This audit is for RACP basic trainees preparing the Divisional Written Examination in Adult Internal Medicine who are considering PassGP and, in particular, its AI Virtual Tutor. Used with discipline, an AI tutor can turn a wrong answer into a fast, structured explanation. The principal limitation is that a generative tutor produces fluent text that is not automatically correct or current, so it belongs in a workflow where you answer first, interrogate second and verify against the RACP curriculum third. iatroX covers the written-knowledge and unseen-MCQ layer, not the clinical exam.
What PassGP offers for RACP Adult Medicine right now
PassGP does cover RACP — this is not a coverage-gap audit. The platform runs a dedicated RACP pathway spanning both the Divisional Written Examination and the Divisional Clinical Examination, and it advertises an AI Virtual Tutor. The figures below are vendor-reported and were last checked on 19 July 2026; verify the live details on passgp.au before you buy.
| Feature | Vendor-reported (last checked 19 July 2026) |
|---|---|
| RACP coverage | Dedicated RACP pathway — written and clinical, described as end-to-end |
| Question volume (RACP) | 3,000+ advanced-difficulty "DWE-style" MCQ and EMQ items, plus full mock exams |
| Clinical resources | Short- and long-case preparation; 500+ clinical flashcards across 19 core areas |
| AI features | "AI Virtual Tutor for adaptive learning" — to teach, guide and answer clinical queries |
| Live teaching | Examiner-led tutorials advertised in some packages |
| Price and access | Not fully published; 6- and 12-month programmes; a pre-launch discount and a free trial (15 EMQ, 15 MCQ, 10 clinical cases) were advertised |
Two caveats are worth stating. First, the RACP bank is described as "DWE-style" and is advertised for RACP basic training generally; confirm it is tuned to the Adult Internal Medicine stream you are sitting, rather than Paediatrics & Child Health. Second, the public pages describe the AI Virtual Tutor only in broad terms — "adaptive learning", "answer your clinical queries" — without published detail on grounding, sourcing or how explanations are generated. That thin disclosure is exactly why the verification protocol below matters.
The RACP exam you are actually preparing for
The RACP Divisional Written Examination sits at the end of basic training and tests knowledge in Adult Internal Medicine or Paediatrics & Child Health. As verified against the RACP examination pages, the current Adult Medicine written exam is delivered as two papers: a Clinical Applications paper of around 100 questions (predominantly single-best-answer, with a small number of extended-matching items) over roughly three hours and ten minutes, and a Medical Sciences paper of around 70 questions over roughly two hours and ten minutes. It is delivered on paper with computer-readable answer sheets. Treat these numbers as body-reported and last checked on 19 July 2026, and verify the current question count, format and duration on racp.edu.au, because they are set per diet. A separate Divisional Clinical Examination — long case and short cases — follows the written exam and assesses clinical skills directly.
The line to hold is simple: the RACP curriculum and the College's own examination guides are the authoritative source. An AI tutor is a study aid built on top of that curriculum, and any answer it gives is a claim to be checked against RACP and current Australian guidance, not a substitute for them.
Testing methodology: a fixed rubric across six question types
To audit a tutor fairly, run the same representative set through it and score against a published rubric rather than an impression. Use six item types: a straight-recall fact, a diagnosis vignette, a next-investigation question, a management decision, an ethics or professionalism item, and a deliberately ambiguous stem with no clean single answer. For each, score the tutor on five axes: correctness against RACP-level guidance, whether it shows reasoning or only asserts a conclusion, whether it cites a checkable source, how it handles uncertainty, and whether it leaks the answer before you have thought. Publishing the rubric lets any reader repeat the test on their own subscription instead of taking a review on trust.
| Item type | What a good tutor should do |
|---|---|
| Straight recall | Give the fact and a one-line why, without padding |
| Diagnosis | Reason from the discriminating features, not pattern-match to the buzzword |
| Next investigation | Justify sequence and yield, not just name a test |
| Management | Weigh options and state the decision threshold |
| Ethics / professionalism | Reason to the defensible action; avoid false certainty |
| Ambiguous stem | Name the ambiguity rather than manufacture a confident single answer |
Grounding audit: where do the explanations come from?
