Skip to content

How we use AI

Last updated: 30 October 2026. This statement explains which parts of TheatrePass use artificial intelligence, what that AI can and cannot do, how we check it, and how to tell us when it gets something wrong. It forms part of the terms of service and should be read with the privacy policy, which explains what happens to your data.

Where AI is used

  • Virtual Tutor. A voice model reads the question, turns your spoken answer into text and reads the model answers. A large language model then compares the text of your answer with the mark scheme for the station and writes your marks and feedback.
  • Timed mock interviews. A voice model plays the interviewer. It speaks the questions and the follow-ups we have written for the station, listens to your answers and moves the interview on. The same kind of language model marks the transcript afterwards and writes the report.
  • Your CV. If you upload your CV, a language model reads its text and lists the experience on it, and you check the list. When a portfolio answer is marked, the same kind of model compares that answer with your list, separately from the marking, and points out experience you left out and anything you said that your CV does not show.
  • Examination revision tutor. Where an examination product includes a tutor chat, a language model answers your questions using only our own revision notes, and tells you which note it is drawing on. It has no access to the wider internet.
  • Drafting our written material. Some material, including mark schemes and question explanations, is drafted with the help of AI and then reviewed by a person before it is used for marking or shown to you.

Apart from the uses listed above, nothing on the platform is driven by AI. Reading the question banks, frameworks and guides, tracking your progress, editing your CV list, peer practice and billing involve no AI at all, and no AI makes any decision about your account. If we add an AI feature we update this page before it goes live.

The providers behind it

Marking, feedback, the tutor chat and reading your CV run on large language models from an established AI provider. The interviewer's voice and the transcription of your speech run on a specialist voice provider. Each processes your data only to provide the service to us, under a written contract. Our privacy policy describes every kind of provider we use, and we will name the providers that hold your data if you ask us.

How marking is calibrated

  • Every station has a mark scheme written on the scoring domains the real panel publishes for that interview and the 0 to 5 scale those panels use. Each scheme describes what an answer at each band contains, so the model is comparing your answer with a written standard rather than forming its own view.
  • The model runs with fixed settings, so the same answer marks the same way each time. It is told the specialty and the station it is marking and receives only the scheme for that station.
  • We test two things and keep the results: stability (the same answer marked several times lands the same bands) and discrimination (a model answer scores higher than a partial answer, which scores higher than a thin one, and an empty attempt is not marked at all). We rerun these checks whenever we change the marking instructions.
  • A mark is a measure against our scheme. It is not a prediction of the score a real panel would give you.

What the AI cannot do

  • It can be wrong. Marks, feedback, model answers and explanations can contain errors, leave things out or sound more certain than the evidence allows. Treat a surprising statement the way you would treat a colleague's recollection: check it before you rely on it.
  • It can lag current practice. A model answer or explanation can fall behind the guidance you follow at work. Our material illustrates interview and examination technique, not a current clinical protocol.
  • It can mishear. Transcription is less accurate with background noise, some accents and unusual drug or procedure names, and a mistranscribed word can lower a mark. Read the transcript, and rerun the question if it is wrong.
  • The interviewer is not a person. It follows the questions we have written and may not respond to an unusual answer the way a real panel would.
  • It cannot use your CV to change a mark. The check against your CV runs apart from the marking and never sees your marks, so it cannot change a mark, a band or a model answer. It can misread a CV, so check what it says against your own record.
  • TheatrePass is an educational tool. It is not a medical device, not clinical decision support and not a source of clinical advice, and it must never be used to make decisions about real patients.

Decisions about you

The AI makes no decisions about you. Its output cannot affect a real interview, examination or application; no examiner, employer or recruitment body sees it; and you can rerun any question or ask us to look at a specific attempt. We never use an AI score as the basis for a decision about your account: every action under the abuse policy is decided by a person.

Your data and the AI

  • What is sent: the text of your answer or the transcript of it, with the station's mark scheme. Your name and email address are not sent with it. During a spoken session, audio passes to the voice provider in real time and is not stored by us.
  • Your CV: to make your list, the text of your CV is sent, including any personal details on it, such as your name. The file itself is not sent and is never stored. To check a portfolio answer, your list is sent with that answer, its question and the station's mark scheme. We do not use your CV or your list to train our own or anyone else's AI models.
  • Training: We do not use your recordings, transcripts or answers to train AI models, and we use our providers' settings that keep your data out of their training wherever they offer them.
  • People: nobody at TheatrePass reads your transcripts or listens to your recordings as a matter of routine. The privacy policy lists the four situations in which a member of staff can open a specific attempt.

If we find a problem

If an AI feature produces unsafe or persistently wrong output we can pause or withdraw it while we fix it. If that stops you using a paid feature for more than seven days in a row, the terms of service entitle you to an extension or a pro-rata refund of that feature.

Report a wrong or unsafe output

If a mark, a piece of feedback, an explanation or anything the interviewer says seems wrong, unsafe or offensive, email hello@theatrepass.co.uk with the subject line Unsafe AI output: attempt reference, replacing the attempt reference with the one shown on your report, and tell us what you saw. A person will reply within 5 working days. Depending on what we find we correct the mark scheme, rerun your mark at no cost, withdraw the content, or pause the feature. You can also report it with Feedback, in the menu under your name inside the app.