Generative AI has rapidly become part of higher education. Many students are already using tools such as ChatGPT, Claude, Gemini, or Mistral to support their studies, whether for brainstorming ideas, understanding difficult concepts, or obtaining feedback on their work. At the same time, universities face important questions around academic integrity, privacy, trust, and how AI should be integrated into learning.
These questions motivated our latest publication in Computers & Education: “A five-study design-based research approach to co-design an institutional AI digital assistant”, which reports on a 13-month research and development journey involving 315 students and 20 staff at The Open University.
Free access to the article (50 days): Read the full paper
Why build an institutional AI assistant?
Most discussions about AI in education focus on publicly available tools such as ChatGPT, Claude, Gemini, or Mistral. However, universities have unique learning materials, pedagogical expertise, and an understanding of their learners that cannot easily be replicated by general-purpose AI systems. At the same time, there are legitimate concerns about sharing institutional content and student data with commercial platforms.
Our research therefore explored whether an institutional AI Digital Assistant (i-AIDA) could provide students with trustworthy, curriculum-aligned support while operating within a framework of institutional governance, privacy, and academic integrity.
Rather than developing a system behind closed doors, we adopted a design-based research approach, working iteratively with students and staff throughout the entire development process. Over five studies, participants shared their expectations, tested prototypes, provided feedback, and helped refine the system.
Students want more than just another chatbot
One of the most surprising findings was that students were not primarily interested in having another chatbot.
While conversational support was appreciated, participants consistently valued features that directly supported learning. In particular, students responded positively to:
- Flashcards for revision
- AI-generated quizzes
- Formative feedback on learning activities
These features helped students test their understanding, identify gaps in their knowledge, and prepare for assessments. In fact, the quiz and flashcard functionalities often received more positive feedback than the chat function itself.
This finding challenges the common assumption that the future of educational AI lies purely in conversational interfaces. For many learners, structured learning support may be more valuable than open-ended dialogue.
Trust, privacy and academic integrity matter
Students clearly saw advantages in having an institutionally developed AI assistant. They appreciated the idea that responses could be grounded in official university materials rather than drawing from unknown internet sources. Participants also valued institutional oversight regarding privacy and data governance.
At the same time, students raised important concerns about:
- Academic integrity
- Data privacy
- Ethical use of AI
- The impact of AI on learning
- Potential reductions in human interaction
These concerns did not disappear during the project. Instead, they highlighted the importance of transparency and of positioning AI as a complement to human support rather than a replacement for educators.
Experience changes perceptions
Perhaps the most interesting result emerged during the final beta-testing phase. Some participants initially expressed scepticism about AI in education. However, after interacting with a functioning version of i-AIDA and experiencing how it could support learning, participants became significantly more positive about its usefulness. The majority of students who tested the system believed it would benefit their studies, and many rated it favourably compared to other generative AI applications.
This finding suggests that conversations about educational AI should be grounded in direct experience rather than hypothetical assumptions. What people imagine AI might do and what they experience a well-designed AI system doing can be very different.
Eight design principles for educational AI
Beyond the development of i-AIDA itself, a key contribution of the paper is a set of eight evidence-based design principles that emerged from the research. These principles emphasise:
- Supporting learning through curriculum-aligned activities and feedback.
- Using institutionally governed AI systems.
- Ensuring ease of use for learners with varying AI experience.
- Addressing privacy and academic integrity concerns transparently.
- Providing real-time, 24/7 support.
- Adapting to disciplinary contexts and learner characteristics.
- Designing for accessibility and usability through iterative testing.
- Involving students directly in the co-design process.
Collectively, these principles offer practical guidance for institutions seeking to implement AI in ways that enhance learning rather than simply introducing new technology for its own sake.
Looking ahead
While our findings are encouraging, they should not be interpreted as an argument that AI is universally welcomed by all learners. Throughout the project, a proportion of students remained strongly sceptical about AI in education. These perspectives provide valuable insights and remind us that successful educational innovation requires engaging with concerns as well as enthusiasm.
The broader lesson from this work is that educational AI should not start with technology. It should start with learners, educational goals, and thoughtful design.
Rather than asking “How can we use AI?”, perhaps the more important question is:
“How can we design AI that genuinely supports learning?”
Our work on i-AIDA represents one attempt to answer that question, and we hope the lessons learned will be useful to educators, researchers, learning designers, and institutions navigating the rapidly evolving landscape of generative AI in higher education.
Leave a Reply