QBankMD is a Canadian MCCQE Part I question bank that pairs a conventional bank with an AI Assistant chatbot and an adaptive engine. It suits candidates — particularly international medical graduates — who want on-demand explanation and translation while they practise. Its principal limitation is the one every generative tutor shares: the chatbot layer can produce fluent, confident text whose provenance you cannot see, so the feature that feels most helpful is also the one you must interrogate hardest. This review scores the tutor on grounding, reasoning, calibration and exam fidelity, and gives you a protocol to verify it.
We did not have privileged access to QBankMD's internal models, and this is not a claim to have read its source code. It is an audit method: a fixed set of probe types, a published rubric, and the failure modes to watch, so you can reproduce the assessment on the live product and reach your own verdict.
What QBankMD offers for MCCQE Part I right now
| Attribute | Vendor-reported state (last checked 19 July 2026) |
|---|---|
| Question volume | Approximately 3,927 practice questions (product page); the vendor blog cites "3,000+" |
| AI Assistant | An "intelligent chatbot" for asking questions about exam content, requesting clarifications, explanations and translations |
| Adaptive engine | An adaptive learning algorithm on all plans, said to focus on weak areas, with personalised study recommendations |
| Explanations | Physician-written, in-depth for correct and incorrect options, described as evidence-based with references to medical literature and a Canadian-guideline focus |
| Access periods | 1, 2, 3, 6, 9 or 12 months |
| Price (CAD, vendor-reported) | From $135 (1 month) to $480 for 12 months (excl. tax), roughly $40/month equivalent; free 20-question trial |
| Components supported | Written multiple-choice practice only |
Every figure above is vendor-reported and should be re-checked on the product page. Note the useful distinction the table draws: the static explanation base is described as physician-written and referenced, which is a good provenance story, whereas the AI Assistant is a generative chatbot, whose per-answer grounding is exactly what an audit has to test. Those are two different layers with two different trust profiles.
The exam you are actually preparing for
Since April 2025 the MCCQE Part I has been multiple-choice only, the earlier clinical decision-making component having been removed. The MCC currently describes 230 MCQs in two sections of 115 items, with up to two hours and forty minutes per section; confirm the live count and duration on the MCC site. The exam is built on the MCC Examination Objectives under the CanMEDS roles and sampled against a two-axis blueprint: Dimensions of Care (Health Promotion & Illness Prevention, Acute, Chronic and Psychosocial, each targeted near 25%) and Physician Activities (Assessment/Diagnosis about 20%, Management about 35%, Communication about 25%, Professional Behaviours about 20%). A tutor that helps you on diagnosis but neglects management, communication and professional reasoning is helping you with a minority of the blueprint.
How we audited the tutor: four tests and six probes
The rubric has four scored dimensions. Grounding: can you see where each claim comes from? Reasoning: does the tutor build understanding or just reveal the answer? Calibration: does its confidence track the actual strength of the evidence? Fidelity: does its advice respect MCCQE Part I jurisdiction, terminology, timing and blueprint weighting? Score each from 0 to 3 and re-run periodically, because model behaviour changes silently between versions.
Run the same six probes every time, drawn from your own practice so no copyrighted stem is reproduced: a straight-recall item, a diagnosis item, a next-investigation item, a management item, an ethics or medico-legal item, and one deliberately ambiguous item where the honest answer is "it depends." The recall and diagnosis probes test baseline accuracy; the investigation and management probes test whether reasoning respects Canadian sequencing; the ethics probe tests jurisdiction; and the ambiguous probe is the most revealing, because a well-calibrated tutor should refuse to feign certainty.
Grounding: where do the words come from?
Grounding is the audit's centre of gravity. QBankMD's static explanations are described as physician-written and referenced, which, if accurate, gives that layer a defensible provenance. The generative chatbot is a different matter. When you ask the AI Assistant to elaborate, you need to know whether it is quoting those physician-written explanations, paraphrasing an external guideline, or generating uncited model output that merely sounds authoritative.
Test it directly. Ask the Assistant, "What is the source for that recommendation?" and "Is that from a Canadian guideline, and which one?" A well-grounded tutor names a source you can check or openly says it cannot attribute the claim; a poorly grounded one produces a plausible citation that dissolves when you look for it. Do not accept a reference you have not seen; a named guideline is a lead to verify, not proof. The aim is not to reproduce copyrighted content but to confirm that a real, checkable source stands behind the advice.
