Education & Young scientists

ChatABG: designing a Socratic AI chatbot to support undergraduate students of animal breeding

---

Abstract

Students increasingly use large language models (LLMs) for learning. However, the default behavior of LLMs conflicts with educational objectives. LLMs provide direct and helpful answers, while students should be encouraged to think and struggle first before reaching for help or instruction. This is particularly important for undergraduate animal breeding courses that require students to apply knowledge by selecting appropriate formulas, performing calculations, and interpreting results. Providing immediate answers reduces productive struggle, which might undermine the learning process. Although using LLMs by students might be harmful for learning, it also provides opportunities to improve the learning process. Research shows that students who received one-to-one tutoring performed about two standard deviations better than students taught in a typical classroom setting. Although individual tutoring by human teachers is expensive and hard to scale, we might be able to offer a personalized tutor to each student using LLMs. To support students and preserve the learning process, we developed ChatABG. This custom LLM-based AI tutor serves the Animal Breeding and Genetics (ABG) course at Wageningen University & Research. This Bachelor's course relies heavily on tutorials and practicals, where students are asked to recall course concepts and formulas, perform calculations, and interpret results. We designed ChatABG to 1) align notation and explanations with course materials, 2) use a Socratic tutoring style that guides students rather than giving answers, and 3) provide accurate support for calculations. The primary aim of ChatABG was to support students during tutorial and practical sessions, and help them prepare for the exam in a fun and engaging way. We used an iterative five-step workflow to develop ChatABG: 1) define capabilities and criteria, 2) plan evaluation using test materials, evaluation forms, and version control, 3) write the system prompt, 4) iteratively evaluate and refine the system prompt, and 5) field test the tool in a live course setting to gather feedback. We tracked system prompt versions using git, and each version was tested against evaluation criteria using 20 test questions. We field tested ChatABG by offering the tool to the student cohort of academic year 2025-2026. Usage frequency was tracked using a voluntary registration system. User statistics showed that the tool was most heavily used during the first two weeks of the course, and most conversations were registered on days that tutorials and practicals took place. In the course evaluation, students differed greatly in opinion of ChatABG. About 30% of students reported that ChatABG helped them to a great extent, while about 45% reported that ChatABG did not help them at all. The main challenge in developing ChatABG was to incorporate good didactic characteristics in the design of an AI tutor, rather than the technical implementation itself.

Keywords: 2026

How to Cite:

Duenk, P., Lont, D. & Bovenhuis, H., (2026) “ChatABG: designing a Socratic AI chatbot to support undergraduate students of animal breeding”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2342685. doi: https://doi.org/10.31274/wcgalp.24481

Rights: 1