This workflow is for Italian medical graduates preparing for the Concorso SSM who use Secret SSM's AI tutor (marketed as SELF) and want it to sharpen their reasoning rather than replace it. Used well, an AI tutor is a fast Socratic partner. Used lazily, it becomes an answer-vending machine that leaves you fluent at reading explanations and weak at generating them under time. The principal limitation to manage is answer leakage: the tutor gives you the reasoning you were supposed to produce.
What Secret SSM offers for the Concorso SSM right now
Vendor-reported unless attributed to the exam body; last checked 19 July 2026.
| Feature | What we found (vendor-reported, 19 July 2026) |
|---|---|
| Question bank (free) | ~500 unpublished items plus ~500 official past ("ministeriali") items |
| Question bank (paid) | 30,000+ unpublished items with explanations, plus ~1,700 annotated official past items |
| AI tutor | "SELF", described as a smart tutor that learns from top students' strategies |
| Simulations | Free: 1 assessment plus 12 official-paper simulations; paid: 50 custom plus 12 official, with explanations |
| Analytics | Score-curve tracking; weak-area identification |
| Price | Free starter kit; the paid simulator package reported at approximately €299 |
| Access period | Not clearly stated on the pages we reviewed — verify current terms on secretssm.com |
SELF is the reason to read this article. An AI tutor is powerful precisely because it will answer whatever you ask — which is also why it needs rules. The rest of this piece is a set of rules that keep the tutor on the teaching side of the line.
The exam anchor: decree, syllabus and released questions
The Concorso SSM is the national specialty-entrance examination governed each year by a ministerial decree from the Ministero dell'Università e della Ricerca (MUR). The format is fixed: 140 multiple-choice questions, each with five options and one correct answer, in 210 minutes, delivered by computer nationally. Scoring is +1 for a correct answer, −0.25 for a wrong answer and 0 for a blank; the maximum test score is 140, with up to 7 further points from your curriculum, giving a ceiling of 147. Your total places you in the national ranking (graduatoria) that decides specialty and location. There is no official syllabus split into chapters — the content spans internal medicine, general surgery, pharmacology, clinical pathology, imaging and the major specialties — and the ministry releases past question sets each year. Those official released questions, not any third-party bank, are your calibration standard. Confirm the current year's decree and rules before you rely on any figure here.
Prompt taxonomy: six jobs an AI tutor should do
Do not ask "why is the answer C?" That invites leakage. Instead, ask the tutor to do one of six defined jobs, each of which forces you to keep thinking:
- Mechanism — explain the pathophysiological or pharmacological mechanism behind an option, without naming which option is correct.
- Discriminating features — contrast two plausible diagnoses on the one or two features that separate them.
- Option elimination — give the reason each distractor is wrong, one at a time.
- Guideline verification — state the current first-line management and the guideline it rests on.
- Counterfactual — change one clinical detail and ask how the best answer changes.
- Retrieval testing — have the tutor quiz you on the same concept in a new stem, with no answer shown until you commit.
Because the negative marking penalises rash guessing (−0.25), option-elimination and discriminating-feature prompts pay off directly: they train the judgement of when to answer and when to leave a question blank.
Twenty-five prompts, grouped by the six most common error types
Keep these reusable and free of any copyrighted question text. Paste your own paraphrased scenario where a prompt calls for it.
Error type 1 — Mechanism gaps
- "Explain the mechanism linking [finding] to [condition] in three steps; do not tell me the answer."
- "What receptor or pathway does [drug class] act on, and what predicts its main adverse effect?"
- "Give the physiological reason [sign] appears in [condition], then ask me to predict one related sign."
- "Walk me through why [lab derangement] follows from [mechanism], without referencing any option."
Error type 2 — Differential / discriminating-feature errors 5. "List the two features that best separate [diagnosis A] from [diagnosis B]; do not say which is correct." 6. "Which single finding most changes the probability between these two diagnoses, and why?" 7. "Give me three conditions that share [presentation] and the one discriminator for each." 8. "If I only had [one investigation], how would it distinguish these differentials?"