Reasoning behaviour: does it teach or does it tell?
A tutor earns its name by improving your reasoning, not by shortening the path to the letter. Probe four behaviours. Does it ask a useful diagnostic question before answering, or does it immediately reveal the option? Does it prematurely leak the answer when you wanted to reason first? Does it handle uncertainty honestly on the ambiguous probe? And, most diagnostic of all, does it correct a false premise?
That last test is the sharpest. Assert something wrong — "since the first-line treatment here is X, why isn't the answer Y?" when X is not first-line — and watch what happens. A tutor with good reasoning discipline stops and corrects the premise; a sycophantic one accepts your error and builds a confident, wrong explanation on top of it. Generative chatbots are prone to this agreement bias, so a single false-premise probe tells you more about a tutor's real value than a dozen questions it answers correctly.
Exam fidelity: Canadian, current, time-aware
Fidelity asks whether the tutor's advice would survive contact with the actual MCCQE Part I. Check jurisdiction first: on your ethics and guideline probes, does it default to Canadian framing — consent and capacity as practised in Canada, Canadian screening intervals, Canadian immunisation schedules — or does it drift to United States conventions, a common failure for models trained on USMLE-heavy corpora? QBankMD advertises a Canadian-guideline focus for its explanations; verify that the chatbot honours it too.
Check terminology, and check timing. The tutor should reinforce that this is a next-best-step examination weighted heavily toward Management, sat at roughly one minute per item across two 115-item sections, and it should not encourage a depth of deliberation you cannot afford on the day. A tutor that teaches beautiful, exhaustive workups is not preparing you for a paced MCQ paper.
Failure modes to watch
Five failure modes recur with generative tutors, and you should log each when it appears. Hallucinated citations — confident references to guidelines or trials that do not say what is claimed. Overconfident wording on genuinely uncertain questions, where hedging is the correct answer. Outdated guidance, because a model's training cut-off may predate a guideline change. Answer leakage, where the tutor reveals the option before you have reasoned, quietly converting practice into passive reading. And plausible-but-unexamined elaboration — fluent detail that pads the explanation without being tied to any source. None of these means the product is bad; all of them mean the chatbot output is a draft to verify, not a verdict to trust.
A safe-use protocol: answer first, interrogate second, verify third
Adopt a fixed three-step routine so the tutor accelerates learning without eroding calibration.
First, answer first. Commit to your own choice before you open the Assistant, so the tutor cannot short-circuit the retrieval that makes practice work. Second, interrogate second. Ask the tutor to explain the reasoning, then challenge it: "What is the source?", "Would this differ under current Canadian guidance?", and at least once per session a deliberate false premise to test whether it corrects you. Third, verify third. For any management or guideline claim that would change your answer, confirm it against a Canadian primary source before you internalise it. High-value prompts share a shape: they ask for provenance, for the Canadian-specific position, and for the one situation in which the stated answer would be wrong.
A seven-day plan for international graduates
Give QBankMD one job — adaptive, explained practice with on-demand clarification — and iatroX a different one: unseen, timed measurement and Socratic review of missed items. No proprietary-algorithm claims are made for either; the point is to learn on one surface and measure on another.
| Day | QBankMD job (learn) | iatroX job (measure) |
|---|---|---|
| 1 | 40 adaptive items; use the Assistant only after committing an answer | — |
| 2 | 40 items; run the six-probe rubric on five explanations | 20-item unseen mixed block, timed |
| 3 | 40 items weighted to Management-type stems | — |
| 4 | 30 items; log any hallucinated citation or false-premise failure | 20-item unseen mixed block; Socratic Tutor on every miss |
| 5 | Redo flagged items; verify three guideline claims against Canadian sources | — |
| 6 | 30 items on your two weakest blueprint domains | 30-item unseen block; compare with Day 2 |
| 7 | Light review; no new volume | Short unseen block; read the trend |
Decision checklist: continue, supplement, switch or stop
Continue if the tutor grounds its claims, corrects false premises, respects Canadian framing, and your unseen iatroX blocks are trending up. Supplement — with primary-source reading or a second, non-overlapping bank — if the explanations are strong but the chatbot's provenance is thin, or if a blueprint domain stays weak. Switch primary bank only for a measurable, repeated fidelity failure, such as systematic United States framing on Canadian ethics items, not because the AI feels impressive. Stop leaning on the Assistant when you notice it answering before you reason; at that point it is feeding answer leakage and eroding the calibration you came to build. Decide on measured behaviour, never on novelty or sunk cost.
