This audit is for UK physician associates preparing the clinical component of the Physician Associate Registration Assessment (PARA) who are weighing up the Geeky Medics AI virtual-patient simulator. Used deliberately, it is a useful way to rehearse history-taking and communication under conversational pressure. The principal limitation is worth stating plainly at the outset: there is no PARA-specific Geeky Medics simulator, and a generative virtual patient cannot reproduce a standardised sixteen-station OSCE circuit, examiner variability or hands-on physical examination. iatroX covers the written knowledge layer, not the OSCE.
What Geeky Medics offers for PARA right now
The most useful thing to establish before you spend money is what actually exists. Geeky Medics does not, as of the last check, sell a product badged for the Physician Associate Registration Assessment. The "Para–Physician Associate Exam Practice" item that surfaces in its learning library is a ten-question community quiz created by a student user, not an official, blueprinted PARA bank — treat it as incidental, not as coverage. What Geeky Medics does offer that is relevant to a PA sits in two buckets: a general AI virtual-patient simulator built for OSCE-style rehearsal, and a set of multiple-choice question banks whose content overlaps PARA knowledge without being mapped to the PARA blueprint.
The figures below are vendor-reported and were last checked on 19 July 2026. Verify the live counts, access period and price on geekymedics.com before you buy, because these change between releases.
| Feature | Vendor-reported (last checked 19 July 2026) |
|---|---|
| PARA-specific product | None advertised; a 10-question "PA exam practice" quiz exists but is user-generated |
| Virtual-patient simulator | AI-Patients ("SimChat") — 900+ virtual patients; text, voice and video-avatar modes |
| OSCE station library | 1,300+ expert-authored OSCE stations; several hundred AI OSCE scenarios |
| AI examiner | Assesses the interaction, completes a checklist and returns written feedback |
| Branching | Conversation adapts dynamically to what the candidate actually says |
| Timing / scoring | No fixed station timer or standardised numerical score stated on the simulator pages |
| MCQ content relevant to PARA | Tens of thousands of free MCQs; 3,000+ MLA AKT items with two mocks; 600+ PSA prescribing items |
| Price and access period | Not published on the pages checked — verify current bundle pricing on geekymedics.com |
Two points deserve naming. First, the simulator is an OSCE-and-communication tool; it maps to the PARA clinical exam, not to the written applied-knowledge paper. Second, because none of it is PARA-badged, the responsibility for choosing the right cases and matching them to the PARA blueprint falls on you, not on the product.
The PARA you are actually preparing for
The Physician Associate Registration Assessment is delivered by the Royal College of Physicians (RCP) Assessment Unit, and it matters because the General Medical Council began regulating physician associates on 13 December 2024. From that point the PARA became the gateway to GMC registration, not an optional badge. The exam has two components. The written applied-knowledge exam — the Knowledge-Based Assessment — is a single-best-answer paper of around 200 questions delivered online, split into four one-hour papers of 50 questions each (roughly four hours of testing plus breaks). The clinical exam is an OSCE of 16 stations: 14 clinical scenario stations and two rest stations, with about two minutes of reading and eight minutes inside each station. These figures are body-reported and were last checked on 19 July 2026; verify the current question count and OSCE station count on rcp.ac.uk, and note the transitional arrangements running from 13 December 2024 until 13 December 2026, during which hybrid blueprints apply across curriculum versions.
The distinction between official requirements and third-party claims is the whole game here. The GMC content map and the RCP's own OSCE blueprint and candidate regulations are the authoritative documents. No third-party simulator, Geeky Medics included, can "be the OSCE"; it can only rehearse some of the skills the OSCE samples. Hold every vendor claim against the RCP blueprint rather than the other way around.
Format map: what the simulator reproduces, and what it does not
An honest audit inventories the tasks. The table maps PARA clinical tasks against what the Geeky Medics simulator can rehearse.
| PARA / OSCE task | Simulator reproduces it? |
|---|---|
| Opening a consultation, structured history | Yes — this is its core strength |
| Focused questioning and red-flag screening | Yes, if your case script includes them |
| Ideas, concerns and expectations | Yes — conversational and adaptive |
| Explanation, counselling, shared management (verbal) | Yes |
| Breaking bad news / difficult conversations | Yes — a named use case |
| Physical examination findings | No — you cannot examine a virtual patient |
| Procedural and practical skills stations | No |
| Standardised bell-driven timing and rest-station rhythm | No — self-imposed only |
| Marking against the official OSCE domain descriptors | No — returns a checklist, not the RCP rubric |
| Trained-actor and examiner variability across a live circuit | No |
The pattern is clear. The simulator is strong on the verbal, cognitive and interpersonal half of an OSCE station and structurally silent on the physical, procedural and standardisation half. That is not a flaw so much as a boundary, and knowing where the boundary sits is the point of the audit.