The single most important question for any AI tutor is provenance. Does an explanation cite the RACP curriculum, a named external guideline, or nothing checkable at all? PassGP's public description does not commit to a grounding model, so inspect it yourself: ask the tutor to name its source and link or quote the specific guideline for a claim. If it produces a real, checkable reference, provenance is good. If it produces a plausible-sounding citation you cannot find, or none, treat the explanation as uncited model output and verify independently. You do not need to reproduce copyrighted content to do this — you only need to confirm that a named, current source actually says what the tutor claims.
Reasoning behaviour: does it teach or just tell?
A tutor earns its place if it improves your reasoning rather than shortcutting it. Test four behaviours. Does it ask a useful diagnostic question — "what would change your management here?" — before revealing the answer, or does it dump the conclusion immediately? Does it handle uncertainty honestly, or force a single answer onto an ambiguous stem? Does it correct a false premise if you feed it one — for example, an incorrect drug mechanism embedded in your question — or does it agree and build on your error? Answer leakage is the quiet failure: a tutor that shows the answer the moment you open the explanation trains recognition, not recall, and recognition is not what the written exam rewards.
Exam fidelity: jurisdiction, terminology, timing and weighting
RACP is an Australasian exam, and a general-purpose model often defaults to United States or United Kingdom framing. Test whether the tutor respects Australian and New Zealand terminology, local guideline sources and the RACP blueprint weighting, or whether it quietly imports overseas thresholds and brand names. Test whether it respects the exam's pace — the written papers reward efficient reasoning, not exhaustive essays — and whether its explanations map to the blueprint's emphasis rather than to whatever the model finds most interesting. A tutor that is clinically reasonable but jurisdictionally off can teach you the wrong local standard.
Failure modes to expect
Five failures recur with AI tutors, and naming them makes them easier to catch. Hallucinated citations — confident references to guidelines or trials that do not say what is claimed. Overconfident wording — "always", "never", "the definitive investigation" — where real medicine is conditional. Outdated guidance — a model reflecting a superseded recommendation. Answer leakage — revealing the key before you reason. And plausible-but-unexamined elaboration — fluent extra detail that sounds authoritative but drifts beyond what the item, or the exam, actually tests. None means the tool is worthless; all mean it needs a verification step.
Safe-use protocol: answer first, interrogate second, verify third
Use a three-step routine on every item. Answer first: commit to your answer and your reasoning before you open the tutor, so it cannot leak the key into a decision you have not yet made. Interrogate second: ask the tutor to justify the key and to say why each distractor is wrong, then push back — "what if the patient were older?", "what is your source for that threshold?" High-value prompts include "cite the specific guideline and its year", "give the discriminating feature that rules the alternative out", and "where is this recommendation different in Australia?". Verify third: for anything guidance-sensitive or surprising, confirm against a current RACP-appropriate source before you commit it to memory. This routine keeps the tutor as an accelerant, not an authority.
A worked seven-day plan: PassGP for one job, iatroX for measurement
This loop assumes a busy trainee revising around clinical work. It uses PassGP for one defined job — teaching and explanation on missed items — and iatroX for the different job of measuring transfer on unseen, timed single-best-answer questions. No proprietary-algorithm claims are made.
- Day 1 — 30 PassGP items in your weakest system; for each miss, run the answer-first, interrogate, verify routine. That evening, 30 unseen timed items in iatroX and log misses by system.
- Day 2 — Review only yesterday's verified explanations; 30 unseen iatroX items weighted to those systems.
- Day 3 — 30 PassGP items in a second weak system; interrogate the tutor on two ambiguous stems specifically. 30 unseen iatroX items.
- Day 4 — Data and image interpretation practice; verify any threshold the tutor gives against a current source. 30 unseen iatroX items.
- Day 5 — Ethics and management items; test whether the tutor forces false certainty. 30 unseen iatroX items.
- Day 6 — One timed mixed block at paper length; map misses to the blueprint.
- Day 7 — Rest or light review; decide the next action from the trend, not from a single block.
Remember that your iatroX percentage is not your exam score; it is a trend line and a coverage map. Use the Socratic Tutor in iatroX to interrogate a missed item before you read the explanation, applying the same verify-first discipline.
Decision checklist: continue, supplement, switch or stop
- Continue with PassGP if its explanations survive your verification routine and your unseen scores are trending up.
- Supplement with a second, independent unseen bank so your measurement is not taken on the same items you learned from — the two-bank principle.
- Switch emphasis toward full timed mocks and, for the clinical exam, supervised long- and short-case practice once your written knowledge is reliable.
- Stop relying on any tutor explanation that repeatedly fails verification; a fluent wrong answer is worse than no answer.
Decide on measurable gaps — verification hit-rate, unseen-score trend, blueprint coverage — not on novelty or sunk cost.
Frequently asked questions
Is PassGP enough for RACP Adult Medicine on its own? It is a substantial, RACP-specific resource with a large question bank and an AI tutor, and for many trainees it can carry most of the written-exam preparation. On its own, though, it shares the weakness of any single bank: you end up measuring your readiness on the same items you learned from, and the clinical exam needs supervised long- and short-case practice that no bank reproduces. Treat it as a strong core with an independent measurement layer added, not as the whole plan.
Which RACP Adult Medicine component does PassGP not reproduce well? The Divisional Clinical Examination. PassGP offers case-preparation material, but the long case and short cases are examiner-observed, real-patient encounters, and no platform reproduces the interpersonal and physical-examination reality of that assessment. Its AI Virtual Tutor is a written-knowledge aid; it is not a clinical examiner, and its published functional detail is thin, which is why every explanation needs verifying.
How should I verify PassGP AI answers for RACP Adult Medicine? Use answer first, interrogate second, verify third. Commit to your own answer before opening the tutor, then ask it to cite the specific guideline and year and to explain why each distractor is wrong, and finally confirm anything guidance-sensitive against a current RACP-appropriate source. Pay particular attention to jurisdiction — check that thresholds and drug choices reflect Australian and New Zealand practice rather than imported United States or United Kingdom defaults.
When should I stop using PassGP and move to mixed mocks? When your first-attempt accuracy on unseen items has plateaued and your remaining errors are pace and discrimination rather than knowledge, shift the balance to full timed mixed blocks at paper length. That is usually the final few weeks. Keep the tutor only for targeted repair of a specific recurring error, not as a daily default, because at that stage volume of unseen practice matters more than more explanation.
How should I combine PassGP with iatroX without duplicating practice? Give each a distinct job: PassGP for teaching and explanation on missed items, and iatroX for unseen, timed measurement of whether that teaching transferred. Because you should never measure readiness on the same items you learned from, keeping the learning bank and the measurement bank separate is the point, not an inconvenience — compare the error type across the two rather than repeating the same questions.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026. Vendor figures are vendor-reported and change between releases; verify the current PassGP question counts, AI Virtual Tutor detail, price and access period on passgp.au, and confirm the RACP written-exam format and question counts on racp.edu.au before relying on any number here. Disclosure: iatroX operates a competing question bank covering the RACP written-knowledge layer; it is not a clinical-exam simulator, and this audit confines iatroX's role to the unseen-measurement job that PassGP's tutor does not claim. Corrections are welcome via the feedback route on iatrox.com.
References: RACP — Divisional Written Examination and examinations pages (racp.edu.au); PassGP RACP pathway (passgp.au/racp); iatroX Australia question bank (/australia); "Your Q-Bank Percentage Is Not Your Exam Score" (iatrox.com/blog/qbank-percentage-not-your-exam-score); how to audit an AI medical exam tutor (iatrox.com/blog/how-to-audit-an-ai-medical-exam-tutor-grounding-answer-leakage-hallucinations-and-retention); iatroX comparison hub (/compare).