Naohiro Tashiro (pictured left) is a board-certified family physician at the Department of General Medicine, Okayama University Hospital, Japan, with an interest in medical education and reflective practice in the age of AI.
Yuki Otsuka (pictured right) is a board-certified general physician and Assistant Professor at the Department of General Medicine, Okayama University Hospital, Japan, where he serves as director of undergraduate and postgraduate medical education.
About a year ago, I (NT) used AI while writing the portfolio required for my specialty certification. Much of the time, I used it to search the literature, but occasionally I asked it to critically look at a reflective entry I had drafted. I pasted in my draft and asked for critical feedback, and within seconds a question came back:
“You are treating one of the two things the patient showed — A and B — as if only one of them were true. If you held both as equally true, would your thinking change?”
The question highlighted how I had unconsciously simplified the patient’s complex picture. I rewrote the entry immediately. Another question arose:
“You have turned the sense of falling short — of not quite reaching this patient — into ‘a lesson for next time.’ But what if you sat with that feeling instead of resolving it? What would you write then?”
This exchange occurred many times within a single hour. I wrote, was questioned, rewrote, was questioned again — a loop turning faster than any dialogue with a busy supervisor could. None of the questions handed me an answer; each one pushed my thinking further than I would have taken it alone.
Dialogue with a supervisor is precious, but there are physical limits to how often and how long it can occur. AI was always available, ready to work through my reflections whenever I needed.
As a trainee, I was fortunate to have supportive supervisors. Yet they too were busy, and, as Lewis and Hayhoe have noted, time constraints and professional isolation were real barriers to my reflective practice.¹ Dialogue with a supervisor is precious, but there are physical limits to how often and how long it can occur. AI was always available, ready to work through my reflections whenever I needed.
A recent BJGP Life article by Nurani lays bare a fundamental problem of assessing reflection in the age of AI.² In an era when AI can generate reflective writing aligned to the RCGP curriculum, the assumption that the words in a portfolio mirror a trainee’s own thinking can no longer hold. I agree with this. However, this debate has largely been led from the assessor’s side of the table. I want to add the learner’s view, drawn from my own experience of using AI to write my certification portfolio: that the same AI, used differently, can deepen reflection in ways that human supervision alone cannot. Shah and Singh describe AI in a consultation not as a decision-maker but as a tool that “facilitated reflection,” creating the cognitive space to articulate unease and resist premature closure.³ My experience points the same way, in written reflections rather than the consultation itself.
Thus, what turns the speed of the loop into depth of reflection? The key is where the reflection begins. The use of AI in portfolios spans a spectrum: searching for literature and knowledge, organising one’s thinking through dialogue, and outsourcing the reflection itself to AI. We call all of these “using AI,” yet they are entirely different as reflective processes. Only when learners enter the loop as agents of their own reflection — responding to the AI’s questions and rewriting — does the reflection deepen. The moment reflection is handed over to AI, the loop stops.
The question to ask, then, is not whether AI was used but how learners engaged with it. What questions did they ask?
However, the finished product does not reveal where a learner stands on that spectrum. Between the learner who keeps the loop of reflection turning and the learner who hands reflection itself to AI, a wide gap may grow over time, which is invisible in the completed portfolio.
The question to ask, then, is not whether AI was used but how learners engaged with it. What questions did they ask? How did they rewrite their reflections in response to this? How many times did the loop turn? The evidence of whether reflection deepened lies in that process, not in the texts they produced.
AI can be a tool that strips reflection away or a partner that deepens it. For learners to wield this double-edged sword well, we need to understand what makes AI deepen reflection rather than replace it and what gets in the way. Before practice outpaces policy, we must study the process of reflecting with AI.
References
- Lewis M, Hayhoe B. The digital Balint: using AI in reflective practice. Educ Prim Care 2024;35(6):198-202. doi:10.1080/14739879.2024.2372606
- Nurani R. If AI can ‘reflect,’ what are we assessing? BJGP Life. 7 April 2026. Available from: https://bjgplife.com/if-ai-can-reflect-what-are-we-assessing/
- Shah R, Singh A. Using artificial intelligence to create meaning. Lancet Prim Care 2026; 2: 100129. doi:10.1016/j.lanprc.2026.100129
Author’s AI disclosure: Claude (Anthropic) was used for translation from Japanese to English and language editing. Paperpal was used for English language refinement. All content reflects the authors’ own ideas.
Featured photo by Inna Nasonova on Unsplash
