Best AI for OSCE Practice: Stations, Plans and Differentials

Which AI tools help with OSCE practice?

OSCE practice with AI splits into two jobs, and no tool in this guide does both. The first job is performing a station: taking a history, explaining a diagnosis or counselling a patient out loud against the clock, which needs someone or something to play the patient and mark you. Geeky Medics and PLABable's OSCE AI do this with AI patients, and ChatGPT's voice mode can hold a spoken role-play. The second job is checking the medicine behind the station: whether your management plan matches current guidance, whether your differential put the dangerous causes first, and what a result actually means. AskMed does the second job, answering from journals, clinical guidelines such as NICE, and drug labels, with a link after every claim. AskMed is not a patient simulator and does not mark stations, so pair AskMed with a simulator, a study partner or a course. The table shows what each tool is and what it is best for.

Geeky Medics

Geeky Medics' App Store listing describes more than 200 OSCE guides with step-by-step instructions, videos and checklists, and more than 1,300 OSCE scenarios. Its virtual patients are over 900 AI patients that, in the listing's words, "listen and respond like a real patient, by text, voice or video", with AI examiner feedback on how to improve. Group practice assigns candidate, patient and examiner roles so you can run stations with friends. The listing also names MLA AKT questions, PSA prescribing questions, and radiology, blood test, blood gas and ECG cases. Premium content needs an auto-renewing annual subscription.

PLABable OSCE AI

PLABable's OSCE AI page says you talk to an AI patient, take a history and get marked against a marking guide. It lists 153 stations, timed eight-minute stations in the same format as PLAB 2, a patient that answers your questions in real time, and a full assessment report with pass or fail for every station. Its sample report scores data gathering, clinical management and interpersonal skills, the three domains PLAB 2 is marked on. Because it is built around PLAB 2, it suits international graduates heading for UK practice most; best AI tools for PLAB 2 compares it with other PLAB 2 options.

ChatGPT

ChatGPT is OpenAI's general AI chatbot. Its App Store listing describes Advanced Voice Mode: tap the soundwave icon for a real-time spoken conversation. Give ChatGPT a scenario, ask it to play the patient and take the history out loud. It has no OSCE station bank or official mark scheme and is not limited to medical sources, so check its clinical content before you learn it. Is ChatGPT accurate for medical questions? covers what studies found.

What does an OSCE test in the UK, the US and elsewhere?

An OSCE, or objective structured clinical examination, tests what you do with a patient in a fixed time across a series of stations, each marked against set criteria, and the format differs by country. In the UK, the GMC's PLAB 2 for international graduates has 16 scenarios of eight minutes each, marked on data gathering, technical and assessment skills, clinical management, and interpersonal skills. UK medical students sit the clinical and professional skills assessment (CPSA), one of the two parts of the Medical Licensing Assessment, often called the UKMLA; each medical school sets and delivers its own CPSA, and the GMC quality assures it. In Australia, the AMC clinical examination has 16 assessed stations, each with two minutes of reading and eight minutes of assessment. In the US, the USMLE ended Step 2 Clinical Skills in January 2021, so there is no national licensing OSCE and clinical skills exams are set locally. The table compares these formats.

Sources: the GMC's PLAB 2 guide and CPSA page, the AMC's clinical examination page and the USMLE's announcement. Whatever the exam, the guidance that counts is the one that applies where you sit it. AskMed's guideline sources include NICE, the US Preventive Services Task Force, CDC, WHO and the ESC, but no Australian guideline body, so check Australian guidance directly for the AMC.

How do you practise a history-taking station with AI?

A history-taking station is practised best as a loop: perform the station out loud, find what you missed, fix the knowledge behind the gap, then perform it again. Step one is the performance, with a study partner, a course or an AI patient such as Geeky Medics' virtual patients or PLABable's OSCE AI, timed to your exam's station length, which is eight minutes in PLAB 2. Step two is the gap: write down the questions you forgot to ask, such as red flags, a drug history or the patient's concerns. Step three is the knowledge, and that is where AskMed fits: ask which features of the presenting complaint matter and why, for example "Which headache features need urgent investigation, according to NICE?", and AskMed answers from the guideline with a link after every claim. Step four is the next attempt at the same or a similar scenario. The loop below shows the order; AskMed's part is the checking, not the role-play.

