AI SCA practice, with honest limits

How the AI SCA simulator works

The AI Patient Simulator lets you consult a simulated patient without a partner. You pick a case, by specialty and consultation type, and the AI takes on the patient's role from that case's script. It reveals the story only when you ask the right questions, keeps its ideas, concerns and expectations, and plants the cues a role-player would.

1. Choose a case and a patient manner. As well as standard, you can train the human side with an angry, tearful, health-anxious or evasive patient, a dominant third party, or low health literacy.
2. Prepare for 3 minutes with the records, as in the exam. Exam-day mode adds a connecting screen and a strict clock.
3. Consult for 12 minutes, by voice or text. Hands-free voice lets you just talk (it needs a microphone, works best in Chrome or Edge, and the patient can speak aloud). Or type throughout. The marking is the same either way.
4. End and get feedback. You get examiner-style feedback across the three domains, with specific points to change, plus the full transcript to reread.

When you are ready for stamina, the AI Mock Exam Circuit chains several AI patients back to back with one report card at the end. See how to run an SCA mock exam.

What the feedback covers

Feedback is organised by the three domains the RCGP marks: data gathering, clinical management and relating to others. It is written against the published RCGP domain descriptors. It points to moments in your transcript: a cue left unanswered, a plan that was announced rather than shared, a safety-net that was too vague to act on. It is most useful for habits that show up in words, such as open questions, exploring ideas and concerns, explaining clearly, checking understanding and specific safety-netting.

What AI marking cannot do

• It cannot see you. Eye contact, facial expression, pauses, warmth and the feel of rapport are a large part of how a real examiner experiences a consultation. The AI judges the words, not the room.
• It can be wrong. Like any language model, it can misread what you meant, give credit you did not earn, or miss something good you did. If feedback looks wrong, check it against the domain descriptors and the case's model consultation.
• The patient is fiction. The AI may add clinical details that are not in the case. Never carry its clinical content into real practice; use the case's knowledge section and UK guidance for that.
• It does not replace people. Practise with colleagues too, through group mocks or role-play packs, and get your trainer to watch real consultations. Real-consultation feedback can then help you find patterns in de-identified transcripts.

Privacy: what happens to what you say

From our AI safety and data page and privacy policy:

• What you type or say is sent as text through our server (Cloudflare) to the AI provider (Anthropic) to generate the patient's replies and the feedback. Anthropic does not use data sent through its API to train its models.
• Inputs and outputs of the simulator are not kept on our side beyond the request. Anything you choose to save stays in your own browser, or in your account if you sync it.
• No tool on the site records or uploads audio. Voice mode uses your browser's own speech recognition to turn speech into text, and your browser provider's terms apply to that step.
• The cases are fictional, so you never need real patient details. Never enter anything that could identify a real patient.

Frequently asked questions

More SCA guides