How to Use AI Mock Vivas for MRCPsych CASC Prep

AI mock vivas for MRCPsych CASC work best when you shrink them to the size of a real station. The exam is a 16-station OSCE-style assessment completed in two circuits on the same day, with seven minutes in each station, so your rehearsal should feel brief, focused, and slightly pressurised rather than like a long revision chat with a chatbot.

That matters because this exam tests specific parts of a psychiatric encounter, not a whole clinic review. Each station focuses on history taking, examination, or patient management, and communication is assessed in every station, so your AI setup should ask for one task, one role-player voice, and one short debrief.

Why this works for CASC

Used well, AI gives you volume. You can run ten first-attempt stations in an evening, rehearse awkward explanations out loud, and repeat the same scenario until your opening minute sounds calm instead of scrambled.

Used badly, it teaches the wrong reflexes. If you let it reward long speeches, generic empathy, or textbook monologues, you will get better at sounding polished and worse at actually completing the station.

The real exam is interactive. RCPsych states that trained role players may respond differently depending on the skill the candidate shows, which is one reason AI should sit alongside human practice, not replace it. Use AI for repetition and structure; use colleagues, supervisors, or study partners for realism and emotional nuance.

Practical takeaway: let AI be your reps machine, not your final judge.

Build AI mock vivas that match the real MRCPsych CASC station

Your prompt should mirror the station format, not your anxiety. The current structure is still two circuits: a morning circuit of six management stations, one examination station, and one history station with four minutes' reading time, followed by an afternoon circuit of four history and four examination stations with 90 seconds' reading time. Linked stations no longer form part of the exam.

A reusable prompt template

Ask the AI to do five things every time:

give you a candidate instruction only, not the answer
state the setting clearly, such as ward review, emergency department liaison, memory clinic, CAMHS assessment, or tribunal-related discussion
play one role consistently, for example patient, relative, nurse, GP, or consultant
enforce the correct reading time and a hard seven-minute stop
give feedback under short headings: task focus, structure, communication, risk or safety, and closure

A simple prompt is enough:

Simulate an MRCPsych CASC station.
Give me only the candidate instructions and station setting.
Wait for my spoken answer.
Stay in character if I ask questions.
Stop me at seven minutes.
Then give concise feedback and one repeat task.

If you use Viva Examiner, build two saved modes: one strict examiner mode and one coaching mode. Strict mode should interrupt less and score the task. Coaching mode should pause after the first minute and tell you what is drifting.

Build balanced sets from the blueprint

Don't keep practising only the stations you like. The current blueprint guarantees coverage across the three main skill areas and also requires each running order to include risk or safety, cognitive or capacity, and physical health content. The blueprint also spans child and adolescent psychiatry, general psychiatry, psychiatry of intellectual disability, old age psychiatry, forensic psychiatry, medical psychotherapy, and addiction psychiatry.

For a six-station AI set, include:

one risk or safety-planning station
one cognitive or capacity-focused station
one physical health station
two management stations
one station from a specialty you tend to avoid

That stops your practice becoming narrow. It also feels more like a real exam day.

Use a simple answer framework, then speak like a doctor

In CASC, structure buys you thinking time. The candidate who sounds tidy usually is tidy.

History stations

Start broad, then narrow fast.

A reliable structure is:

opening agenda and rapport
presenting problem in the patient's own words
targeted symptom clarification
risk, safeguarding, or physical health where relevant
impact on function and supports
brief summary and next step

A good opening sounds like this: I need to understand what has changed, how risky things feel today, and what support you have around you.

Examination stations

Tell yourself what you are trying to prove or exclude. That keeps you from wandering.

Use this sequence:

orient the task
examine the relevant domain only
state key findings clearly
interpret what they mean
link them to immediate management

For a cognitive station, for example, your first sentence might be: I want to assess orientation, attention, and memory because the concern is an acute confusional state.

Management stations

These are where AI practice often helps most, because hesitation shows.

Use a priorities-first structure:

immediate safety and urgency
brief formulation or problem list
biological, psychological, and social plan
legal or capacity issues if relevant
explanation, collaboration, and follow-up

A crisp opening line helps: My priorities are immediate safety, engagement, and whether we need urgent medical or legal action.

