Abstract:

Artificial intelligence (AI) technologies may enhance, simplify, accelerate, and/or revolutionize disparate processes in education, ranging from the design of educational content to its evaluation. The same heterogeneity characterizes the tasks that teachers are expected to carry out; in both approaches, human and AI-enhanced, what remains crucial is the ability to act appropriately in terms of instructional design.

From a traditional perspective, teachers need to take the time to design instruction through the development of a lesson plan, selecting a methodology, objectives, and assessment and filling in a template. Even though this process may seem straightforward, it hides a series of difficulties, also when using structured methodologies. One of the challenges is ensuring that all the phases and activities of structured methodologies are respected. For instance, in team-based learning (TBL), one of the most common structured methodologies, a lesson is organized into a sequence of well-defined phases: pre-class preparation, individual readiness assurance test (iRAT), team readiness assurance test (tRAT), and team application (tAPP).

AI may be helpful in designing teaching activities based on structured methodologies. However, no studies have investigated the relationship between AI and instructional design based on structured methodologies, such as TBL.

Within this context, the present research investigates whether and how Gemini 3.1 Pro can accurately generate lesson plans for teachers according to TBL. The hypothesis is that this model can accurately generate such a document, for instance, by listing the phases in the correct order with complete and effective material for iRAT, tRAT, or tAPP. The aim of this study is to preliminarily test the model in order to critically integrate AI into teaching design with structured methodologies such as TBL.

From a methodological point of view, Gemini 3.1 Pro was selected because it has been considered the best model in terms of accuracy in pedagogical knowledge. One of the most common lesson plan formats used in Italy was inserted into a prompt and submitted to Gemini using the zero-shot prompting technique. Specifically, the prompt was submitted five times in order to balance the variability of the large language model’s output. The final results were then evaluated by three independent experts using a personalized rubric and scores were then averaged to obtain a final value. Ultimately, the results were compared and discussed in light of the literature.

Keywords:

Artificial Intelligence, instructional design, lesson planning, team-based learning.