This workflow is for basic physician trainees preparing for the RACP Divisional Written Examination in Adult Medicine who already use PassGP and its AI Virtual Tutor. It addresses the written-knowledge layer only — the two multiple-choice papers, not the clinical exam. The principal limitation is structural: an AI tutor answers whatever you ask, so if you ask for the answer, it hands it over and your reasoning is never tested. The remedy is prompt design, not more questions.
What PassGP offers for RACP Adult Medicine right now
The figures below are vendor-reported from passgp.au and were last checked on 20 July 2026. PassGP's RACP product appears recently launched (a pre-launch discount was live when checked), so treat every number as provisional and confirm it on the product page before you buy.
| Item | Vendor-reported detail (20 July 2026) |
|---|---|
| Coverage | RACP Divisional Written Examination (DWE) and Divisional Clinical Examination (DCE) |
| Question volume | "3,000+" advanced-difficulty DWE-style questions |
| Question types | MCQs, extended-matching questions and clinical cases |
| Flashcards | "500+" clinical flashcards across 19 core areas |
| AI feature | "PassGP AI Virtual Tutor" for explanation and clinical queries |
| Access | Packages from 3 to 12 months; app-based access |
| Price | Not openly listed at checking; a percentage pre-launch discount was advertised; a limited free trial was offered |
Two honest observations follow. First, PassGP markets preparation for both the written and the clinical exam, but a bank plus an AI tutor is a knowledge layer — it cannot substitute for supervised long-case and short-case practice with patients and examiners. Second, an AI tutor's usefulness depends entirely on how you interrogate it.
The RACP written exam, in brief
The Divisional Written Examination is the end-of-basic-training knowledge hurdle, delivered as two single-best-answer multiple-choice papers sat on the same day. Recent sittings have been reported at roughly 100 questions in about three hours and roughly 70 in about two hours, with a small number of extended-matching items folded in — but these figures move between diets and the delivery mode has varied, so verify the current count, format and duration on racp.edu.au before planning. A separate Divisional Clinical Examination (a long case and short cases) follows. Keep the distinction sharp: PassGP's AI tutor and iatroX both sit on the written-knowledge side; neither reproduces the clinical exam, and any claim that a chatbot prepares you for a long case is marketing, not method.
Why prompt design decides what an AI tutor is worth
The default interaction with any tutor — "why is the answer C?" — teaches recognition, not reasoning: you read a fluent explanation, feel the click of understanding, and test nothing, and on unseen items that click does not transfer. The pillar on auditing an AI medical exam tutor names the failure modes — answer leakage, ungrounded assertion and shallow retention — that the prompts below are built to close.
Group your prompts by the error that produced the wrong answer, not by topic. Most written-paper errors fall into six types: a genuine knowledge gap; a missed discriminator; faulty elimination; an outdated guideline or threshold; premature closure or a probability error; and a retention or transfer failure. Diagnose the error first, then reach for the matching prompt.
Twenty-five reusable prompts, grouped by error type
Use these verbatim or adapt them. Do not paste proprietary question text into any tool; describe the clinical scenario in your own words.
Knowledge gap (mechanism prompts)
- "Explain the mechanism linking this finding to the diagnosis in three steps. Do not reveal the correct option — give me the pathophysiology only."
- "What is the single physiological reason this treatment works here? Give the causal chain, not the guideline recommendation."
- "Teach the underlying principle this item tests using a different clinical example, without naming the correct answer."
- "List the three facts I would need to answer this without guessing, then quiz me on each one before you explain."
Missed discriminator (feature prompts) 5. "Which single detail in this stem changes the diagnosis? Withhold the answer and ask me to name it first." 6. "Give me two vignettes that are identical except for the one feature separating these two conditions." 7. "Rewrite the stem with the discriminating clue removed and tell me what the answer becomes — without revealing the original answer." 8. "What is the examiner testing with this specific phrase in the stem? Ask me before you explain."
