blog

Answer-first chatbots optimise this minute; Socratic tutors optimise next month; the evidence on productive struggle says the difference shows up at the exam.

The randomised evidence on AI tutors is thinner and more mixed than the marketing suggests: no clear overall knowledge advantage, better practical skills, and low certainty through...

Postgraduate exam preparation is a time problem before it is a knowledge problem; the best AI tools for working doctors are built for thirty-minute days.

The AMC CAT tests Australian practice, not generic English-language medicine; the best AI preparation in 2026 keeps jurisdiction front of mind.

The MCCQE Part I is not a reskinned USMLE; the best AI preparation respects Canadian context, from CanMEDS framing to preventive care and ethics.

UWorld remains the USMLE bank of record; the 2026 question is which AI layers around it, and where a low-cost additional adaptive bank fits.

MRCP Part 1 rewards mechanism-deep question practice, and the SCEs reward specialty-specific depth; one subscription covering all 13 SCE specialties changes the maths.

The AKT and SCA reward different machines: guideline-grounded question practice for one, consultation rehearsal for the other, with daily general practice feeding both.

PLAB 1 rewards UK-guideline question practice and PLAB 2 rewards consultation rehearsal; the best AI stack for IMGs in 2026 trains each with the right machine.

The UKMLA is two different exams wearing one name; the best AI preparation in 2026 pairs a blueprint-mapped AKT bank with virtual-patient CPSA practice.

AI can deepen placement learning or quietly replace it; here is the line between educational use and unsupported clinical decision-making, and workflows that stay on the right side...

Revision between shifts is won on session length, one-handed usability and offline behaviour; here are the apps built for ten stolen minutes.