AI Research Still Ignores the Clinicians Who Need It Most

Key Takeaways

• A scoping review discussion put a hard number on the AI CPD gap: only 3% of 278 AI-in-medical-education papers addressed continuing professional development.
• The micro-learning discussion was narrower, but operationally useful: short resources work best when placed inside existing meetings, apps, and clinical routines.
• For CME providers, the common thread is not “make shorter AI content.” It is to design for practicing clinicians’ workflow, then measure application rather than attendance.

A BEME scoping review discussed this week found that only 3% of 278 AI-in-medical-education papers addressed CPD, while about half focused on undergraduate medical education. The evidence posture is narrow — one underlying review, discussed in educator-led formats — but the mismatch is directly relevant to CME providers: AI learning is being studied where trainees are, while its most immediate use may be where practicing clinicians work.

AI education is still built around trainees

In the PAPERs Podcast discussion of the BEME review, the panel described a literature base of 278 papers, with CPD representing only a small fraction of the corpus (source). A parallel YouTube version of the same discussion framed the finding as a gap in how the field is studying AI’s educational value for busy practitioners, not just students (source). The discussion occurred in an oncology-adjacent education context, but the CPD implication is portable across specialties. For CME teams, the point is not simply that AI belongs in more activities. It is that CPD needs its own evidence base: how clinicians verify AI outputs, when they use AI inside a clinical workflow, what risks they notice, and what behavior actually changes after education. That extends an earlier brief on AI ethics training, but sharpens the question. Ethics and verification competencies may be widely endorsed, yet they remain under-studied in CPD settings. The concrete implication: audit AI-related education for whether it is measuring practicing-clinician behaviors, or just borrowing trainee-oriented assumptions.

Shorter formats need a workflow home

A separate educator-led Faculty Factory conversation gave the delivery-side version of the same problem. The discussion centered on 10- to 15-minute micro-talks, just-in-time teaching tips, app-based resources, and inserting brief segments into meetings that already exist rather than asking clinicians to attend another standalone session (source). This is a single academic faculty-development source, so it should not be treated as broad clinician consensus. Still, it names a format constraint CME providers already feel: time pressure changes what counts as usable education. The most useful detail was not the duration alone. It was the combination of a short resource, a specific use case, and a chance to apply it. The discussion also distinguished usage analytics from learning: opens, downloads, and time-in-app are not the same as improved teaching or practice behavior. For CME providers, micro-learning should not become a library of tiny assets with thin outcomes. The concrete question is: where will the learner use this, what will they do differently, and how will the provider know?

What CME Providers Should Do Now

• Separate AI activities for practicing clinicians from trainee-oriented content, and define CPD-specific outcomes before launch.
• Convert one long-form activity into 10- to 15-minute segments tied to a concrete clinical or team workflow.
• Pair each micro-resource with a practice prompt, verification step, or follow-up question that can be measured beyond attendance.

What CME teams should reconsider

This week’s signal is not broad, but it is useful. CME providers are being handed two linked cautions: do not assume AI education research answers CPD questions, and do not assume shorter content changes behavior by itself. The opportunity is to design from the realities of practice first — time, workflow, verification, and measurable application — and then decide what format and technology belong there.