Mark Cooper is a teacher working at Newfriars College, a specialist post-16 college in Stoke-on-Trent providing pathways into adulthood for young people with a broad range of disabilities and learning difficulties. Mark is a dyslexia advocate with lived experience. The college’s TechAbility champion, Mark describes himself as an ‘EdTech Pragmatist’!
The tension between doing and proving
I am a teacher working in Stoke-on-Trent, and I view the world through a very specific lens: I am Dyslexic and Neurodivergent. In the Specialist Further Education (FE) and SEND sectors in which I work, the daily administrative demands on an educator can be demanding – paperwork, tracking, and evidence gathering – but are necessary for quantitative quality assurance. Yet every minute spent buried in a laptop is a minute taken away from our learners.
I wanted to find a way to reduce admin to spend more time with my learners and so I started to explore if I could use a technique to convert recorded spoken notes into text using AI. My journey developing this technique didn’t happen overnight. It was born out of a need to build a bridge. I needed a bridge between the dynamic, messy, beautiful reality of SEND delivery and what can be the rigid, structured requirements of quality processes. This is how we used technology not to replace the human element, but to protect it.
Purpose & philosophy: more than just ‘admin’
Quality in specialist provision isn’t about ticking boxes; it’s about capturing the nuance of progress. In our sector, progress isn’t always linear. It’s not a straight line on a graph; it’s often a complex web of small victories.
At Newfriars College, we use Evidence for Learning to track these victories against Education, Health and Care Plan (EHCP) outcomes. The challenge is how to capture high-quality, person-centred evidence while still prioritising the learner.
For example, when in the community, supporting a student in a retail environment, I was not able to stop, pull out a laptop, and type up a detailed observation. Previously, this meant relying on memory to write notes hours later, which inevitably leads to a loss of detail and quality. However, the AI approach outlined below makes for a better outcome for all.
The strategic journey: capturing the lightning
We started asking: How can we capture the ‘lightning’ moment of a learning breakthrough without disrupting the session or the learner on placement?
The breakthrough came when we combined three elements: wireless technology, dictation built into Microsoft Word, and Generative AI. We moved from simply ‘recording notes’ to a workflow I call Contextualised Speech-to-Structured-Text AI (CSST-AI).
The goal was clear:
- reduce workload for staff
- improve the quality and consistency of feedback
- ensure that evidence is rigorously mapped to learner outcomes.
The process: a workflow for quality
Here is how the bridge between the dynamic, messy delivery and the structured quality process works in practice. It transforms raw, spoken feedback into structured, high-quality evidence.
Analysis: AI is not perfect
This process sounds magical, but it requires strict quality control. AI is not perfect!
The hallucination risk
Early on in developing this process, we identified the risk of ‘hallucinations’ – where the AI invents details to fill gaps. To combat this, our custom prompts have rigid constraints built in. The instructions explicitly state: “Only use information present in the raw text. Do not make assumptions. If a learner is not mentioned, state ‘No feedback given’.”
Quality in, quality out
The AI acts as an editor, not an author. If I provide vague or minimal verbal feedback – such as simply saying “Good job” – the AI cannot turn that into high‑quality, meaningful evidence.
The system helps staff strengthen their skills in giving accurate, attentive, and specific observations. To generate strong output, my verbal feedback must also be strong. For example, instead of saying “Good job,” I should say, “Student X, good job using the scanner correctly to price the items without staff support.”
This level of detail enables the AI to produce clear, useful, and learner‑focused evidence.
As the CSST-AI system evolved, so did the specificity and quality of the verbal feedback give to learners. The technology encourages us to be more precise, reinforcing best practice in delivering meaningful praise and constructive feedback.
Practical tips for giving effective verbal feedback are included in the infographic below.
AI requires detailed input for meaningful analysis. Avoid vague positive sentiment.
Directly reference Education, Health and Care Plan targets in verbal feedback to facilitate structured evidence mapping.
The AI cannot infer missing information. Dictate all relevant facts, names and context.
Use full names for AI accuracy, while strictly adhering to data privacy protocols in public settings.
Leverage wireless technology to record feedback instantly, preventing loss of critical detail over time.
The human in the loop
I never remove the human from the loop. The AI produces a draft; the teacher produces the final record. We must check for errors, such as the AI confusing two students with the same first name (solved by using full names).
Impact stories: changing lives and workloads
The neurodivergent teacher
As someone with Dyslexia, my specific learning difficulty manifests most severely when I am tired. At the end of a long day, my ability to write coherent, high-quality observations diminishes. I might miss details or make spelling errors. This workflow has made a big impact to my own practice. It allows me to ‘write’ at the speed of speech. It removes the friction between my thoughts and the page, ensuring that my administrative output reflects my actual teaching ability, regardless of any barriers.
The Job Coach
This approach has supported a Job Coach working with interns in a busy supermarket, where there is little time or space to stop and type notes. Previously, important details were often lost by the end of the day. Using a wireless microphone, the coach now captures evidence on the move, with AI later turning these observations into clear, mapped records of growing independence.
Safe practice remains central. In live retail environments, the coach needs to be alert to who may be listening in or looking over a learner’s shoulder, ensuring no personal information is exposed. She uses Outlook on her work mobile phone to record the speech‑to‑text block and processes the raw feedback away from other people. These routines embed strong data‑protection habits and ensure the method is both effective and responsible.
The learner’s voice
We haven’t just kept this tool for staff. We have used a similar workflow to empower learners with significant literacy barriers – although the CSST-AI process itself is always run by staff to maintain professional oversight and safeguard good practice. I worked with a learner who wanted to contribute to the college newsletter but struggled with writing. The blank page was demoralising; it eroded their self-esteem. By using speech‑to‑text with AI restructuring, they were able to speak their thoughts, which the AI then refined into a newsletter article. Seeing their words in print gave them a sense of pride and confidence that was truly ‘magical’. It wasn’t about cheating; it was about accessibility – removing the barrier of mechanics to reveal the intellect underneath.
Conclusion: continuous improvement
This journey has been about far more than saving time – though the reduction in staff workload and stress is significant. At its core, it is about quality. By automating the structural work of quality assurance, we create the mental space for what truly requires human empathy: connection, teaching, and understanding.
The process is still evolving. We continue to adjust our prompts to ensure they stay aligned with our curriculum and QA cycles. But the central lesson is clear: technology should never add to our burden; it should carry the weight so we can better support learners.
My advice to other colleges? Don’t seek tools that replace professional expertise – choose tools that amplify it. Invest in a set of wireless microphones, experiment with prompts, and begin building your own bridges between real‑world practice and high‑quality evidence.