Error type 3 — Option-elimination failures 9. "For a stem about [topic], explain why each of five generic distractor types would be wrong." 10. "Give the specific reason a plausible-but-wrong option in [topic] is a trap." 11. "What cognitive bias makes candidates pick the wrong answer in [topic] questions?" 12. "Rank these four management options from least to most appropriate and justify the order."
Error type 4 — Guideline / first-line errors 13. "State the current first-line management for [condition] and name the guideline and its year." 14. "Has first-line therapy for [condition] changed recently, and if so, from what to what?" 15. "What is the recommended monitoring for a patient started on [drug], and at what intervals?" 16. "Give the threshold at which management for [condition] escalates, with the source."
Error type 5 — Calculation / pharmacokinetic errors 17. "Show the full working for [dose/rate/clearance calculation], step by step, with units." 18. "What is the commonest arithmetic slip in [calculation type], and how do I avoid it?" 19. "Convert [dose] between routes and explain the factor used." 20. "Quiz me on three variations of this calculation and mark my working."
Error type 6 — Retrieval / recall failures 21. "Ask me to recall the classification of [condition] before you show any answer." 22. "Give me a new stem testing the same concept and wait for my answer and reasoning." 23. "Test me on the same topic in one week's format; do not reveal the answer first." 24. "Turn the concept I just missed into three flashcard questions." 25. "Ask me to teach [concept] back to you in four sentences, then correct my version."
Notice that most prompts end by putting the work back on you. That is the point: a tutor that only quizzes and corrects builds recall; one that hands you finished reasoning builds only recognition.
The anti-answer-leak rule
Before you open the tutor on any item, commit in writing to your answer and a one-line rationale. Only then ask a prompt from the taxonomy above — and start with elimination or mechanism, never "what's the answer?". If your committed answer was wrong, the gap between your rationale and the tutor's is your actual learning point. Skip this step and you will feel productive while learning to recognise good reasoning rather than to produce it under exam pressure. This principle is the whole basis of the iatroX Socratic Tutor and of our guidance on auditing an AI tutor.
Jurisdiction and currency check
AI tutors are trained on mixed international material and may quote guidance that does not match Italian practice or is out of date. Keep one reusable prompt to hand and run it whenever management or thresholds are at stake: "State whether this recommendation reflects current Italian practice, name the specific guideline and its year, and flag any point where international and Italian guidance differ." If the tutor cannot name a source and a date, treat its answer as a hypothesis to verify, not a fact to memorise.
Create a misconception record
Every corrected error earns a four-line entry: one sentence for the wrong rule you were using; one for the corrected rule; one transfer question that tests the same idea in a new context; and a review date a week or two out. This converts a fleeting "ah, of course" into a durable correction and gives you a ready-made unseen retest for later. A dozen honest entries a week is worth more than a hundred explanations skimmed.
Weekly verification sample
Once a week, take a small random set — five to ten — of the tutor's substantive claims and check them independently against official released questions, current Italian guidelines or a primary source. Log every discrepancy. Two or three material errors in a sample is a signal to tighten your jurisdiction prompt and to lean harder on official material for anything guideline-dependent. This is not distrust for its own sake; it is the only way to know how far the tutor can be trusted on the topics that matter to your ranking.
Worked example: a seven-day plan
One job per tool: the SELF tutor teaches; iatroX supplies fresh, unseen, timed items that show whether the reasoning transferred. No proprietary-algorithm claims — just a disciplined loop.
- Day 1 — 40 mixed items under time in Secret SSM. Commit answers first. Mark by the six error types, not just right/wrong.
- Day 2 — Take your three commonest error types into the tutor using only taxonomy prompts; write misconception-record entries.
- Day 3 — Run the jurisdiction prompt on every guideline-dependent correction from Day 2; verify two against official material.
- Day 4 — Retrieval day: have the tutor quiz you (prompts 20–25) with no answers shown until you commit.
- Day 5 — Fresh, unseen, timed iatroX block on the same topics. If you improve on new items, the corrections stuck; if not, revisit the misconception records.
