For durable recall on medical exam blueprints, a question-first, source-grounded, exam-calibrated tutor beats an on-demand explainer or a general chatbot. But that is not the whole picture. Each of these tools has legitimate use cases, genuine strengths, and specific audiences it serves well. An honest comparison says where each wins — including where iatroX is not the right choice.
One caveat before the detail: Pastest Tutor and Osmosis AI features evolve. The descriptions below were verified against their public feature descriptions at the time of writing. If you are reading this months later, check their current feature sets directly at pastest.com and osmosis.org. Feature access dates are noted where relevant.
The Four Approaches, Characterised Fairly
ChatGPT (OpenAI). The general-purpose large language model that most trainees already use. It will explain any medical concept, generate clinical reasoning for any question, produce structured study notes, and answer follow-up questions with impressive fluency. It has no exam-specific pedagogy (it does not know the MRCGP AKT blueprint), no curriculum mapping (it cannot weight questions by exam emphasis), no source grounding in specific UK guidelines (it generates from training data, not from NICE/CKS/SmPC), and no mechanism to diagnose your individual misconceptions (it explains the topic, not your specific error).
ChatGPT's strength is unmatched versatility: it handles any question, any format, any level, any language, with remarkable fluency. Its weakness for exam preparation is precisely the one the crutch-effect evidence warns about: it provides the answer before you have attempted retrieval, removes the cognitive effort that produces durable learning, and does so with such fluency that it feels like effective studying. The Bastani et al. PNAS study showed students using an unrestricted ChatGPT-style interface performed 17% worse when the AI was removed.
Pastest Tutor. Pastest has introduced AI tutoring features alongside its established and well-respected question bank. Pastest has been a dominant UK medical exam preparation resource for over two decades, with large question volumes across MRCP, MRCGP, MRCEM, and other UK postgraduate exams. The AI tutor provides Q&A assistance and concept explanations on demand — a convenience-focused approach that allows trainees to get immediate help when stuck on a question.
Pastest's core strength is its question bank: large, well-established, with detailed explanations built over years of editorial investment. The AI tutor appears to be primarily convenience-oriented — providing answers and explanations when the trainee asks — rather than pedagogically diagnostic (identifying the trainee's specific misconception and forcing retrieval before explaining). The tutor adds a conversational AI layer to an existing product rather than redesigning the learning interaction around Socratic principles.
Osmosis AI. Osmosis is a medical education platform combining video content, question banks, and AI-powered study tools, widely used particularly at undergraduate level in the US. Its AI features support content retrieval and concept explanation — helping trainees find relevant educational material and understand topics through a combination of text and video. Osmosis's strength is its multimedia approach: high-quality video explanations, visual learning aids, and integrated content across the breadth of medical education.
Osmosis AI appears primarily as a content retrieval and explanation system rather than a question-specific Socratic tutor. Its value proposition is "find and understand any medical concept" rather than "diagnose your specific misconception on this specific exam question and force you to retrieve the answer." Different tools for different jobs.
iatroX Socratic Tutor. The iatroX tutor is Socratic-first by default. It binds to the specific question you just answered — it has the question stem, your answer, the correct answer, the official explanation, and the guideline reference as context. It diagnoses your specific misconception (not the general topic, but the exact reasoning error you made) before teaching anything. It withholds the answer to force retrieval (you must attempt the answer before receiving correction). It grounds every clinical claim in NICE, CKS, SmPC/eMC, SIGN, or the exam's own official explanation (not generated from training patterns). And it calibrates per exam — speaking differently for MRCGP AKT, GPhC CRA, MRCP Part 1, DTM&H, or Italian SSM.
Why the Differences Matter: The Pedagogy
The differences between these four tools are not merely cosmetic or feature-level. They map directly onto the evidence about how AI affects learning — and specifically onto the crutch effect documented in the Bastani et al. PNAS study and the 45-day retention deficit.
Retrieval before explanation. Does the tool make you attempt the answer before providing it? Retrieval practice (Roediger & Karpicke, 2006) is among the strongest known drivers of durable learning. Every retrieval attempt — successful or unsuccessful — strengthens the memory trace. Tools that explain first bypass this entirely: the trainee reads the answer instead of producing it. ChatGPT explains immediately. Pastest answers on demand. Osmosis retrieves and explains content. iatroX withholds the answer and asks you first.
Misconception diagnosis. Does the tool identify what you specifically got wrong, or provide a generic explanation? A generic explanation of diabetes management does not help if your specific error was confusing the eGFR threshold for stopping metformin with the threshold for dose review. A diagnostic tutor identifies the exact reasoning error and targets it — which is more efficient and more effective than re-explaining the entire topic.
