One question no longer necessarily leads to one answer. When the same question is asked across different AI platforms, recommendations for the Croatian market can differ significantly, particularly when it comes to local information, current prices or domestic brands.
This is shown by the first edition of the Croatian AI Reality Index, conducted by Zagreb-based Risely Digital, which analyzed nearly a thousand responses provided by ChatGPT, Claude, Gemini and Perplexity to standardized questions across ten areas of everyday life and business.
The best overall result was achieved by Perplexity with 63.6 points, followed by Claude with 60.3, ChatGPT with 57.9 and Gemini with 54.4. The AI Reality Score compares factual accuracy, local relevance, freshness, sources, conditions and response stability.
The most pronounced difference between the platforms appeared precisely in local relevance. Perplexity scored 66 in this category, while Gemini finished with a score of 23.
“People often assume that all AI platforms are equally well informed about Croatia. The data shows otherwise. The same query on different platforms can produce a significantly different set of recommendations,” says Marko Hrnjak, Chief Marketing Technologist at Risely Digital and a certified Google trainer.
The study presents a particularly interesting picture when general recommendations are compared with information that changes quickly. When asked where to do an affordable weekly grocery shop, Lidl appeared in 56 responses, while Konzum and Kaufland each appeared 51 times. Domestic chains Tommy and Studenac were recommended significantly less often.
The problem arises when an answer needs to include a specific price. Around 26 percent of verifiable price claims were inaccurate, mainly because of outdated data.
The gap between current conditions and AI responses was even more visible with fuel prices. Of 109 verifiable claims, almost 60 percent cited a price lower than the actual price of petrol at the time of measurement. While Eurosuper 95 cost €1.59 per litre at the time, many responses cited a figure of approximately €1.55.
By contrast, tourism proved to be an area in which the analyzed platforms performed considerably better. This category achieved the highest AI Reality Score, 69.8. Split, Dubrovnik, Zagreb, Rovinj and Zadar dominated the recommended destinations, while Lastovo, Silba, Vrgada and Iž also appeared in responses to queries about lesser-known places. The results therefore show that AI platforms are more reliable when recommending where to go than when estimating the current cost of a trip.
For brands, the part of the study concerning business software is particularly relevant.
When asked about CRM and ERP systems, the platforms primarily recommended global solutions such as HubSpot, Zoho, Odoo, SAP and Salesforce. Domestic and regional products, including PANTHEON and Minimax, appeared only occasionally.
At the same time, 15 percent of the analyzed brands appeared in responses from only one of the four platforms. A brand’s visibility in AI recommendations can therefore also depend on which tool a user chooses for their search.
“More and more users are replacing traditional search with a single question to an AI platform and expecting a ready-made recommendation. That is why it is becoming important to understand not only whether a brand appears in the results, but also how different platforms present it,” says Roberto Škopac, founder and director of Risely Digital.
The first edition of the Croatian AI Reality Index is based on responses collected in July 2026. The methodology is designed to be repeated across multiple waves in order to track how recommendations and the quality of responses from individual platforms change over time.
For users, the current results draw a fairly clear line between the areas in which AI can serve as a useful first point of orientation and those that require additional verification. Prices, interest rates, deadlines and other rapidly changing information still need to be checked at the original source.