• Chest pain. "Which features in a chest pain history point to acute coronary syndrome, aortic dissection or pulmonary embolism?"
• Abdominal pain. "In a woman of reproductive age with lower abdominal pain, what must be ruled out first, and with which test?"
• Feverish child. "Which features in a feverish child under 5 does NICE class as high risk?"
• Breathlessness. "Which questions separate heart failure, COPD and asthma in an adult with breathlessness?"

How do you check an OSCE management plan against guidelines?

To check an OSCE management plan, type out the plan you gave at the station, name the setting and the country, and ask AskMed what current guidance would add or change. AskMed is built to give complete management answers: before finishing, the AI checks its own draft so that every guideline-recommended option is named, the whole treatment pathway is visible, and the year of every guideline recommendation is stated. When guidance has changed, the answer gives the newest position and says what changed and since which year, which matters in an exam marked against current practice. Say "UK" or "NICE" for PLAB 2 and the CPSA, because US and UK guidance can differ. Use follow-ups to rehearse the twist an examiner adds: "and if she is pregnant?" is read together with your last question and answer. AskMed never gives doses, so for a prescribing element take the amounts from the BNF or the current drug label.

• Paste your plan. Use Paste note, or attach a PDF or a photo of printed notes; attachments explains what is read.
• Ask for the whole pathway. First-line and alternative options, what to do after no response, monitoring, and special situations such as pregnancy.
• Name the setting. A GP telephone consultation, an emergency department and a ward call for different plans.
• Check the year. A current guideline outweighs an older review; open the citation behind the step you will say aloud.

How can AI help you review differentials and explain findings?

For differentials, AskMed is instructed to list the can't-miss causes first and then the common ones, each with its distinguishing feature and first test, which is the order an OSCE examiner expects when you present a case. Describe the patient the way you would to the examiner, with age, the presenting complaint, key positives and negatives, and the setting, and ask for the differential. For findings, AskMed explains a lab test with its normal and decision thresholds, a stepwise interpretation, the causes of an abnormal result and common pitfalls, and gives both unit systems where countries differ, such as HbA1c in % and mmol/mol. AskMed reads attached images for their printed text only, so it can read a printed lab report but cannot interpret an X-ray, an ECG tracing or a skin photo; describe those findings in words. For explanation stations, use AskMed for the facts, then put them into plain words yourself and practise saying them aloud.

• Present, then ask. "Adult with pleuritic chest pain, tachycardia and a recent long-haul flight, in the emergency department: differential and first tests?"
• Interpret a result. "Low sodium with low serum osmolality: stepwise interpretation and causes?"
• Prepare an explanation. "What does a new diagnosis of type 2 diabetes mean for monitoring and treatment options?" Then explain it to your partner without jargon.
• Describe images in words. Rate, rhythm and ST changes for an ECG, or the zone and pattern of shadowing for a chest X-ray.

How AskMed answers lists every source AskMed searches and the rules each answer follows.

What are the limits of AI for OSCE preparation?

The main limit of AI for OSCE preparation is that an OSCE is a performance in front of an examiner, and no app can examine an abdomen, feel a pulse or practise a procedure on a manikin for you. Examination and procedure stations need hands-on practice in a clinical skills session or with a partner. AI patients and AI examiner feedback are written by AI too, so a mark from a simulator is a practice signal, not the judgement of a real examiner; the official guidance for your exam, such as the GMC's PLAB 2 guide, the MLA content map or your school's handbook, sets what is marked. Every AskMed answer is written by an AI language model from the sources its search found, and no person reviews it before you see it, so open the citation behind any point you will say to an examiner. AskMed searches a fixed list of medical sites, so guidance published elsewhere may not appear, and it gives no doses.

• Follow your exam's country. NICE and other UK guidance for PLAB 2 and the CPSA; the guidance your school teaches elsewhere.
• Practise with people. Communication, empathy and timing are marked, and a partner or examiner shows them best.
• Report what looks wrong. Tap the thumbs down under an AskMed answer and pick a reason, such as Inaccurate or Sources don't match.

Best AI tools for PLAB 2 covers the UK OSCE for international graduates in detail, and best AI apps for USMLE covers the US written exams.