Source grounding. Does the tool generate explanations from training data, or anchor them to verifiable clinical sources? For medical exam preparation, ungrounded generation risks creating false clinical knowledge — a plausible-sounding explanation that is subtly wrong about a threshold, a contraindication, or a monitoring interval, and that the trainee accepts because it sounds authoritative.
Loop closure. Does the tool test you afterwards? If the tutoring session ends with an explanation and no follow-up retrieval event, the knowledge remains fragile. A closed learning loop includes a delayed retrieval test that feeds back into spaced repetition — confirming that the learning was durable, not just momentary.
Feature Comparison Table
| Feature | ChatGPT | Pastest Tutor | Osmosis AI | iatroX Socratic Tutor |
|---|---|---|---|---|
| Socratic-first default | No — explains immediately | No — answers on demand | No — retrieves/explains | Yes — asks you first |
| Diagnoses your specific misconception | No — generic explanation | Limited | No | Yes — question-specific |
| Source-grounded (NICE/CKS/SmPC + official explanation) | No — training data | Partial (Pastest editorial content) | Partial (Osmosis content library) | Yes — guideline + exam grounded |
| Per-exam calibration | No — general purpose | Partial — UK exam coverage | Limited — primarily US/UG | Yes — per exam, per register |
| Retrieval challenge → spaced repetition | No | No | No | Yes — feeds adaptive engine |
| Confidence calibration | No | No | No | Yes — metacognitive scaffolding |
| Anki export | No | No | Flashcards available | Yes — /anki |
| Mobile revision | Via app/web | Via app | Via app | Yes — React Native iOS/Android |
| UK exam coverage breadth | None (general) | Strong UK postgrad | International, US-weighted | 15+ UK/intl exams |
| Free tier | Yes (limited) | No | Limited | Yes — free UK core exams |
Who Each Is Genuinely Best For
ChatGPT is best for trainees who want fast, unconstrained factual lookups across any topic without exam-specific grounding — and who are comfortable managing the crutch-effect risk through their own discipline. ChatGPT is also genuinely valuable for tasks that are not exam-specific retrieval practice: generating study materials, summarising textbook chapters, explaining concepts at different levels, exploring clinical reasoning outside a structured exam context, and creating mnemonics or revision aids. It is the best general-purpose medical AI tool. It is not the best exam preparation tutor.
Pastest is best for trainees who want an established UK Q-bank with a large question pool, detailed explanations, and the reassurance of a brand that has supported UK medical exam candidates for decades. The AI tutor adds convenience; the core value is the question bank itself. If the trainee's primary need is volume of practice questions with comprehensive written explanations, Pastest delivers — and the AI layer adds an additional route to those explanations.
Osmosis is best for visual learners, particularly at undergraduate level, who benefit from video explanations and multimedia medical education. Osmosis's AI complements its content library. If the trainee learns best through watching rather than reading, and primarily needs concept understanding rather than exam-specific retrieval practice, Osmosis is a strong fit.
iatroX Socratic Tutor is best for trainees who want question-first, diagnostic, Socratic tutoring that forces retrieval, identifies specific misconceptions, and grounds answers in UK clinical sources — particularly trainees preparing for UK postgraduate exams who need guideline-concordant answers from NICE/CKS/SmPC. The combination of Socratic pedagogy, source grounding, per-exam calibration, and spaced-repetition-integrated loop closure is what no single competitor replicates.
iatroX is not the right choice for someone who only wants fast factual lookups with no retrieval demand (ChatGPT is better). Someone who wants video-based learning as their primary revision mode (Osmosis is better). Someone who specifically wants the largest possible question volume in a single UK exam and does not value Socratic pedagogy (Pastest may have more questions in some established exams). Someone preparing exclusively for US exams who wants the deepest US-specific question bank (UWorld remains the benchmark for USMLE volume and explanation depth). Someone who finds the Socratic approach frustrating and genuinely prefers to read explanations — which is a legitimate preference, even if the evidence suggests it produces less durable learning.
Those honest "not us" lines matter. The comparison is credible because it acknowledges where each tool genuinely fits — including where iatroX does not. Trust is built by honesty, not by claiming superiority in every dimension.
The Combination That Matters
The strongest single differentiator is not any individual feature — it is the combination. Socratic-first + source-grounded + per-exam calibrated + misconception-diagnostic + retrieval-closing + spaced-repetition-integrated + free UK core exams + 15+ exam coverage. No single competitor currently replicates all of these in one product. ChatGPT has none of the exam-specific features. Pastest has the questions and UK coverage but not the Socratic pedagogy or source-grounded AI. Osmosis has content but not question-specific diagnostic tutoring. iatroX combines them into one integrated revision workflow.