Appendix D — Review personas

The six readers every chapter is reviewed against, from curriculum/personas.md in the repository. See How this book was made for how they are used.

Six readers to hold in mind when reviewing any chapter. Three are students; three are the people who would adopt, adapt, or clear the course. Review a chapter by reading it once as each persona and answering that persona’s questions. A chapter passes when none of the six would put it down.

Names are placeholders. Every student is a fourth-year medical student; see design-constraints.md for what that fixes.

D.1 Students

D.1.1 Marcus — the early adopter

MS4 matching into diagnostic radiology, the specialty where the “AI will replace you” talk is loudest, and he has decided to be on the winning side of it. Uses ChatGPT or Claude daily: differentials, drafting notes on rotations, studying, emails. Pays for a subscription. Has already been told once by an attending not to paste anything into it, and did not fully understand why. Wants to try every tool on the day it launches and assumes the course will show him more of them.

What he needs from a chapter: a reason to slow down that is not a lecture on caution. He learns from being shown a failure he did not expect in a tool he already trusts.

His questions on review:

Is there anything here I could not have found by asking the model itself?
Does the chapter show me a concrete case where my current habit gets the wrong answer, or does it just tell me to be careful?
Does it treat me as a user to be restrained, or as someone who could evaluate a tool for the team?
Where is the thing I can do with this tomorrow?
When the chapter talks about my specialty’s future, does it cite anything, or is it vibes in either direction?

Fails the chapter when: it is all warnings, or it explains something he already does at a slower pace than he does it.

D.1.2 Elena — the avoider

MS4 matching into general surgery. First in her family in medicine; English is her second language, and she reads technical prose slower than she thinks. Has not opened a chatbot in a year. Tried one in MS2, got a confidently wrong answer on a pharmacology question, and concluded it is unreliable for anything that matters. Has fixed ideas about what AI is good and bad at: good at summarizing, bad at reasoning, dangerous for patients. Some of those ideas are right, some are two years out of date. Uses only free tiers, on a five-year-old laptop, and will not give a phone number to sign up for anything. Privately worries that peers who use these tools are getting ahead, and that admitting she does not use them looks like falling behind.

What she needs from a chapter: a low-stakes way in, a clear statement of what the tool cannot do, and permission to keep her skepticism as a skill rather than shed it as a bias.

Her questions on review:

Does the chapter assume I already use these tools? If so, where do I start?
Is the first activity something I can do in ten minutes on a free account that does not ask for my phone number?
Does it acknowledge that my fear is partly justified, and separate the justified part from the outdated part?
Is there a moment where being skeptical is the right answer, not the obstacle to overcome?
Do the results depend on how fluently I phrase the prompt, and does the chapter say so?
Do the comics and pictures make sense without already knowing the field, or are they in-jokes?

Fails the chapter when: it opens with a demo she cannot follow, uses vendor vocabulary without defining it, requires a paid tier, or reads as enthusiasm.

D.1.3 Priya — the clinician-in-waiting

MS4 matching into emergency medicine. Neither excited nor afraid of AI. Has already met it in the EHR without being asked: on her sub-internship she silenced a sepsis alert that fired on a patient who obviously did not have sepsis, and watched an attending send an AI-drafted inbox reply without reading it. Nobody told her whether either of those was fine. Uses chatbots a little. Has one question: how do I use this without hurting a patient or getting myself in trouble. Cares about workflows, consent, documentation, and who is responsible when the tool is wrong. Has no interest in how models work beyond what she needs to answer that question.

What she needs from a chapter: the practical rule, the reason behind it, and the boundary case where the rule breaks.

Her questions on review:

If I skip every section about how the technology works, does the chapter still tell me what to do on the wards?
Is there a scenario I recognize from a rotation?
Does it say who is accountable, and does it say what to document?
What is the one sentence I would tell a co-intern?

Fails the chapter when: the “why it matters” is about the field rather than about her patient, or the practical guidance is buried under mechanism.

D.2 Faculty and staff

D.2.1 Jordan — the instructional designer

Works in the medical school’s education office. Not a clinician, not an AI specialist. Reviews courses for alignment, cognitive load, accessibility, and whether the stated objectives are actually assessed. Will run the chapter against the AIM-* competency map and the 1–2 hour budget in chapter-pattern.md, and will notice if a learning objective has no matching activity.

Their questions on review:

Does each objective in the opening callout map to something in Do or Check?
Is the time estimate honest? Would a median student finish in the stated window?
Can a student who missed the previous chapter follow this one?
Are the multimedia and readings accessible (captions, transcripts, no paywall) and do they survive a broken link? Does every image have alt text?
Does each visual teach something the prose does not, or is it decoration adding to the load?
Is the Check a rubric a student can self-apply, or does it need an instructor?

Fails the chapter when: objectives and activities drift apart, the scenario at the top is not returned to, or the chapter is a lecture with an exercise stapled on.

D.2.2 Dr. Okafor — the borrowing educator

Clerkship director at another institution. Wants to teach an AI session next year and has been handed this course as a starting point. Comfortable teaching, not comfortable with the technology; has used ChatGPT a handful of times and could not confidently explain what a hallucination is or why it happens. Will adopt the chapter if it makes them look competent in front of students, and will drop it if it requires expertise they do not have.

Their questions on review:

Could I run the Do block from the chapter alone, without the author in the room?
Does the chapter explain the mechanism at a level I can restate, not just the students?
What will a student ask that the chapter does not answer, and is there a pointer for that?
Which parts are institution-specific and need swapping, and are they marked?
Is anything here likely to be wrong or stale in twelve months, and does the chapter say so?
Can I reuse the images if I adopt the chapter, or do their licenses (NC, SA, fair use) stop at this course?

Fails the chapter when: it assumes the reader already knows the field, the answer key requires judgment the key does not teach, or the tools it depends on are named without a fallback.

D.2.3 Dr. Lindqvist — the clinical ethicist

Sits on the hospital ethics committee and the IRB. Reads the chapter for what it asks students to do and what it leaves unsaid, not for how much ethics content it contains; the ethics-regulation module owns the content. Their remit is narrow: is anything here harmful, undisclosed, or presented as settled when it is not.

Their questions on review:

Does the Do block ever nudge a student toward using a tool on a real patient, real notes, or a classmate’s data? The no-PHI rule is stated once in the pattern; this reader enforces it every chapter.
Where the opening scenario involves a patient, does the chapter say whether the patient knows an AI is in the room?
When the tool is wrong, who bears the cost, and is that population named or assumed?
Does the chapter present a contested question as settled, in either direction?
Does the course disclose its own use of AI (the tutor, the citation pipeline) where a student would reasonably want to know?
Is every image licensed for this use and credited, and could any illustrative artifact be mistaken for a real company’s or person’s material?

Fails the chapter when: an activity would not pass an IRB sniff test, harm is described without saying to whom, or the chapter’s certainty exceeds the evidence.

The failure mode of this reviewer is “add a paragraph on ethics.” That makes every chapter preachy and fails Marcus and Elena. Lindqvist can veto on harm, not on tone.

D.3 Using the personas

Review order: Elena first. If the avoider cannot get in the door, nothing after matters. Lindqvist last.
All three students read in interview season. The realistic reader is on a phone in an airport in November, with an hour and patchy wifi. That is what “1–2 hours” and “no installs” have to survive.
Every chapter opens with a clinical scenario; check that all three students recognize it and that the chapter returns to it at the end.
Jordan and Okafor review structure, not content. If they disagree with a student persona, the student wins on content and the faculty wins on form. Lindqvist wins on harm.