My Action Research Project involved substantial qualitative data collection: two focus group sessions, one group semi-structured interview and six individual semi-structured interviews, totalling over five hours of recorded dialogue. Given the compressed timeline of this ARP, I made the pragmatic decision to use AI transcription software to convert audio recordings into text.
To mitigate accuracy concerns, I adopted a verification process, where I listened back to all audio recordings in full, cross-referencing against the AI-generated transcripts to correct misattributed speakers and mishearings. This was particularly critical for the focus group discussions where multiple student voices overlapped, and for moments where students used discipline-specific architectural terminology that AI software frequently misinterpreted. I also manually cleaned all transcripts to remove automated timestamp markers and formatting inconsistencies, ensuring readability whilst preserving the authentic flow of conversation.
This hybrid approach, AI for initial transcription, human verification for accuracy, allowed me to balance methodological rigour with practical time constraints. The AI tool provided a foundation, but the quality of my analysis ultimately depended on my own careful listening, contextual understanding, and reflexive engagement with participants’ voices.
However, I want to acknowledge the methodological limitations of transcribing. Whilst automated transcription tools offer speed and efficiency, they lack the nuance, tone, and contextual understanding inherent in human conversation. Transcripts, be it manually or with AI, cannot capture hesitations, emotional shifts, laughter, or the pauses that often signal deeper reflection. All of these are elements that matter when analysing how students develop consciousness around discussion topics, such as inclusive design. As Braun and Clarke (2022) acknowledge, transcription is itself an interpretive act; what we choose to capture shapes what becomes analysable data.
A learning from my ARP, if I sought further research, would be allowing for more time to consider how I interpret conversations for analysis and how to share conversations beyond transcripts, but as engaging and knowledgeable content.
Please refer to ARP Blogs 4, 5, 6 and 7 for attached transcripts.
References
Braun, V. and Clarke, V. (2022) Thematic Analysis: A Practical Guide. London: SAGE Publications.