Faulty elimination (option prompts) 9. "For each of the five options, state the one finding that would make it correct, then ask me which finding is present here." 10. "Rank the distractors from most to least plausible and explain why the second-best option is wrong, without confirming the best one yet." 11. "Name the trap each wrong option is designed to catch, then let me match my error to a trap." 12. "If I had chosen my wrong option, what would I have had to believe? Show me the flawed assumption."
Outdated guideline or threshold (verification prompts) 13. "State the current Australian guideline threshold for this parameter, name the source and its date, and flag explicitly if you are unsure." 14. "Does the answer depend on a guideline that has changed in the last three years? If so, give both the old and current position." 15. "Cite the primary source for this recommendation and tell me whether it reflects Australian practice specifically." 16. "Where guidelines differ internationally on this point, which position does the RACP curriculum expect? Say plainly if you cannot verify it."
Premature closure or probability error (counterfactual prompts) 17. "Change one variable in the stem so the best answer becomes wrong. Which variable, and what is the new answer?" 18. "What pre-test probability should I have started with here, and did the stem raise or lower it?" 19. "If this patient were thirty years older or younger, would management change? Make me reason before you answer." 20. "Give the most dangerous 'nearly right' answer for this presentation and explain the exact situation in which it would be correct."
Retention or transfer failure (retrieval prompts) 21. "Generate a new vignette that tests this same concept, hide the answer, and re-present it to me in 48 hours." 22. "Turn this item into three one-line recall questions I can self-test without options." 23. "Write a harder, second-order version that applies this principle to a different organ system." 24. "Quiz me in reverse: give me the management and ask me to reconstruct the presentation." 25. "Summarise what I got wrong as a single testable rule, then set me a transfer question that uses it."
The anti-answer-leak rule
Before you open the tutor on any item, write down two things: the option you chose and one sentence of rationale. Only then may you prompt. This single discipline turns the tutor from an answer vending machine into an examiner, because every explanation now lands against a stated prediction. If you cannot articulate a rationale, that is itself the finding — the item belongs in your weak-topic queue.
The jurisdiction check
Australian internal medicine practice diverges from UK and US positions on drug availability, screening intervals and thresholds. Keep one reusable prompt for this: "Before you explain, state whether your answer reflects current Australian practice and RACP curriculum expectations, name the guideline you are relying on and its publication date, and flag any point where Australian and overseas guidance differ." An AI tutor otherwise defaults to the most common training-data position, often US-centric. Treat any unsourced threshold as unverified until checked against Therapeutic Guidelines, an Australian specialty society or a primary trial.
Create a misconception record
For every confirmed error, log four lines and nothing more: the incorrect rule you were carrying, the corrected rule, one transfer question that uses it, and a review date. A worked example: incorrect rule — "start empirical antibiotics before culture in all suspected infective endocarditis"; corrected rule — "in a stable patient with suspected subacute endocarditis, obtain blood culture sets before antibiotics unless the patient is unstable"; transfer question — "when does the stable-versus-unstable distinction change the order of investigation and treatment?"; review date — seven days on. Four lines is the ceiling; longer notes do not get reread.
Weekly verification sample
Once a week, take a small random sample of the tutor's outputs — five is enough — and check each against an official or primary source, logging any discrepancy: a wrong threshold, an out-of-date recommendation, a confident claim with no basis. This is how you calibrate how much to trust the tool, mirroring the audit-an-AI-tutor pillar. If the discrepancy rate climbs, tighten every verification prompt and lean harder on primary sources.
A seven-day plan for trainees revising around clinical work
This loop assumes a busy on-service week and no claim to any proprietary algorithm — it works with any bank and any tutor.
- Monday: One 30-minute PassGP block in your weakest domain. Commit answers and rationales before reviewing.
- Tuesday: Feed only the wrong items to the PassGP AI Virtual Tutor using the mechanism and discriminator prompts. Write four-line misconception records.
- Wednesday: Run the jurisdiction-check prompt across any guideline-dependent item from Monday. Verify two thresholds against primary sources.
- Thursday: A fresh, timed, unseen RACP block in iatroX. Because these items were not in your PassGP practice, the score measures transfer, not memory.
- Friday: Re-test Monday's misconceptions using the retrieval prompts (transfer questions, reverse quizzing).
