What is it about?

Most of what we know about how people relate to AI chatbots comes from studies in one country, the United States. But do those results generalize? We took a widely-cited US study on ChatGPT users and ran it again, this time with 413 people from Bosnia and Herzegovina, Croatia, Montenegro, and Serbia. The results surprised us. In the US study, users who valued efficiency also found ChatGPT a bit creepy, as if the tool being “too good” felt unsettling. In the Balkans, that link vanished. Efficiency was just efficiency. No unease, no emotional charge. Playfulness, it turned out, built trust and reduced discomfort. Users tested the chatbot’s limits, joked around with it, and explored what it could do. US users seemed driven away by creepiness, but Balkan users kept coming back if they trusted the tool, regardless of any unease. Why the difference? When we read what people wrote in their own words, the picture was clear. For Balkan users, AI is a tool rather than a friend. They call it an “assistant,” a “secretary,” a better search engine. They verify its answers. They worry about AI’s effect on society but compartmentalize those worries and keep using it for practical tasks. We call this the instrumentalist mindset, and we think it is rooted in the region’s post-socialist history, where technology has long been evaluated by what it delivers rather than what it symbolizes. Anyone building AI products for global audiences should stop trying to make chatbots feel human. Users in these markets want reliable tools, transparent sources, and easy ways to check the AI’s work. The most trusted AI behaves like a good instrument, not a person.

Featured Image

Why is it important?

Nearly all research on how people interact with AI chatbots has been done in the United States and other Western, English-speaking contexts, yet the results are routinely treated as universal. Universalizing them is a problem. AI products built on Western assumptions are now being deployed to billions of users worldwide. Millions in Southeast Europe alone interact with Gemini through Android phones and ChatGPT in their daily work. If designers assume that “friendly,” human-like AI is what everyone wants, they may be building the wrong product for most of the world. Our study is the first to test an established model of AI chatbot adoption in the Balkans, and one of very few to examine generative AI perceptions in a post-socialist context. The main patterns fail here. Efficient AI does not feel creepy to Balkan users, and creepiness does not drive them away. Balkan users treat chatbots as tools, not companions. The EU AI Act is changing how AI is regulated, and companies are racing to localize AI products. Our results give concrete, evidence-based guidance for building AI that fits how non-Western users actually think.

Perspectives

I hope this paper nudges the field toward treating cross-cultural replication as a basic standard rather than a niche pursuit. Too often, results from US samples end up as the default assumptions in AI products used by billions. If our work makes even a few researchers pause before generalizing, or a few designers question whether “human-like” is really the goal, it will have done its job. I also hope it encourages more researchers from underrepresented regions to test Western models against their own contexts. The most interesting research happens in the gaps between what we assume and what we actually find.

Dr. Marko Galjak

Read the Original

This page is a summary of: Trust Over Creepiness: AI Chatbot Perceptions in Balkan Countries, International Journal of Human-Computer Interaction, March 2026, Taylor & Francis,
DOI: 10.1080/10447318.2026.2638548.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page