ChatGPT Compares Product Reviews — and Now Actively Offers It to Czech Users. What Should Your E-shop Do?

This week OpenAI started sending Czech users a marketing email — in Czech — with the pitch: "Share a few items you're browsing and go through the reviews before you decide. Ask ChatGPT to compare what customers say about your options. It can pull out ratings, review themes and common complaints, then rank the best choices in a simple comparison table."

In other words: OpenAI is actively teaching Czech shoppers to make the final decision between three electric kettles not on Heureka (the leading Czech review aggregator) or on your e-shop, but in ChatGPT. This article breaks down how the feature works, where ChatGPT gets reviews of Czech products — and what to do so your product isn't missing from that comparison table (or worse: represented only by its "common complaints").

What exactly ChatGPT can do with reviews

This isn't a brand-new feature but the culmination of two steps OpenAI merged into a single shopping experience:

• Shopping research (November 2025) — a specialized variant of GPT-5 mini trained on shopping tasks. It asks clarifying questions, browses the web (prices, availability, specs and, crucially, reviews) and builds a personalized buying guide. It works for signed-in users, including the Free tier.
• The redesigned shopping experience (March 24, 2026) — visual product browsing, conversational filtering and side-by-side comparison tables: price, features, ratings, "best for whom."

What's new from the Czech market's perspective is the localization: OpenAI promotes the feature in Czech, it works in Czech, and Czech products genuinely compete in it. And because the model also extracts themes and common complaints from reviews, there's an uncomfortable novelty for every merchant: ChatGPT won't just tell your customers what you claim about the product — it will tell them what other people complain about.

Where ChatGPT gets reviews (and what it means for the Czech market)

ChatGPT has no review database of its own — it assembles reviews from what its crawlers read and what it deems trustworthy. Available analyses show a clear hierarchy:

1. Review platforms and aggregators. According to a citation analysis published in June 2026 by Lupa.cz, Heureka.cz is ChatGPT's fourth most-cited source in the Czech Republic (for Perplexity it's second, right after YouTube). The most-recommended domains include Allegro.cz (8.3 % of recommendations), Alza.cz (4.7 %), Zboží.cz (2.9 %) and Heureka.cz (2.3 %). A review profile on the aggregators is therefore not a "nice to have" — it's a direct input into AI answers.

2. Community content. Shopping research is trained to prioritize organic content including review sites and Reddit — a real user's experience from a forum can weigh more than ten ratings on your product page. Wikipedia and Reddit together account for over 25 % of ChatGPT citations (US data); YouTube appears in 16 % of AI search answers.

3. Your product pages. An analysis of roughly a million ChatGPT Shopping listings (Profound) showed that 88 % come from crawling product pages — with a growing share of official feeds. ChatGPT also displays rating counts for products, which it reads from structured data.

4. Who's missing from the system: Amazon. It blocks all OpenAI crawlers, so millions of Amazon reviews are invisible to ChatGPT. For Czech e-shops it's a reminder that visibility is a choice — and plenty of Czech websites block AI bots without ever having treated it as a decision.

Overview: where AI reads about your products in the Czech Republic

Context that should worry most of the market: according to the same Lupa.cz analysis, roughly 91 % of Czech e-shops are invisible to AI — out of some 67,000, AI systems mention only about 6,000.

The most common mistake: reviews that technically don't exist

Here's the detail that decides the game and almost nobody addresses it: OpenAI's crawlers (GPTBot, OAI-SearchBot) do not execute JavaScript. They read the raw HTML exactly as the server sends it.

Yet most Czech e-shops display reviews via a third-party JavaScript widget that only loads in the browser. The result: an e-shop with thousands of honestly collected ratings is, to ChatGPT, an e-shop without a single review — no stars in shopping cards, no trust signals, no chance in the "what do customers say" comparison. In the comparison table it's represented by an empty space, or by Heureka reviews it doesn't control.

The 30-second test: open your product page, hit "view page source" (not the inspector — the raw source) and search for the text of any review. If it's not there, it doesn't exist for AI either.

