AI for medical educators: the production-side stack for faculty, consultants, and clinical teachers in 2026

How to think about AI as a clinical teacher

Medical education is a content-production problem wrapped in a curation problem. A single hour of teaching costs six to ten hours of preparation when you build from scratch, and most of that time goes into format work, not clinical thinking: laying out slides, drawing the same diagram for the third year running, writing distractors, formatting mark schemes. AI is excellent at format work and mediocre at clinical reasoning. Use it for the first, never trust it for the second without verification.

The mental model that works: AI drafts, the educator finishes. A drafted lecture deck, a drafted diagram, a drafted SBA bank, a drafted block of feedback. The clinical accuracy, the pedagogical judgement, and the trainee-specific personalisation are the parts only you can do, and those are also the parts the GMC holds you accountable for under the 2024 generative AI guidance for doctors.

The five highest-leverage uses for faculty

1. Teaching slide decks and lecture material

Gamma, Beautiful.ai, and Canva's AI presentation features turn a typed outline into a 20-slide deck in under two minutes. The structure they produce is usually serviceable. The medical imagery they generate is usually wrong: invented anatomy, mislabelled vessels, the wrong number of cranial nerves. Generate the structure, then strip out every clinical image and replace it with material you trust from Radiopaedia, BMJ Best Practice, or your own labelled diagrams. The deep workflow lives in ai-presentations-for-medical-school.

2. Original teaching diagrams

Sketch the structure, procedure, or pathology on paper or an iPad; a sketch-first tool renders it photoreal and labelled. The output is original to you (cleaner under most institutional IP policies than stock images), accurate (because you authored the topology), and reusable across slides, handouts, and exam papers without per-figure BioRender pricing. Angiosome is purpose-built for this; see ai-medical-illustration for the wider category.

3. SBA and EMQ question writing

GPT-5 and Claude Sonnet 4.5 are competent at first-draft single best answer items and distractors at preclinical level, weaker at finals-grade items where distractors need to be plausibly close. The realistic ratio is one keeper for every three drafts. Generate twenty, bin twelve, rewrite eight, ship five. You still save 60% of the time versus a blank page, and the items that survive curation are yours.

4. Feedback drafting for trainee reflections

AI can convert your bullet-point comments into structured prose feedback (strengths, areas for development, suggested actions) in seconds. Useful when you have 20 reflections to mark on a Sunday evening. Always personalise before sending: trainees can spot generic AI prose at fifty paces, and the trust cost of a flat reply is far higher than the time saved.

5. OSCE and OSCA station drafting

AI is excellent at producing station stems, mark schemes in a consistent format, and simulated-patient scripts. You curate the clinical content; the model handles the formatting and the consistent voice across a station bank. Saves a teaching half-day per station, especially when you are converting a clinical scenario you already know into examinable form.

AI tools for medical educators, by job-to-be-done

A realistic teaching-week workflow

The point of an AI workflow is not to use AI everywhere, it is to use it on the tasks where the format work dominates the thinking work. A realistic week for a consultant who teaches half a day:

1. Monday: outline next week's teaching in a text doc. Drop it into Gamma; generate a first-draft deck. List the diagrams you will need.
2. Tuesday (45 min): sketch each diagram on paper or iPad. Render via a sketch-first AI tool. Drop into the deck, replacing the stock medical images.
3. Wednesday (60 min): draft 20 SBAs in Claude or GPT-5 against the learning objectives. Bin half, rewrite the rest.
4. Thursday: deliver teaching. Capture three improvements for next round in a notes app.
5. Friday (30 min): AI-draft prose feedback from your bullet comments on 15 trainee reflections. Personalise each. Send.

GMC, institutional, and data-protection limits

The GMC's 2024 guidance on generative AI for doctors is the most relevant single document for UK faculty. It does not ban AI use; it makes clear the doctor remains accountable for any output used in patient care or training, and for the confidentiality of any data uploaded. In practice that means three rules: never upload patient identifiable information into a consumer AI tool; verify every clinical claim in AI-drafted teaching material before students see it; and disclose AI assistance in resources where your institution requires it.

Prompts worth saving in your notes app

The single highest-return faculty habit in 2026 is keeping a short library of prompts that you tweak per teaching block, instead of writing each one from scratch. Five worth stealing:

• 'Generate 10 SBAs for [topic] at [year level]. One best answer, four distractors, each distractor a common clinical misconception. Include a one-sentence rationale per option.'
• 'Convert these bullet-point comments on a trainee reflection into 150 words of structured feedback: strengths, areas for development, two specific suggested actions. Comments: [paste].'
• 'Draft an OSCE station for [scenario] including: candidate instructions, simulated patient script, examiner mark scheme with clear pass criteria, total 10 minutes.'
• 'Summarise this lecture transcript into a one-page student handout: learning objectives, three key concepts, two clinical applications, five recall questions. Transcript: [paste].'
• 'Re-write this paragraph at [year level] reading age, keeping the clinical accuracy. Flag any claim you are not confident about. Paragraph: [paste].'

Modelling good AI use to your trainees

If you teach trainees, they will adopt your AI habits, your shortcuts, and your blind spots. The right defaults at faculty level shape what an entire cohort thinks AI is for. Three habits worth modelling: cite sources alongside AI-drafted summaries, show the verification step you took before trusting an AI claim, and be explicit when a slide or handout was AI-assisted. The trainees who watch a senior clinician verify and disclose AI use are the ones who become safe AI-using doctors.

The inverse is also true. A consultant who pastes patient notes into ChatGPT in front of an FY1 has just trained that FY1 to do the same thing on call, without the legal cover. The cultural transmission is faster than any e-learning module the trust will roll out.

Sources