How Often to Run AI Mock Vivas Before Exam Day

The best AI mock viva frequency for most trainees is little and often: two to three short mocks a week when you are more than a month out, rising to four or five focused sessions a week in the last fortnight, then easing off in the final 48 to 72 hours. That pattern gives you repeated retrieval under pressure without tipping into the dead zone of tired, robotic answers.

There is no single official college rule on how often to run AI mock vivas. What the current official exam examples do show is that spoken clinical assessments are timed, structured, and multi-domain: MRCP(UK) Part 2 Clinical Examination (PACES23), the MRCGP Simulated Consultation Assessment, and the MRCOG Part 3 Clinical Assessment all sample performance in short stations or consultations rather than in one long conversation.

Why frequency matters

In these exams, knowledge is only part of the mark. You are also being judged on how quickly you frame the problem, whether you prioritise safety, how clearly you explain your reasoning, and whether you stay organised when interrupted. RCGP's SCA toolkit breaks performance into consultation domains plus global skills such as structure and time management, while MRCOG Part 3 assesses patient safety, communication, information gathering and applied clinical knowledge across its circuit.

That is why one giant mock on a Sunday is usually less useful than shorter, repeated rehearsals through the week. Official prep resources also lean heavily on targeted review: the SCA toolkit is designed for use with a trainer or educational supervisor and asks candidates to identify weak areas before practising again, while RCOG offers circuit practice, example videos and marking-familiarisation resources. AI fits neatly in that gap between bigger human-led sessions.

AI mock viva frequency: a simple schedule

Use this as a starting template, not a law. If your main problem is freezing on the first question, you need more reps; if your main problem is nuance or factual accuracy, you need better feedback rather than more volume. The logic is spaced practice plus rapid correction, not endless case accumulation.

Six to eight weeks out

Run 2 to 3 AI mock vivas a week.
Keep most sessions short: one stem, one opening answer, one follow-up sequence.
Rotate case types: acute prioritisation, communication, uncertainty, management planning.
Finish by rewriting or re-speaking your weakest 60 to 90 seconds.

This phase is about building fluency. You want enough repetition that your opening structure becomes automatic, but not so much that every answer starts to sound memorised.

Two to four weeks out

Increase to 3 to 4 AI sessions a week.
Make at least 2 of them fully timed to match your real station length.
Add interruptions: ask the AI to challenge your differential, change the observations, or play a worried relative.
Keep 1 human mock each week if possible.

This is the sweet spot for most trainees. Timed repetition exposes pacing problems early enough to fix them, while human feedback stops the AI from becoming your only standard setter.

Final 7 days

Aim for 4 to 5 focused AI mocks across the week, not 4 to 5 every day.
Prefer short sets of 1 to 3 cases rather than marathon circuits.
Rehearse your first minute, your summary, and your safety-netting language.
Book 1 or 2 human mocks if you can, especially if tone, empathy or pacing are still shaky.

In the last week, the goal is sharpness. You are trimming hesitation and cleaning up recurring errors, not cramming every possible topic.

Final 48 to 72 hours

Do 1 light confidence-preserving session a day, or skip AI entirely if it winds you up.
Use familiar stem types, not wild-card cases designed to scare you.
Stop any session that turns into unfocused doom-scrolling through prompts.
Sleep, logistics and calm matter more than squeezing in one more bad mock.

That taper is deliberate. A tired final rehearsal can dent confidence without teaching you much, especially when there is no time left to correct the habit it reveals.

This build, peak and taper plan is an editorial recommendation, not an awarding-body policy. The colleges set the exam format; they do not publish a universal AI mock viva prescription.

Make each AI mock viva count

One good AI mock has five parts: a realistic stem, a timed answer, at least two interruptions, a short debrief, and an immediate rerun of the weak bit. The rerun matters. Repetition without correction is just noise; repetition with feedback is what starts to change performance.

Ask for a stem at the right level for your exam.
Answer out loud, not in your head.
Get the AI to interrupt you once for clarification and once for escalation.
Mark yourself under four headings: structure, safety, prioritisation, communication.
Rerun only the weak segment, then move on.

For many stems, a reliable opening sounds like this: state the setting and main risk, give your working diagnosis or task aim, say what you will do next, then flag escalation or safety-netting. That simple shape works in an acute station, a consultation, or a structured discussion.