Bottom line
QBankMD's combination of physician-written explanations, an adaptive engine and a generative Assistant is a reasonable MCCQE Part I package, and the AI layer can genuinely speed up clarification and translation for international graduates. But the chatbot's value is conditional on its grounding, and grounding is invisible unless you test it. Run the six probes, insist on sources, plant a false premise, and verify anything that would change your answer. Used with that discipline, the tutor is an accelerant; used credulously, it is a confident narrator of unverified claims. Your bank percentage, AI-assisted or not, is still not your exam score — measure readiness on unseen material.
FAQ
Is QBankMD enough for MCCQE Part I on its own? For most candidates a single bank is not sufficient on its own, and QBankMD is best treated as a strong practice-and-explanation resource rather than complete preparation. Its adaptive engine and AI Assistant support learning, but neither guarantees blueprint coverage nor replaces objectives-led reading and unseen mock practice. Its usefulness also depends on how well you verify the chatbot's output rather than accepting it at face value.
Which MCCQE Part I component does QBankMD not reproduce well? There is no separate CDM component to reproduce, since the exam is now multiple-choice only, so the harder gap is qualitative: a generative tutor reproduces confident explanation more readily than reliable provenance, and it can under-serve the ethical, medico-legal and population-health reasoning weighted heavily in the Physician Activities axis. Treat the chatbot's fluent output on those topics with particular caution and verify against Canadian sources.
How should I verify QBankMD AI answers for MCCQE Part I? Use the answer-first, interrogate-second, verify-third routine: commit your own answer, then ask the Assistant for its reasoning and its source, then confirm any decision-changing claim against a current Canadian primary source such as Hypertension Canada, Diabetes Canada, the Canadian Paediatric Society or Choosing Wisely Canada. Add a deliberate false-premise prompt at least once per session; a tutor that accepts your error and elaborates on it has failed the most important test.
When should I stop using QBankMD and move to mixed mocks? Move to unseen, mixed, timed mocks in the final two to three weeks, or sooner if you notice the Assistant answering before you reason and your accuracy on familiar items plateauing. Adaptive practice on a familiar bank increasingly rewards recognition, and the only trustworthy readiness signal is performance on material you have not seen, sat under the real pacing of two 115-item sections.
How should I combine QBankMD with iatroX without duplicating practice? Assign each tool a distinct job: learn and clarify on QBankMD, then measure transfer on iatroX with unseen, timed blocks and the Socratic Tutor on missed items. Do not re-answer identical items across both platforms, because the measurement value depends on the material being unfamiliar. In effect QBankMD builds and explains, while iatroX provides the independent read-out that tells you whether the explanation actually stuck. See the AI-tutor audit and calibration guides below.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026. Vendor-reported figures — question counts, AI and adaptive features, and prices — were taken from the QBankMD product page and blog on that date and can change without notice; confirm the current details before purchase, and treat AI-generated explanations as drafts to verify rather than settled fact. Disclosure: iatroX operates a competing MCCQE Part I question bank with its own Socratic Tutor, so this audit confines iatroX's role to a job QBankMD's chatbot does not claim — unseen, timed measurement and source-first review of missed items. Corrections are welcome through the feedback route on iatrox.com.
References: Medical Council of Canada, MCCQE Part I overview, multiple-choice format and blueprint (mcc.ca); QBankMD product page and blog (qbank.md). Internal: How to audit an AI medical-exam tutor: grounding, answer leakage, hallucinations and retention; Calibrating automated feedback before you trust the score; Your Q-Bank Percentage Is Not Your Exam Score; iatroX MCCQE Part I bank; compare Q-banks.
Open a missed MCCQE Part I item in the iatroX Socratic Tutor →