Artificial intelligence (AI) chatbots are increasingly used in higher education, particularly in STEM disciplines, where they assist students with programming, mathematical problem-solving, and research tasks. This study investigates the adoption, application areas, and perceived effectiveness of AI chatbots among undergraduate STEM students at University North in Croatia, focusing on those in their first three years of study. The research includes students enrolled in Electrical Engineering, Computing and Informatics, Logistics and Sustainable Mobility, Mechatronics, Multimedia, Design and Application, Mechanical Engineering, Civil Engineering, and Geodesy and Geomatics programs.

Data were collected through an online survey, where students reported their chatbot usage patterns, the challenges they encounter, and their perspectives on the accessibility of free versus paid chatbot versions. The results indicate that a significant number of students rely on AI chatbots for coding assistance, understanding complex concepts, and preparing for exams. However, they also highlight concerns regarding the accuracy of chatbot-generated responses, the depth of explanations, and the limitations of free versions. Many students find the cost of premium versions prohibitive, making affordability a key barrier to accessing advanced AI features. Despite these challenges, most respondents perceive AI chatbots as valuable tools that enhance their learning experience.

Survey results reveal that ChatGPT is the most frequently used AI chatbot, followed by DeepSeek, Gemini, Copilot and Perplexity AI. The primary applications include solving mathematical problems, programming, and researching academic topics. While many students report improved understanding of STEM concepts through chatbot use, concerns persist regarding inaccurate responses, lack of specialized knowledge, and surface-level explanations. Additionally, the majority of students believe that advanced AI chatbot versions should be freely accessible for academic purposes, with many emphasizing that chatbots should be free specifically for students.

The widespread use of AI chatbots has unexpectedly forced educators to rethink traditional assessment methods. As chatbots can generate quick and often well-structured answers, standard written exams and take-home assignments are increasingly vulnerable to AI-assisted solutions. In response, lecturers have begun shifting towards more interactive and applied assessment formats, including oral examinations, project-based assessments, and problem-solving tasks that require deeper conceptual understanding rather than rote memorization.

This study provides empirical insights into how undergraduate STEM students at University North utilize AI chatbot technology, as well as how their usage is reshaping academic evaluation. The findings offer valuable implications for educators and developers seeking to optimize AI-assisted learning environments while maintaining academic integrity.

Keywords: AI chatbots, STEM education, higher education, student learning, research tasks.