Three principles do most of the heavy lifting in memory research: retrieval practice (testing beats re-reading), spacing (distributed practice beats massing) and adaptive sequencing (practice should chase your weaknesses, not your comfort). The 2026 tools implement these to very different depths, and the biggest split is between scheduling cards and adapting questions.
Card scheduling versus question-level adaptation
Card scheduling is the Anki lineage: each flashcard carries its own review clock, and the algorithm decides when it returns. It is superb for atomic facts. Question-level adaptation is a different machine: the system watches your performance across a curated bank and decides which question, topic and difficulty you meet next. One optimises when a fact returns; the other optimises what your limited practice time samples. Serious revision wants both.
Who does what
iatroX builds both into the bank itself: spaced repetition returns questions as forgetting approaches, and the adaptive engine reshapes sequencing after every answer, with the Socratic Tutor feeding missed-question misconceptions back into the queue. The revealing test is what happens after a wrong answer: on iatroX, the error changes next week's practice, not just this minute's explanation.
AMBOSS pairs its Q-bank analytics with library-linked review, strong on identifying weak domains, lighter on automated return scheduling.
Neural Consult generates flashcards from your materials and supports Anki export, effectively delegating scheduling to the best card scheduler ever built while GLIA handles the tutoring.
Anki-supported workflows remain the gold standard for card scheduling, free and infinitely customisable, with the standing costs of manual deck curation and zero connection to any exam blueprint.
Question-bank analytics elsewhere (UWorld, Lecturio, Osmosis) tell you your weak areas with varying automation of what to do about it; the burden of acting on the dashboard often stays with you.
As we argued in our piece on why retrieval practice beats reading AI answers, the platform matters less than whether effortful recall actually happens on schedule. But platforms that automate the schedule make it happen far more often.
A seven-day cycle to copy
Day 1: 40 adaptive questions, no topic filter; let the engine find you. Note the two weakest topics it exposes. Day 2: 20 questions targeted at weakness one; run the Tutor on every miss. Day 3: 20 on weakness two, same rule; ten spaced reviews from day 1 will start resurfacing. Day 4: light day, reviews only, ten minutes on mobile. Day 5: 40 mixed adaptive questions; watch whether the week's misses return and whether you convert them. Day 6: one timed mini-mock or block under exam conditions. Day 7: reviews plus a fifteen-minute read of the guideline behind the week's most stubborn topic. Then repeat, letting the engine re-diagnose.
The cycle is deliberately boring. Spacing works precisely because it is distributed, unglamorous and relentless, and the best tool is whichever one makes the boring loop automatic for you.
Frequently asked questions
Anki or built-in scheduling?
Decide by where your content lives and how much curation you enjoy. Anki remains unmatched as a pure scheduler and costs nothing; it also makes you the librarian of your own deck forever. Bank-integrated scheduling, the iatroX model, trades that control for automation: the questions, the misses and the returns live in one system that needs no maintenance. The wrong answer is the common one, half-running both so that neither system sees your whole history.
How many reviews a day is sustainable?
Whatever number you will actually clear on your worst day, not your best. Reviews compound: a target calibrated to good days collapses in the first busy week and the backlog then poisons motivation. Most working candidates sustain twenty to forty question-level reviews daily; set caps accordingly, and let new material slow down before reviews get skipped, never the reverse.
What do I do when I fall behind?
Triage rather than binge. Clearing a five-hundred-review backlog in one grim evening feels virtuous and re-teaches your brain that the system produces misery. Better: cap daily reviews, let the scheduler re-sort by urgency, and accept a fortnight of catch-up, which is precisely the failure mode adaptive platforms handle for you by redistributing rather than accumulating.
Does spacing work for imminent exams, or only long campaigns?
The full power needs weeks, but even a fortnight benefits: short intervals still beat massing, and the final-week priority becomes converting recent misses rather than meeting new material. If your exam is truly days away, run your error list on short cycles and forgive the theory; spacing is a strategy, not a religion.
Do I keep reviewing after the exam?
For content your career keeps using, a light maintenance schedule is one of the quiet best habits in medicine, and platforms that persist across exams make it nearly free; for exam-specific minutiae, let it go with a clear conscience. The skill you should definitely keep is the loop itself, since the next exam, and the next guideline update, will want it again.