For the past several months, we've been running one of the most comprehensive analyses of AI-generated product recommendations in the Czech market. The question we started with was deceptively simple: when Czech shoppers ask AI models where to buy something, which brands actually show up — and does that match what Google says?
The answer turned out to be far more interesting than we expected. We tested 1,150 shopping prompts across 19 product categories, covering over 10,000 Czech e-commerce domains, and compared results from three AI models against two traditional search engines. What we found challenges some basic assumptions about the relationship between AI search and traditional SEO.
Below is a summary of the key findings from the Czech AI E-commerce Visibility Index — the full interactive report is available if you want to dig into individual brands and categories yourself.
The Gap Nobody's Talking About
Here's the number that stopped us cold when we first ran the analysis: only 39.6% of brands recommended by AI models also appear in traditional search results for the same queries. The Jaccard similarity index — a standard measure of overlap between two sets — sits at just 25.7%.
That means the brands Google surfaces and the brands AI models recommend are largely different populations. This isn't a minor discrepancy. The AI ecosystem and the SERP ecosystem are operating on distinct views of the Czech e-commerce market, and for most companies, optimizing for one doesn't automatically help with the other.
"Over 60% of AI brand recommendations have no counterpart in traditional search results for the same query. Ranking well on Google tells you almost nothing about your AI visibility."
The degree of divergence varies by category, and some of those differences are striking. Auto & Moto, Travel, and Home & Garden show roughly 54% overlap — meaning AI and traditional search at least partially agree on who the relevant brands are. But Electronics sits at just 11.1% overlap, and the E-shop Comparison and Cross-category segments are at 5.3% each. In electronics, AI search is effectively running a different show from Google entirely.
AI Search Is a Winner-Take-Most Game
If the AI-SERP gap is surprising, the concentration of AI recommendations is almost jarring. The single top domain — alza.cz — captures 4.6% of all AI brand mentions across every category and every model we tested. The top 10 domains together account for 20.2% of all recommendations. The top 50 take 36.8%.
This is a textbook power-law distribution — the kind you see in social media engagement, citation networks, and app store downloads. A small number of brands are capturing a wildly disproportionate share of AI visibility, while the vast majority of e-shops are essentially absent from the conversation. Notably, the remaining ~10,000+ domains share the other half of all mentions between them.
The practical implication is uncomfortable: a solid organic search presence doesn't translate to AI visibility. You could rank consistently in Google's top 10 and still be completely absent from AI-generated recommendations. Building AI presence requires a different playbook than SEO, and right now, the gap between those two playbooks is wide.
Who's Actually Winning Czech AI Search
Alza.cz stands alone. With 1,217 AI mentions and the top position across all models, it's the only Czech e-commerce brand that can genuinely be called an AI search leader — and it leads traditional search equally strongly. Below alza, the picture gets more complicated.
Mall.cz ranks second in AI mentions with 974, a number that would be impressive if mall.cz still operated as an independent Czech e-shop. It doesn't — it was absorbed into the Allegro group in 2023. Heureka.cz, a price comparison aggregator, comes third with 828 mentions. Traditional search tells a different story: zbozi.cz climbs from sixth in AI to second in SERP, while drmax.cz and benu.cz crack the SERP top 10 without appearing in AI's top 10 at all. These are meaningful divergences with real business implications.
The Ghost Brand Problem
One of the more unsettling findings in this research is what we've started calling "ghost brands" — e-shops that have closed, exited the Czech market, or ceased operations, yet continue to receive active recommendations from AI models. Shoppers who follow these recommendations land on dead pages, generic redirects, or sites that no longer serve Czech customers.
We identified eight defunct Czech e-commerce businesses still regularly recommended by at least one AI model. Together they account for 1,840 ghost mentions — nearly 7% of all AI recommendations in our dataset. Mall.cz (closed 2023), kasa.cz (ceased 2022), electroworld.cz, okay.cz, euronics.cz, nay.cz, bibloo.cz, and tesco.cz — which exited the Czech market entirely in 2022 — are all still appearing in AI-generated shopping advice.
