GenAI Integration into Education: An Evaluation Framework
The growing presence of generative AI (GenAI) in university classrooms raises important questions about its pedagogical value, its alignment with institutional goals, and the conditions required for its responsible adoption. That’s why we at the Leiden Learning and Innovation Centre (LLInC) have developed a practical framework to help evaluate AI tools and understand their potential in educational contexts.
This project used the Community of Inquiry (CoI) framework (Garrison, Anderson, & Archer, 2000), which emerged from our literature review as the leading model for understanding meaningful learning in technology‑enhanced environments.
The Col model conceptualises meaningful learning as the interplay of three core elements: teaching presence (how instructors design, facilitate, and support learning), social presence (how learners communicate, connect, and build a sense of community), and cognitive presence (how learners construct meaning and engage in critical thinking). In addition to the literature review, we analysed institutional guidelines, national frameworks such as the Npuls EduGenAI Framework, and insights from our teacher professional development programmes (including BKO).
Using thematic analysis as well as open and axial coding, we identified the most frequently occurring themes and organised them according to the three presences of the Col model. Some recurring themes did not fit within these presences, particularly those related to ethical standards such as transparency and bias. To account for this, we added a fourth dimension to our model: ethical presence. Each presence then became one dimension of our framework, with the identified themes forming the categories within each dimension.
In our designed sessions, we created practical, pedagogically meaningful scenarios and AI-supported use cases to test and refine the rubric, ensuring it accurately reflected educational practice and challenges. These were used to test the clarity, relevance, and discriminative power of our rubric indicators and band descriptors across the four Col dimensions (Teaching Presence, Cognitive Presence, Social Presence, and Ethical Presence). Each AI tool (LUCA, Copilot, and Perplexity) was evaluated across the four previously defined dimensions using a three-level Likert Scale in relation to the designed scenarios. Qualitative notes complemented these results to capture nuances in performance.
Results and Implications
The project resulted in a practical framework which helps teaching staff understand the potential of GenAI tools in supporting their teaching and enhancing student learning. This framework offers a meaningful way to make informed, responsible choices in higher education when it comes to GenAI.
The results show that the particular GenAI tools under analysis in this project can improve teaching and assessment by providing consistent criteria-based feedback that aligns with learning objectives. In doing so, these tools support instructional and pedagogical alignment by reinforcing the coherence between intended learning outcomes, learning activities, and assessment practices. When feedback is systematically anchored in predefined criteria and outcomes, it helps students better understand performance expectations and directs their learning in ways that are congruent with the course design. For instructors, such alignment facilitates and supports reflective teaching by making explicit how instructional decisions translate into student learning and intake.
Additionally, our evaluation indicates that while these AI tools may provide accurate and reliable content, their ability to stimulate higher-order thinking, scaffold learning, and promote learner autonomy is inconsistent. Further, our findings indicate that supporting student independence remains challenging. Most of the analysed GenAI tools do not give students many chances to make choices, ask questions, or solve problems on their own. If teachers want to integrate GenAI effectively, they should use it in ways that encourage students to take ownership of their learning rather than simply receiving ready-made answers.
Moreover, the findings show that current GenAI tools have significant limitations in supporting inclusivity, equity, and differentiation. While these tools can offer basic explanations and step-by-step support, they cannot adequately tailor learning to individual needs, especially for students with sensory or cognitive differences. This means teachers remain essential for actively reviewing and adapting AI-generated content to ensure it is inclusive, accessible, and culturally responsive. Additionally, GenAI cannot replace the emotional and motivational support which human teachers provide. Although it can generate encouraging messages, it cannot fully respond to students’ frustration, emotions, or personal contexts. Teachers are still needed to create a supportive learning environment through empathy, encouragement, and relationship building.
Finally, the findings of this project demonstrate that GenAI tools raise several ethical and practical concerns. They often lack transparency, making it difficult for teachers and students to judge the accuracy of AI-generated information. This means teachers must guide students in verifying sources and using GenAI responsibly. Since data privacy and intellectual property are also risks, clear rules are needed on what students can safely share.
For educators, programme leaders, or institutions interested in this work, the full framework, including the rubric, scenarios, and methodological background, is available upon request. Workshops, presentations, or collaborative discussions can also be arranged to explore how this framework can support GenAI adoption within courses, programmes, or faculty-wide initiatives.
Feel welcome to reach out to Zohre Mohammadi Zenouzagh (z.mohammadi.zenouzagh@llinc.leidenuniv.nl) if you would like to engage further, adapt the rubric for your context, or explore future research collaborations!