Worked example

Say the AI gives you this station: a 67-year-old man with severe depression has stopped eating and tells you his family would be better off without him.

First run: just complete the task inside time.

Second run: improve the order. Open, clarify suicidal thinking, ask about plans and means, check psychotic symptoms and self-neglect, ask about supports, then summarise risk and immediate action.

Third run: give a 20-second closing summary out loud. That is often the difference between sounding busy and sounding appointable.

Debrief like an examiner

RCPsych's candidate guidance explains that each station has a construct defining what competent performance looks like, and candidate instructions may include explicit negatives telling you what not to do. In other words, extra talking is not automatically extra credit. If the task is to explain lithium monitoring, don't drift into a full bipolar history.

After every AI station, ask only five debrief questions:

Did I actually do the task asked?
Did I find the key risk, legal, or physical-health issue early enough?
Did I sound organised after the first 30 seconds?
Did I use plain English and respond to cues?
Did I close with a safe, realistic next step?

Then write one line only in your logbook:

keep
change
repeat

That tiny loop is better than pages of vague reflection.

Keep AI in its lane

AI is excellent for repetition, reframing, and speed. It is much less reliable for official rules, subtle human affect, and the standard of a borderline pass.

Use it safely:

de-identify every clinical scenario
don't memorise canned empathy phrases
ask it to challenge you with silence, anger, vagueness, or derailment
use it to shorten your answer, not inflate it
verify any exam-policy question on the RCPsych website, especially dates, places, regulations, and adjustments, because these do change over time

RCPsych also reminds candidates that day-to-day clinical work, educational supervision, observation of experienced clinicians, video, and feedback are valuable preparation. That fits a sensible hybrid model: AI for daily drilling, humans for final calibration.

Common mistakes

using AI as a passive tutor instead of a timed station simulator
asking for model answers before you've attempted the station yourself
practising only management stations because they feel easier to script
forgetting physical health, capacity, and safeguarding content
sounding empathic but failing to make a decision
answering the stem you hoped for rather than the task actually set
never replaying a weak station after feedback
trusting AI on current exam policy without checking RCPsych

Practice workflow

A realistic weekly workflow for a time-poor trainee looks like this:

two short solo AI sessions each week, 25 to 30 minutes each
in session one, run three morning-style stations with longer reading time
in session two, run four afternoon-style stations with rapid reading and tighter starts
do one human mock each week or fortnight with a colleague who will interrupt, challenge, and score you honestly
keep a running list of your recurring failures, such as weak openings, vague risk summaries, poor closure, or rambling formulations
on Viva Examiner, create a repeat queue of failed stations and rerun them 48 to 72 hours later

A useful rhythm is first attempt, feedback, immediate repeat, delayed repeat. That is how you turn insight into behaviour.

Summary

Use AI mock vivas for MRCPsych CASC as seven-minute station drills, not long revision chats.
Match the real format: single task, timed reading period, hard stop, short feedback.
Build your practice set around the blueprint, not around your favourite topics.
Debrief against task focus, structure, communication, safety, and closure.
Keep human mocks in the loop, because CASC is interactive and tone-sensitive.

How to Use AI Mock Vivas for MRCPsych CASC Prep

Why this works for CASC

Build AI mock vivas that match the real MRCPsych CASC station

A reusable prompt template

Build balanced sets from the blueprint

Use a simple answer framework, then speak like a doctor

History stations

Examination stations

Management stations

Worked example

Debrief like an examiner

Keep AI in its lane

Common mistakes

Practice workflow

Summary

References

Practice Psychiatry with a live AI examiner

Why this works for CASC

Build AI mock vivas that match the real MRCPsych CASC station

A reusable prompt template

Build balanced sets from the blueprint

Use a simple answer framework, then speak like a doctor

History stations

Examination stations

Management stations

Worked example

Debrief like an examiner

Keep AI in its lane

Common mistakes

Practice workflow

Summary

References

Related guides

Practice Psychiatry with a live AI examiner