For our March community meetup, we were joined by Craig Scott, Hydrogen & Welding Tutor and IQA at Riverside College. Craig delivered an informative session taking us through a practical workflow using AI tools to refine resources in order to support teaching, retrieval and closing knowledge gaps.
Continue for a recap of the session and discussions, you can also access the full recording and transcript on our YouTube channel.
The challenge
Craig started by explaining the challenge he faced last September – four months to teach students eight health and safety learning outcomes, at the end of which they would take an online multiple-choice exam with questions sourced from an expansive question bank. A key challenge would be parsing the volume of material provided by the awarding body, prioritising content that learners would most need to be taught directly. He noted the material was often in PowerPoint format and could be quite text-heavy, providing a high volume of information but in a way that students may find difficult to process.
The Workflow
Analysing source materials
The first step in the workflow was to use Copilot to analyse the material, prompting the tool to review the resources and help him pinpoint the content with the highest priority for teaching. Craig emphasised that he uses Copilot for all steps involving uploading materials or data like this, as their college’s institutional license provides the required data security.
Building focused teaching resources
The next step was building the teaching resources, and Craig shared a series of tips for creating and refining content using a mix of AI tools.
For each learning objective, Craig used ChatGPT to craft a strong prompt for Gamma, an AI tool that generates content including presentations and even web pages from the user’s prompt. Gamma could produce a more focused PowerPoint in under a minute. He then uploaded the output from Gamma back to ChatGPT though to quality-check it against the original criteria, an important step since the first output did not always fully match what was needed.
Craig also explained his use of NotebookLM to transform written content into infographics, providing a clearer, visual way to access the often dense information. He had also experimented with its ability to create videos and shared examples of two types of video output, a standard subscription video and a higher subscription tier “cinematic video” – the difference in quality was noticeable. He noted though that the feature that generated the most excitement is the interactive podcast mode, where students can interrupt a generated audio podcast mid-playback to ask a specific question and get an immediate, contextual answer. Craig described this as genuinely valuable for learners who want to get to the crux of a topic without listening to a full recording.
Reviewing outputs
Members were interested in how well these tools represented health and safety and other industry-specific visuals in the imagery and content generated. Craig shared that image generation could indeed still be a bit hit and miss, but Gamma helpfully comes with the option to use royalty free web-based images and there is also the ability to add your own images.
The question of the accessibility of generated materials was also raised. This was recognised by community members as a challenge with many tools which do not automatically create things like alternative text or use accessible colour palettes. Craig advised that outputs certainly may need alterations for accessibility, which can often be done within the tool itself. Adding accessibility requirements to the prompts was also advised by members with similar experience.
These conversations brought us to an interesting point around managing our expectations of AI tools, and the continuing importance of having a human review any generated content.
Checking for understanding
Craig then used Copilot to analyse the new PowerPoints and generate multiple-choice and open-ended questions from them to test students’ understanding of the content after he has taught it.
He explained how he has used platforms like Microsoft Forms and Nearpod to deliver questions. Nearpod, he noted, was great for in class use, as it allows students to complete assigned questions live on individual devices with the results visible to him as they come in. Allowing him to see all student responses in real time on a single screen, making the knowledge gaps across the class visible but without putting individual students on the spot.
Craig mentioned a particular highlight was Nearpod’s Time to Climb feature: a time-based quiz where students race to answer questions correctly and quickly, climbing a leaderboard in real time. Craig described it as genuinely changing the energy in the room, with “competitive retrieval” raising attention in a way a standard quiz does not.
Analysing the gaps
After the lesson the Nearpod reports and Forms results were then uploaded to Copilot, which Craig used to help him find and group the weakest question areas, a process that used to be manual and time-consuming. Craig shared one of the prompts he used, asking Copilot to sort questions by highest percentage incorrect and identify patterns. He emphasised the value of adding a final instruction for Copilot: “Ask me no more than two questions in order to do your best work.” This kept the interaction focused rather than spiralling into many back and forth iterations.
Closing knowledge gaps and reassessing
With this data Craig could then use ChatGPT again to create a new, targeted prompt for Gamma. Creating a new resource which was focused on the student’s knowledge gaps. A key feature of the new presentations was asking Gamma to provide explanations for why incorrect answers selected by students were incorrect and similarly providing justification for why the right answer was the correct one.
Using these new resources the weaker areas could be re-taught, and students tested again using the original questions to see whether the knowledge gaps had been fully addressed. If there were gaps still present then the previous steps in the workflow could be revisited, though Craig mentioned this was rarely needed. When ready the cycle could be used again focusing on the next learning outcome
What did AI add?
Craig ended by reflecting on what AI had added compared to a workflow without it. He noted it had helped him sharpen decisions about which content to prioritise, and the infographics and student-generated questions brought real engagement to a subject, health and safety, that can sometimes be a tough sell. Quiz data also became something he could act on, rather than information that sat unanalysed after a lesson.
He was clear that generated content always needed to be reviewed and iterated on, and that across all the tools he used, the same principle held: AI works best as a thinking partner, supporting teaching decisions rather than replacing them.
With his first cohort of 16 students, 87% passed at merit or distinction, with 25% achieving distinctions, and he has since repeated the approach with several classes with similar results. “It didn’t just help me make better resources and save time,” he said. “It genuinely helped learners achieve stronger results.”
Thank you to Craig for joining us and delivering a fantastically insightful session full of practical, replicable ideas and tips.
Join us in April
Our next AI in FE community session will be on 28 April, 12.30–1.30pm – back to our usual time. This one will be an interactive AI Ideas Exchange: a discussion session where we crowdsource topics via Padlet, vote on what to discuss, and break into smaller focused groups to explore the chosen topics.
As always, sessions are open to Jisc FE member institutions. Join our AI in FE Jiscmail list to receive invites.
If you would like to present a future AI in FE community session, please get in touch via ai@jisc.ac.uk.
Find out more by visiting our Artificial Intelligence page to explore publications and resources, learn more about our communities and sign up for our AI Literacy training.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk