This audit is for RACGP registrars deciding whether to rely on the PassGP AI tutor for the AKT and KFP — and, by extension, the CCE. It addresses the written knowledge components most directly, and its principal limitation is stated up front: an AI tutor can accelerate your understanding of content, but it cannot certify its own accuracy, so the value depends entirely on whether you audit it. This article gives you the protocol to do that, rather than a verdict you should take on trust.
We have not been granted evaluation access to the current PassGP tutor, so we do not report its specific outputs; we describe the audit you should run yourself and label every vendor claim as vendor-reported.
What PassGP offers for RACGP Fellowship right now
The following is vendor-reported and was last checked on 19 July 2026; confirm the current inclusions and pricing on passgp.au, as course contents and prices change between diets.
| Item | Vendor-reported position (19 July 2026) |
|---|---|
| AKT bank | 2,595 "advanced-difficulty" AKT questions |
| KFP bank | 1,395+ KFP questions |
| CCE | 500+ CCE cases with model answers |
| AI tutor | "PassGP AI Virtual Tutor," described as trained for AKT, KFP and CCE |
| Mocks & feedback | 14 mock exams; mocks with examiner feedback; a free 1:1 mock in the full package |
| Other assets | 500+ flashcards across 19 clinical areas; short video "masterclass" reels; unlimited small-group teaching; a podcast |
| Components | AKT, KFP and CCE all covered |
| Access & price | Free trial; "Crash Course 2026.2" reported at $595; a 12-month Fellowship package covering all three exams |
That is a broad offering, and the small-group teaching and human-marked mocks are genuine strengths that pure banks lack. The AI tutor, however, is the component that needs auditing, because it is the one that can be confidently wrong.
The exam you are actually preparing for
The RACGP Fellowship written examinations are the Applied Knowledge Test (AKT) — 150 single-best-answer questions — and the Key Feature Problem (KFP) — 70 individual multiple-selection questions, each case independent. Both are currently delivered on paper with computer-readable answer sheets; the RACGP sets the time allowed, and because published durations have varied you should confirm the current timing on racgp.org.au rather than trusting a third-party figure. The Clinical Competency Exam (CCE) is the clinical/consultation component, now reported in score bands rather than a single mark; its detailed structure is set by the college, and no third party should be inventing it.
Hold the official material as your authority. The RACGP curriculum, the examination report and the college's sample questions define the standard; the PassGP tutor is a tool to be measured against them, not a substitute for them.
Testing methodology: a fixed probe set and a published rubric
To audit any AI tutor fairly, put the same fixed set of item types through it and mark each against the same rubric. Use at least these six probes:
- Straight recall — a fact with a single correct answer (for example, a first-line drug).
- Diagnosis — a stem that requires selecting the most likely diagnosis from a realistic differential.
- Next investigation — an item that turns on the single most appropriate next test.
- Management — a question asking for the best next management step in Australian general practice.
- Ethics or professionalism — consent, confidentiality, mandatory reporting or a boundary issue.
- A deliberately ambiguous item — one with no clean answer, to see how the tutor handles uncertainty.
Score each response against the same rubric: is it correct; is it grounded in a citable source; does it reason before it concludes; does it flag uncertainty; and does it resist a false premise you plant in the question? Publishing your rubric — even just to yourself — is what turns an impression into an audit.
Grounding audit
Grounding is the heart of it. For each probe, ask where the answer comes from. Does the tutor cite an identifiable authority — the RACGP curriculum, Therapeutic Guidelines, an Australian clinical guideline — or does it offer fluent, uncited prose that reads authoritatively but points to nothing? You can inspect provenance without reproducing copyrighted content: ask the tutor to name its source and the date, then check that source yourself. If it cannot name one, or names one that does not say what it claims, treat the answer as ungrounded. Our pillar on how to audit an AI medical exam tutor sets out the full grounding, answer-leakage and retention checks; apply them here directly.
Reasoning behaviour
A good tutor reasons; a weak one asserts. Assess four behaviours. Does it ask useful diagnostic questions before committing, or does it prematurely reveal the answer the moment you open the item, destroying its value as practice? Does it handle uncertainty by giving a differential and a rationale, or does it force a single answer onto an ambiguous stem? And — the most revealing test — if you deliberately state a false premise ("given that this patient has confirmed X"), does it accept and build on the error, or does it correct you? An AI that cheerfully elaborates on a false premise is one you cannot trust unsupervised.
Exam fidelity
Content accuracy is not the same as exam fidelity. Test whether the tutor's advice respects the specifics of the RACGP Fellowship: Australian jurisdiction and terminology, the primary-care framing of the questions, the time constraints of the real papers, and the blueprint weighting across the curriculum. A tutor that gives excellent hospital-medicine answers can still be a poor fit for a general-practice exam if it consistently reaches for secondary-care thresholds and referral patterns that do not match Australian primary care.
Failure modes to watch for
- Hallucinated citations — plausible-looking references to guidelines that do not say what is claimed, or do not exist.
- Overconfident wording — a uniform tone of certainty that does not distinguish a settled fact from a contested one.
- Outdated guidance — advice that reflects a superseded recommendation, with no date attached.
- Answer leakage — revealing the answer before you have had to reason, which quietly converts practice into passive reading.
- Plausible-but-unexamined elaboration — extra detail that sounds authoritative, adds risk, and was never checked.
A safe-use protocol: answer first, interrogate second, verify third
Use the tutor inside a fixed three-step routine. Answer first: commit to your own answer and reasoning before you read anything the tutor says, so it cannot lead you. Interrogate second: ask the tutor to justify its answer, name its source and date, and explain why the distractors are wrong — high-value prompts include "which Australian guideline supports this, and from what year?" and "what would change your answer?" Verify third: for anything guidance-sensitive or surprising, check the named source yourself before you commit it to memory. The routine costs a minute per item and removes almost all of the failure modes above.
A seven-day pattern for Australian GP registrars
This pattern gives PassGP one job — teaching and explaining content through its bank, tutor and small groups — and iatroX a different job: measuring whether that content has transferred, on unseen AKT-style single-best-answer items you did not write. Be clear on scope. iatroX covers RACGP AKT-style knowledge and is explicitly not a CCE consultation simulator; it will not rehearse your consulting. Treat any adaptive sequencing it offers as vendor-described and judge it only by whether your unseen scores move; make no assumptions about the algorithm.
- Day 1: Work a themed PassGP AKT set; run every missed item through the tutor using the answer-first routine.
- Day 2: KFP practice in PassGP, focusing on precise selection; then 20 unseen AKT-style items in iatroX on the same theme to test transfer.
- Day 3: Attend a small-group session or watch a masterclass reel; log two management principles to test later.
- Day 4: A timed AKT block in PassGP; then a short unseen iatroX block to check the same knowledge holds cold.
- Day 5: CCE case work in PassGP with the model answers; book or rehearse a real consultation with a supervisor, because this is the part software cannot mark.
- Day 6: A mixed AKT/KFP mock under time; verify any surprising tutor claims against Australian guidelines.
- Day 7: Review the week's verified corrections; sit one unseen iatroX AKT-style block to confirm retention, and set next week's themes from the errors.
Decision checklist: continue, supplement, switch or stop
- Continue with PassGP if its tutor survives your audit — grounded, reasons before concluding, corrects false premises — and its human-marked mocks and small groups are moving your scores.
- Supplement with iatroX for unseen AKT-style measurement, and with supervised consultation practice for the CCE, if your knowledge is strong but your transfer or your consulting is untested.
- Switch emphasis toward examiner-marked mocks and CCE role-play as the exam nears, because the AI tutor cannot certify your consultation performance.
- Stop buying more content when the limiting factor is consolidation, not coverage — a stable unseen score, not an unfinished bank, is the signal.
Frequently asked questions
Is PassGP enough for RACGP Fellowship on its own? For the written AKT and KFP it is a broad and well-resourced option, and its human-marked mocks and small-group teaching add something banks alone cannot; but "enough on its own" is an overstatement for any single product, particularly for the CCE. The CCE is a live consultation assessment, and no bank or AI tutor can certify how you actually consult, so most registrars should pair PassGP's content with supervised consultation practice. Treat it as a strong core, not a complete solution.
Which RACGP Fellowship component does PassGP not reproduce well? The CCE consultation performance. PassGP can teach CCE content and structure through its cases and model answers, and that is useful, but the exam assesses a live, examiner-judged interaction — communication, timing, safety and rapport in real time — which an AI tutor and a case bank cannot replicate. Use supervised role-play and real consultations, ideally observed, to cover the part the platform describes but cannot mark.
How should I verify PassGP AI answers for RACGP Fellowship? Use the answer-first, interrogate-second, verify-third routine: commit to your own answer before reading the tutor, then ask it to name its source and date and justify why the distractors are wrong, and finally check that source yourself for anything guidance-sensitive. If the tutor cannot name a current Australian source, or names one that does not support the claim, do not commit the answer to memory — an unverified AI explanation is a hypothesis, not a fact, however fluently it is written.
When should I stop using PassGP and move to mixed mocks? Move to full mixed mocks once your themed practice is consistently strong on unseen items and the tutor is no longer surfacing genuine knowledge gaps. At that stage the limiting factors are pacing, stamina and cross-topic switching across a whole paper, which single-theme practice does not train — usually the final few weeks. Keep some human-marked mocks in that window so your judgement, not only your recall, is tested.
How should I combine PassGP with iatroX without duplicating practice? Assign non-overlapping jobs and never run the same items twice. PassGP teaches and explains — bank, tutor, small groups, CCE cases — while iatroX measures transfer on unseen AKT-style questions and keeps the underlying knowledge retrievable; because iatroX is a knowledge and unseen-measurement layer and not a CCE simulator, the two do not collide. The two-Q-bank rule is the method for keeping the second source clean rather than duplicative.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026; PassGP question counts, AI-tutor descriptions, inclusions and prices are vendor-reported and change between diets — verify the current figures on passgp.au before relying on them, and note that we describe an audit protocol rather than reporting the tutor's specific outputs, to which we do not have evaluation access. Disclosure: iatroX operates a competing question bank and its own Socratic tutor; for the RACGP Fellowship its role here is confined to unseen AKT-style measurement and knowledge retention — jobs PassGP's CCE consultation product does not claim to certify — and iatroX is not a CCE simulator. Corrections are welcome via the feedback route on iatrox.com.
References: RACGP AKT, KFP and CCE examination pages and curriculum (racgp.org.au); the PassGP RACGP product pages (passgp.au) for vendor-reported features and pricing; iatroX internal references — how to audit an AI medical exam tutor, Your Q-Bank Percentage Is Not Your Exam Score and the RACGP Fellowship bank landing page.