- Day 6 — One full official-paper simulation under strict 210-minute timing; practise the leave-or-answer decision against the −0.25 penalty.
- Day 7 — Weekly verification sample; update your prompts; schedule next week's transfer retests.
Continue, supplement, switch or stop
- Continue if your unseen scores are rising and your misconception records are shrinking week on week — the tutor is teaching, not spoon-feeding.
- Supplement if you have leant so heavily on explanations that you can recognise reasoning but not generate it; add more no-answer retrieval and unseen timed blocks.
- Switch the way you use the tutor (not necessarily the product) if your weekly verification keeps surfacing jurisdiction or currency errors — route guideline questions to official material instead.
- Stop any prompt habit that leaks answers before you have committed; it is quietly training the wrong skill.
Bottom line
Secret SSM's SELF tutor is a capable, well-stocked study partner for the Concorso SSM, and its large item pool plus official-paper simulations give you real volume. The risk is not the tool but the habit: an AI that answers on demand will happily do the thinking the exam expects from you. Commit before you consult, prompt for jobs rather than answers, verify anything guideline-dependent against official material, and keep a separate unseen source to prove transfer. Do that, and the tutor accelerates your reasoning instead of replacing it.
Frequently asked questions
Is Secret SSM enough for the Concorso SSM on its own? Its paid tier reports a very large item pool with explanations and official-paper simulations, which is a strong content base for one platform. What no single bank guarantees is that your reasoning transfers to unseen items under the −0.25 penalty, so most candidates keep the official released questions as their calibration standard and add an unseen source for timed measurement. Prices and access terms were vendor-reported on 19 July 2026 and should be reconfirmed on secretssm.com.
Which Concorso SSM component does Secret SSM not reproduce well? No AI tutor reliably reproduces the discipline the real exam demands: committing to an answer under time and accepting the negative-marking risk with no assistance. The tutor is excellent at explaining after the fact, but the leave-or-answer judgement has to be trained on unseen, timed items and on official-paper simulations, not in a conversation where help is one prompt away.
How should I verify Secret SSM AI answers for the Concorso SSM? Run a weekly verification sample: take five to ten of the tutor's substantive claims at random and check them against official released questions, current Italian guidelines or a primary source, logging every discrepancy. For any answer that hinges on management or thresholds, use a jurisdiction prompt that forces the tutor to name the guideline and its year. If it cannot cite a source and date, treat the claim as unverified.
When should I stop using Secret SSM and move to mixed mocks? Move to full, mixed, timed simulations once your error records are shrinking, your unseen first-attempt accuracy has stabilised, and you have practised the leave-or-answer decision enough that pacing is not a threat. That is usually the final couple of weeks. Do not keep drilling explanations when your limiting factor has become exam-condition performance rather than knowledge.
How should I combine Secret SSM with iatroX without duplicating practice? Give each a distinct role. Use Secret SSM's tutor to diagnose and teach the reasoning behind your errors; use iatroX for fresh, unseen, timed blocks that measure whether those corrections held. Never feed an item you have already seen into the other tool — the second source is only valuable because its questions are new. Open a missed item in the iatroX Socratic Tutor, commit your answer and rationale first, and only then compare.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026. Item counts, the SELF tutor description, pricing and access terms are vendor-reported and change without notice; reconfirm them on secretssm.com. Disclosure: iatroX operates a competing question bank and Socratic Tutor; its role here is confined to a job Secret SSM's tutor does not claim — unseen, timed transfer measurement and answer-first verification. Corrections are welcome via the feedback route on iatrox.com.
References: Ministero dell'Università e della Ricerca, concorso nazionale SSM and annual decree (mur.gov.it); Secret SSM (secretssm.com) and its SSM simulator (secretssm.com/simulatore-ssm); iatroX Concorso SSM bank (iatrox.com/concorso-ssm); how to audit an AI medical exam tutor (iatrox.com); Your Q-Bank Percentage Is Not Your Exam Score (iatrox.com).
Open a missed Concorso SSM item in the iatroX Socratic Tutor →