If you want a simple debrief grid, borrow from what official exams actually reward. The SCA toolkit uses consultation domains and global skills, and MRCOG Part 3 marks across patient safety, communication, information gathering and applied knowledge. A debrief that mirrors those domains is more useful than a vague verdict of good or bad.

Worked examples

A 67-year-old with central chest pain in acute medicine is a good AI stem for first-minute structure: immediate safety checks, high-risk differentials, early investigations, then disposition. A GP telephone consultation for persistent cough tests data gathering, explanation and safety-netting. An obstetric triage stem on reduced fetal movements is useful for risk communication, prioritisation and senior escalation.

The point is not to collect clever diagnoses. It is to make your answer shape dependable under pressure.

When AI should give way to human mocks

AI is strongest for repetition, convenience and low-stakes rehearsal. Human mocks become more important when your marks are being lost on tone, realism, interruption handling or whether your answer actually feels safe to an examiner. That matters because real assessments involve another person judging your pace and response: PACES23 is assessed by two independent examiners at each station, the SCA uses trained role players, and MRCOG Part 3 includes simulated patient or colleague tasks and lay examiners on some stations.

Move toward human-heavy practice if you sound polished to the AI but hesitant in front of a person.
Switch sooner if colleagues keep spotting unsafe omissions the AI misses.
Use human mocks for empathy, shared decision-making, conflict, and explaining risk.
Use AI for volume between those sessions.

A practical split is about 3 AI mocks to 1 human mock six weeks out, drifting towards 1 to 1 in the last fortnight. That ratio is a coaching rule of thumb, not a college rule, but it matches the broader pattern in official prep resources: repeated self-review is useful, and external calibration remains essential.

Use AI safely

Do not paste identifiable patient details into a public AI tool. GMC confidentiality guidance says doctors have ethical and legal duties to protect personal information from improper disclosure, and reflective material should be anonymised as much as possible because people can be identified from combinations of details, not just from a name.

Treat AI feedback as a draft, not as a source of truth. If a management step sounds odd, overconfident or out of date, check it against current guidance or ask a supervisor. That caution fits the way official exam resources frame performance: the SCA toolkit explicitly contrasts safe, evidence-based, up-to-date plans with poorly evidenced or outdated ones.

Common mistakes

Most wasted AI practice falls into a few predictable traps.

Doing too many cases in one sitting and learning none of them properly.
Letting the AI ask essay-style questions that do not resemble your real station.
Practising only your favourite specialty and avoiding weak areas.
Never replaying the poor opening answer.
Memorising polished phrases instead of clear priorities.
Using AI instead of trainer, peer or consultant feedback.

If your AI work is not exposing weak areas, correcting them, and then repeating them, it is probably too passive.

Practice workflow

If you are busy on call, use a rhythm you can actually keep. Three short sessions done well will beat a fantasy schedule you abandon after four days. Spaced rehearsal only works if it is repeatable.

Monday: 1 short AI case on opening structure.
Tuesday: 1 AI case on follow-up questions and interruption handling.
Thursday: 1 human mock or recorded self-mock with debrief.
Saturday: 2-case timed set matching your exam.
After each session: log 1 safety miss, 1 language habit, 1 knowledge gap, and 1 fix for next time.

That last step is the whole game. Debrief, choose one fix, and test it again in the next case. The trainees who improve fastest are usually not the ones who do the most stems; they are the ones who close the loop most reliably.

Summary

Start with 2 to 3 AI mocks a week, rise to 3 to 5 across the final fortnight, then taper.
Keep sessions short, timed and feedback-rich.
Rerun the weak segment straight away.
Use human mocks for realism, nuance and safety calibration.
Never use identifiable patient details, and always fact-check management.

How Often to Run AI Mock Vivas Before Exam Day

Why frequency matters

AI mock viva frequency: a simple schedule

Six to eight weeks out

Two to four weeks out

Final 7 days

Final 48 to 72 hours

Make each AI mock viva count

Worked examples

When AI should give way to human mocks

Use AI safely

Common mistakes

Practice workflow

Summary

References

Ready to rehearse under exam pressure?

Why frequency matters

AI mock viva frequency: a simple schedule

Six to eight weeks out

Two to four weeks out

Final 7 days

Final 48 to 72 hours

Make each AI mock viva count

Worked examples

When AI should give way to human mocks

Use AI safely

Common mistakes

Practice workflow

Summary

References

Related guides

Ready to rehearse under exam pressure?