- Weekend: Weekly verification sample; move any topic still failing after two exposures to next week's Monday slot.
Use PassGP for one job — high-volume DWE-style practice with a tutor for the misses — and iatroX for a different job: unseen measurement, so your readiness signal is not contaminated by recognition. The logic of splitting jobs across two banks is in the two-Q-bank rule.
Decision checklist: continue, supplement, switch or stop
- Continue with PassGP if your domain-level unseen scores are rising and your misconception records are clearing within two exposures.
- Supplement with a second unseen bank if your PassGP percentage is high but your fresh, timed unseen score lags it by more than ten points — that gap is recognition, not knowledge.
- Switch primary bank only if you find repeated factual errors the tutor will not correct even when challenged with a source, or if coverage of a whole blueprint area is thin.
- Stop adding new questions and move to mixed, timed papers once your unseen scores have plateaued and your remaining errors are pacing and exam-technique problems rather than knowledge gaps.
Bottom line
PassGP gives RACP trainees a large DWE-style bank and an AI tutor; whether it accelerates or lulls you depends on whether you make it test your reasoning or simply request answers. Commit before you prompt, verify what it tells you, log misconceptions in four lines, and measure yourself on unseen items elsewhere. The written papers reward disciplined reasoning under time pressure — and the clinical exam is a separate performance test that no chatbot prepares you for.
Frequently asked questions
Is PassGP enough for RACP Adult Medicine on its own? For the written papers it can be a strong core if the vendor-reported volume holds up, but "enough" is the wrong test: what matters is whether your fresh, timed, unseen score is at or above the pass standard, not whether you have finished a bank. A single bank also risks turning completion into recognition, so most candidates benefit from a second, unseen source — and on its own it prepares you for only one of the two exams, the DWE, not the clinical component.
Which RACP Adult Medicine component does PassGP not reproduce well? The Divisional Clinical Examination. A written bank and an AI tutor cannot assess bedside data-gathering, examination technique, or the structured reasoning of a long case under examiner observation. PassGP markets clinical-exam preparation, but that stage needs supervised practice with patients and examiners, not a chatbot. Confirm the current clinical-exam structure on racp.edu.au.
How should I verify PassGP AI answers for RACP Adult Medicine? Run the jurisdiction-check prompt on every guideline-dependent item, demand a named source and date, and take a weekly random sample of five outputs to check against Therapeutic Guidelines, an Australian specialty society or a primary trial. Log discrepancies, and treat any unsourced threshold as unverified until confirmed independently.
When should I stop using PassGP and move to mixed mocks? When your unseen, timed scores have plateaued and your errors are about pacing, misreading stems and stamina rather than missing knowledge. At that point full-length mixed papers under exam timing give you more than another topic block, because the remaining risk is execution across a three-hour paper, not content.
How should I combine PassGP with iatroX without duplicating practice? Assign each a distinct job. PassGP is your teaching-and-drilling bank with a tutor for misses; iatroX is your unseen measurement bank, run only on items you have not seen, so its score reports transfer rather than recall. Never review the same item in both. To interrogate the reasoning behind any single number, read why your Q-bank percentage is not your exam score.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 20 July 2026. Product figures are vendor-reported from passgp.au on that date and, given the product appears recently launched, may change — confirm the live question count, access period, AI features and price on the vendor's page before relying on them. Exam-format details should be confirmed on the official RACP pages, as question counts, delivery mode and durations vary between diets. Disclosure: iatroX operates a competing question bank and clinical-knowledge platform; its role here is confined to unseen written-knowledge measurement, a job PassGP's clinical-exam claims do not cover, and it is not a clinical-exam simulator. Corrections are welcome via the feedback route on iatrox.com.
References: Royal Australasian College of Physicians — Divisional Written Examination and Adult Internal Medicine basic training (racp.edu.au); PassGP RACP product page (passgp.au/racp); iatroX internal resources — the two-Q-bank rule, auditing an AI medical exam tutor and the comparison hub.
Open a missed RACP item and run the verification prompt in iatroX →