Checklist: how to get your reviews into ChatGPT

1. Reviews in server-side HTML. Either render them on the server (SSR) or use a widget that injects the content directly into the HTML. Reviews loaded by JavaScript after the page renders are invisible to AI.
2. Schema.org AggregateRating + Review. Add real aggregate ratings and a few specific reviews in JSON-LD to your Product markup. Roughly 65 % of pages cited by ChatGPT have structured data — and a product without stars loses the comparison even when it's cheaper. Never fabricate values: the markup must match the visible content. Guide: Schema markup for GEO.
3. Check robots.txt. At minimum allow OAI-SearchBot (search and citations) and ChatGPT-User (user actions) — you can block the training crawler GPTBot independently. In detail: robots.txt for AI crawlers.
4. Cultivate Heureka and Zboží.cz. In the Czech market these are two of the most-cited sources in AI answers. Verified reviews, merchant responses, up-to-date data — your aggregator profile is now also your business card in ChatGPT.
5. Answer the "common complaints" before AI surfaces them. The model extracts recurring complaints from reviews. If you have an answer ready (an FAQ on the product page, "who this product is not for," an explanation of trade-offs), your version becomes part of the context. If not, only the complainers speak for you.
6. Support your community footprint. Real mentions on forums, in discussions and on YouTube (review videos have transcripts that AI reads) weigh more than anything on your own website.
7. Measure your baseline. Ask ChatGPT your customer's question ("compare [your product] and [competitor], what do the reviews say?") and save a screenshot. Three months after your changes you'll have a before/after. Systematically: How to measure GEO results.

What it means strategically

Review comparison moves AI shopping into a phase where conversion is decided by trust, not just visibility. Being in the answer isn't enough — next to your product, ChatGPT places a synthesis of what people think about it. E-shops that spent years collecting reviews purely for the conversion rate on their own site suddenly have an asset that works (or does damage) in an entirely new channel.

For the broader context of AI shopping — why OpenAI cancelled in-chat payments, how the product feed works and what Google is planning — see our guide Shopping via ChatGPT in 2026.

Frequently asked questions

Does review comparison in ChatGPT work in Czech? Yes. Both shopping research and the comparison tables work for signed-in users with no geographic restriction, they answer in Czech — and OpenAI actively promotes the feature to Czech users with localized emails.

Does ChatGPT read reviews from Heureka? Heureka.cz is among ChatGPT's most-cited sources in the Czech Republic (4th place per the Lupa.cz analysis, June 2026); for Perplexity it's second. OpenAI doesn't disclose the exact review-extraction mechanism, but citation data shows aggregator content genuinely enters the answers.

My reviews are in a third-party widget — is that a problem? If the reviews load via JavaScript in the browser, they don't exist for AI crawlers. The fix is server-side rendering, an HTML embed, or at least AggregateRating/Review markup in JSON-LD with a ratings summary.

Can ChatGPT decline to recommend my product because of reviews? Yes — the feature explicitly works with "common complaints" and ranks products by what customers say. Poorly handled complaints on Heureka or unanswered grievances can thus show up directly in your customer's purchase decision.

Will more reviews on my own site help me more, or on Heureka? Both, but differently: reviews on your own site must be in machine-readable HTML with schema markup for AI to see them at all. Reviews on third parties (Heureka, forums, YouTube) are weighted by the model as a more independent signal — and Heureka is among the most-cited Czech sources. The ideal is a combination.

How do I find out where I stand? You can run a quick test yourself (ask ChatGPT about your product vs. the competition). The complete picture — visibility across 4 AI platforms, review readability, entities and a competitive comparison — comes from a GEO audit.

Updated: July 3, 2026. Key sources: OpenAI (Shopping research, Powering Product Discovery in ChatGPT), Lupa.cz — AI citation analysis in the Czech Republic (June 2026), Profound — ChatGPT Shopping deep dive, Modern Retail, CNBC.

About the author: Marek Nocar is the founder of NocarStudio, a specialized GEO agency (founded in 2025 as the first GEO studio in the Czech Republic). He helps B2B companies and e-shops become visible in ChatGPT, Perplexity and Google AI Overviews. → More about Marek