Fidelity test: timing, interface, scoring, branching, permitted actions
Compare the simulator against current official candidate guidance on five axes rather than on marketing. Timing: the real OSCE runs on a fixed cycle of roughly two minutes' reading then eight minutes in the station, driven by a bell; the simulator conversation is open-ended, so you must self-impose an eight-minute limit or you will train an unrealistic pace. Interface: the exam is a physical circuit with equipment, a real simulated patient and, where relevant, examination; the simulator is a screen-based chat, voice or avatar exchange. Scoring categories: the OSCE is marked against domain descriptors — data gathering, clinical management, interpersonal skills and patient safety among them — while the AI examiner completes a generic checklist and writes prose feedback that does not map one-to-one onto those domains. Branching: live actors follow a tightly written script with deliberately limited branches so every candidate meets the same case, whereas the simulator branches freely to whatever you say, which is excellent for improvisation but can drift off-blueprint. Permitted actions: in the exam you can request examination findings, use equipment and respond to a real person's non-verbal cues; in the simulator you can only ask verbally. Each mismatch is a place where uncritical practice can quietly teach the wrong habit.
Case-mix audit: are the proportions realistic?
Because you select or generate the cases, the default case mix is simply whatever you find comfortable — and that is the trap. A PARA-representative circuit spans common and acute presentations, a smaller number of chronic-disease reviews, explicit communication and ethics stations, and safety-critical scenarios such as the deteriorating patient or a safeguarding concern. Rare zebras are low yield. Build your rota against a target distribution rather than mood.
| Case type | Rough target share of your practice | Why it matters for PARA |
|---|---|---|
| Common acute presentations | ~40% | Highest-yield; the bulk of the circuit |
| Chronic-disease review and management | ~20% | Tests longitudinal reasoning and safety-netting |
| Communication / breaking bad news | ~15% | An explicit OSCE strength area for the simulator |
| Ethics, consent and capacity | ~10% | Frequently under-practised |
| Safety-critical / deteriorating patient | ~10% | High consequence, easy to avoid |
| Rare or atypical presentations | ~5% | Diminishing returns above this |
If your last twenty simulated cases were all comfortable histories, your fluency is real but narrow. Rotate deliberately.
Feedback audit: observable behaviour, inferred competence, model commentary
Automated feedback comes in three flavours, and they are not equally trustworthy. Observable behaviours — did you screen for red flags, did you summarise, did you check understanding — are the most reliable thing an AI examiner can score, because they are close to a checklist. Inferred competence — "your clinical reasoning was strong" — is a model's guess about an internal state, and it can be confidently wrong. Model-generated commentary is fluent prose that reads like expert judgement but is generated text, not a marked domain score. Treat the checklist-style observations as useful, the competence inferences as hypotheses, and the free-text praise as motivational rather than evaluative. Where the stakes are high, calibrate the automated score against a human before you trust it; our framework piece on calibrating AI-graded OSCE feedback sets out how to do this without over-reading a machine mark.
Repetition risk: false fluency from a small library
The failure mode specific to any simulator is memorising the library. Run the same twelve cases ten times and your scores rise — but you are rehearsing recall of those cases, not transferable consultation skill. The OSCE presents unseen scenarios, so the readiness signal you want is performance on cases you have never met, not a polished run through favourites. Rotate unseen scenarios, generate fresh variants, and reserve a block of stations you have never attempted for the final fortnight. If your score only holds up on familiar cases, you have trained fluency, not competence.
What it cannot test
Be explicit about the gaps so you can fill them elsewhere. The simulator cannot reproduce genuine non-verbal behaviour and the two-way reading of a real person's cues, examiner variability across a live multi-station circuit, local workflow and equipment, or — most importantly — physical examination and procedural skills. Those require real people: peer practice, a study partner playing the patient, and supervised clinical placement. The simulator is a supplement to that human practice, not a substitute for it.
A worked seven-day plan: simulator for one job, iatroX for measurement
This loop uses the Geeky Medics simulator for one defined job — spoken OSCE rehearsal — and iatroX for the different job of measuring written knowledge on unseen, timed single-best-answer items. No proprietary-algorithm claims are made; the value is simply spaced retrieval plus a blueprint-level readout.
- Day 1 — Two simulated history stations, each self-timed to eight minutes. The same evening, sit 40 unseen, timed SBA items in iatroX and log every miss by system.
