How document-grounded AI can support inclusive learning

The rapid adoption of generative artificial intelligence (GenAI) across higher education has raised concerns around its impact on academic integrity and how to ensure students are engaging with reliable sources of information. One promising response is to use document-grounded AI. This approach restricts the AI model to a defined collection of inputs. These might include trusted academic sources such as lecture slides, assessment briefs, reading lists and academic literature selected by the user. 

The wider purpose is to engage students with credible academic materials in a flexible and interactive way to support inclusive learning. Students can use the AI to revisit content, clarify difficult concepts, and prepare for assessments and seminars at their own pace. 

• Here’s how to show students what responsible AI use looks like
• I start my lectures by introducing my teaching assistant: GenAI
• Five tips for using AI in university assessment

Before introducing AI, the first question should be: “What educational problem are you trying to solve?”

I used a document-grounded approach in my Principles of Marketing module, a course that involved about 500 first-year students. The intention was not to replace teaching or independent research but to provide another way for students to engage with trusted academic sources. The challenge was supporting a large cohort from different backgrounds. Some students lacked confidence in navigating academic expectations, interpreting assessment requirements and engaging independently with module content. The interaction with document-grounded AI provided an additional layer of academic support while complementing, rather than replacing, teaching and self-directed learning. 

From my experience, five practical steps can help academics integrate document-grounded AI into their teaching:

1. Start with identifying the sources

I used a two-layer approach to guide students through the research process before they engaged with AI. Students first worked with the core materials selected by the teaching team, then identified and evaluated additional academic sources through recognised academic databases. They uploaded these sources into NotebookLM and used the AI to ask questions, explore concepts and compare arguments. The AI also showed which uploaded sources informed its responses, allowing students to trace information back to the original evidence and understand how references supported different arguments. This sequence keeps students responsible for finding and evaluating evidence while using AI to deepen their engagement with it.  

2. Use AI to build confidence, not dependence

I integrated AI into seminars through short activities lasting 15 to 20 minutes. Students worked independently and then in groups, using lecture materials alongside additional academic sources they identified and uploaded to the AI. Rather than simply revisiting notes as part of regular revision, students could question the material, explore concepts and use their understanding to contribute to group discussions. I integrated AI into smaller seminar groups, allowing tutors to monitor students’ engagement throughout the 12-week module and provide support where needed. 

Tutors could also check the sources students were using and provide formative feedback during weekly activities. This gave less confident students opportunities to develop their understanding before contributing to discussions, initiating conversations and engaging more critically.

3. Develop skills to engage with information in different formats

I first used AI to create 10- to 15-minute video overviews summarising key concepts from one-hour lectures, giving students another way to revisit the content. During seminar activities, I encouraged students to use multimodal AI features themselves to convert longer academic articles into shorter formats, including audio overviews with interactive elements and mind maps. This helped students digest complex concepts in different ways. While the format changed, students continued to engage with the underlying academic content.

4. Introduce new ways to prepare for assessment

Document-grounded AI created a useful bridge between students and tutors during assessment preparation. Students could organise their research into a single source list or separate lists for different sections of the assessment. They were also encouraged to take notes during lectures and add these to the AI as sources. Students could create one or multiple notebooks for their assessment, with each retaining their sources and conversations with the AI. During seminars it gave tutors greater visibility of students’ progress, including how they were engaging with research, writing prompts and using AI responsibly without becoming over-reliant on it. Tutors could then provide formative feedback during seminars based on this activity. 

5. Widen access to academic support

In my module, supporting students from diverse educational and international backgrounds meant providing equitable access to academic support while recognising that students engage with learning in different ways. The document-grounded approach provided a consistent foundation of trusted academic support while offering flexibility in how students engaged with it. Students could also interact with the AI in different languages, reducing language barriers when engaging with complex concepts. 

Qualitative feedback indicated that students valued this approach, particularly for strengthening their understanding of complex concepts and preparing more effectively for assessment. Module outcomes were also encouraging, with the awarding gap reducing from 21.4 per cent to 18.2 per cent. Although these findings do not establish causality, they suggest that the approach may contribute to more inclusive learning as part of a broader pedagogical strategy.

The biggest lesson from my experience is not about the AI tool itself but how thoughtfully it is integrated into educational design. Document-grounded AI created clearer boundaries around how AI supports learning while providing more equitable and flexible academic support for a diverse student cohort.

Imran Khan is a senior lecturer in marketing in the department of management, business, and marketing at Birmingham City University.