The ghost brand problem is almost entirely a GPT-5 issue. Of the 1,840 ghost mentions, 1,378 — 75% — come from GPT-5. Gemini produces just 146 ghost mentions across all queries, Perplexity 316. GPT-5's training data appears to have a significant lag on Czech market realities, while the other models are either more current or better at grounding recommendations in live information.
This also creates a market distortion that's hard for live businesses to fight. If a defunct competitor is still collecting hundreds of AI recommendations per month, those slots are occupied and can't be claimed by brands that are actually open for business. It's a quiet form of AI-generated competitive damage.
Each AI Has Its Own Opinion
One thing that became clear quickly is that GPT-5, Gemini, and Perplexity are not interchangeable views of the same market. Each model has distinct biases, different category strengths, and a noticeably different relationship with real-world search behavior. Running your brand on one and calling it done is like reading one review and assuming it represents the market.
| Model | Total mentions | Ghost rate | SERP overlap | #1 domain |
|---|---|---|---|---|
| GPT-5 | 8,567 | 16.1% | 32.1% | alza.cz / mall.cz (tied) |
| Gemini | 7,965 | 1.8% | 50.0% | alza.cz |
| Perplexity | 10,154 | 3.1% | 50.3% | heureka.cz |
Perplexity generates the most recommendations overall (10,154) and aligns most closely with traditional search at 50.3%. This makes intuitive sense: Perplexity retrieves live web data as part of its responses, so its view of the Czech e-commerce market tends to reflect current conditions. GPT-5, by contrast, shows only 32.1% overlap with SERP — its recommendations appear to be anchored to an earlier snapshot of the market, which explains both the ghost brand problem and the divergence from live search behavior.
The model disagreement gets particularly extreme with specific brands. Kasa.cz — a defunct electronics retailer — has a GPT-5 to Perplexity mention ratio of 19.6x (GPT-5 mentions it 196 times; Perplexity, 10). Okay.cz, another closed brand, sits at 45.7x — 137 GPT-5 mentions versus 4 from Perplexity and 3 from Gemini. These aren't subtle statistical differences. The models are telling completely different stories about which brands exist and are worth recommending.
What This Means for Your Brand
A few things stand out after spending months in this data.
The AI-SERP gap is real and it's large. If you're an e-commerce brand with solid organic rankings and you haven't thought about AI visibility, you're likely missing a channel that's already routing purchasing intent. The 60%+ of AI recommendations with no SERP counterpart represent genuine consumer traffic that traditional SEO can't see or capture.
The concentration effect means most brands are starting from zero. The power-law distribution of AI mentions is steep. There's no gradual gradient where a brand ranked 30th is only slightly less visible than one ranked 5th — the drop-off is sharp. If you're not in the top tier, you're largely invisible in AI search, regardless of your SEO performance.
Model choice matters more than most people realize. Gemini and Perplexity are significantly more current and more aligned with real search behavior than GPT-5 on Czech market data. Brands optimizing their AI visibility need a model-by-model approach. What works on one doesn't automatically transfer to another, and your audience may be asking different models depending on where they are in their purchase journey.
The ghost brand problem is an active market distortion. Eight defunct e-shops are collectively occupying thousands of recommendation slots per month. For their former competitors, this is a real ongoing problem — not just a data curiosity. Until AI training cycles catch up with market realities, live brands in those categories are competing against brands that no longer exist.
This analysis was conducted in March 2026. We tested 1,150 shopping-intent prompts across 19 Czech e-commerce categories against GPT-5, Gemini, and Perplexity, then compared all recommended domains against Google.cz and Bing results for the same queries. 10,303 unique domains were identified across all results. Ghost brand status was manually verified against Czech business registries, company records, and public market data. The full interactive dataset is available in the Czech AI E-commerce Visibility Index.
Explore the Czech AI E-commerce Visibility Index
Search 10,303 domains, filter by category, and compare all three AI models.
Open the Index