- Day 2 — One communication and one ethics station; review the AI feedback, keeping only the observable-behaviour points. Then 30 unseen SBA items weighted to the systems you missed.
- Day 3 — Two acute presentations; add a real study partner for one physical-examination station the simulator cannot do. 30 unseen SBA items.
- Day 4 — Chronic-disease review plus safety-netting; 30 unseen SBA items including prescribing and data-interpretation stems.
- Day 5 — Two unseen simulated cases you have never generated before; note whether your score holds. 30 unseen SBA items.
- Day 6 — A safety-critical deteriorating-patient scenario with a peer; 40 unseen SBA items across mixed systems.
- Day 7 — Rest or a light mixed block. Map the week's misses — spoken and written — and decide the next action: learn, retest, simulate or stop.
Your iatroX percentage is not your exam score; it is a trend line and a coverage map, as the standard caveat article explains. Watch the unseen written score and the unseen-case OSCE performance move together.
Decision checklist: continue, supplement, switch or stop
- Continue the simulator if your consultation scaffolding is still improving and unseen-case performance is trending up.
- Supplement — the right call for most candidates — because the simulator does not touch the written KBA or physical examination. Add a knowledge bank and human OSCE practice.
- Switch emphasis to full-circuit timed mocks with real people once your spoken structure is reliable and you are within two to three weeks of the exam.
- Stop solo simulation if you are only re-running familiar cases and scores no longer move; convert to unseen stations and mixed mocks.
Base each decision on measurable gaps — blueprint coverage, unseen-case performance, error type — not on novelty or sunk cost.
Frequently asked questions
Is Geeky Medics enough for PARA on its own? No, and it does not claim to be. There is no PARA-badged Geeky Medics product; the simulator rehearses OSCE communication and history-taking but does not cover the written Knowledge-Based Assessment blueprint or physical examination, and its MCQ banks are aligned to the MLA and PSA rather than mapped to PARA. It is a strong communication-rehearsal supplement inside a plan that also includes a knowledge bank and human OSCE practice, not a complete PARA solution.
Which PARA component does Geeky Medics not reproduce well? The standardised OSCE circuit and the written exam. The simulator cannot deliver bell-driven timing, examiner marking against the official domain descriptors, physical examination or procedural stations, and it is not the applied-knowledge paper at all. It reproduces the verbal and interpersonal half of a station well and the rest of the exam poorly or not at all.
How many unseen Geeky Medics cases or stations should I preserve for final PARA calibration? Ring-fence a set you have never attempted — around eight to ten stations spread across common acute, chronic, communication, ethics and safety-critical types — and save them for the last ten to fourteen days. Practising only familiar cases inflates your confidence; unseen stations are the honest readiness signal, so protect a reserve that no amount of earlier practice has contaminated.
When should I stop using Geeky Medics and move to mixed mocks? When your consultation structure is reliable across unseen cases and you are within two to three weeks of the exam, shift the balance to full-circuit timed mocks with real people, because stamina across a live multi-station rota, examiner variability and physical examination are precisely what the simulator cannot train. Keep the simulator only for targeted repair of a specific weak communication skill.
How should I combine Geeky Medics with iatroX without duplicating practice? Give each tool the job it is built for: the Geeky Medics simulator for spoken OSCE rehearsal, and iatroX for unseen, timed written single-best-answer measurement of the knowledge layer. When you miss an item, compare the error type — knowledge gap, misread stem, or reasoning slip — rather than the wording, so the two resources measure different things instead of repeating the same practice.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 19 July 2026. Vendor figures are vendor-reported and change between releases; verify the current Geeky Medics counts, price and access period on geekymedics.com, and confirm the PARA format on rcp.ac.uk before relying on any number here. Disclosure: iatroX operates a competing question bank covering the PARA written knowledge layer; it is explicitly not an OSCE simulator, and this audit confines iatroX's role to the knowledge/SBA job that the Geeky Medics simulator does not claim. Corrections are welcome via the feedback route on iatrox.com.
References: RCP — Physician Associate Registration Assessment, information for candidates and OSCE blueprint (rcp.ac.uk); GMC physician associate regulation (from 13 December 2024); Geeky Medics product pages (geekymedics.com); iatroX PARA question bank (/quiz-landing); "Your Q-Bank Percentage Is Not Your Exam Score" (iatrox.com/blog/qbank-percentage-not-your-exam-score); calibrating AI-graded OSCE feedback (iatrox.com/blog/ai-graded-saqs-and-osces-how-to-calibrate-automated-feedback-before-you-trust-the-score); iatroX comparison hub (/compare).