCheck), with one request at a time per store and a pause between. The crawler says who it is.

Product page

From the home page's links. It counts only if it carries its platform's single-product template or og:type product. We do not look at structured data to choose it, so the shares do not lean towards pages that have it.

Rules

The same as Noria's free check. robots.txt is read as RFC 9309 defines: a group per crawler, the most specific wins, the longest rule wins.

Intervals

Every share with its 95% confidence interval (Wilson) and its base.

Limits

  • An automated check from outside. Firewall blocks we cannot see, because we do not pretend to be the crawlers.
  • Of the 3,745 stores, 71 did not let the check read the page, 15 took over 8 seconds and for 17 robots.txt did not answer. They do not count in the page shares.
  • A product page was confirmed in 2,284 stores. Platforms whose templates we do not recognise are under-represented in the product shares.
  • One product page per store. Usually all come from the same template, but we did not check every product.
  • The 7 October study had found product data on 74.7% of pages, because it also counted category pages. With the same definition today we found 74.9%; with confirmed product pages, the share is the one shown here.

Sources

How to cite it

Noria (2026). How ready are Greek online shops for AI assistants? An audit of 3,745 stores (October 2026). noria.gr/en/research/eshop-ai-readiness-2026

The data

Every indicator of the study, in aggregate: for each, how many stores (k) of how many (n), the share and its 95% confidence interval. Nothing per store.

Licence: CC BY 4.0, crediting Noria.

The aggregate data (CSV)