- The interviewer voices, role-play characters and feedback are AI (Anthropic’s Claude models), not a person. No doctor, tutor or medical school checks your feedback.
- Practice stations were drafted by AI and checked by separate AI passes for errors, bias and anything resembling real interview questions. AI can mishear you, miss things you said, or simply be wrong.
- Scores are practice indicators. They are not predictions of any exam, interview or admissions result.
- AI can be wrong. Every question and result has a “Report a problem” button. We look at every report, usually within 5 working days.
- Your answers are not used to train AI models.
Where you meet AI in Mockseat
| What | Made by | Checked by a person? |
|---|---|---|
| The interviewer and role-play characters | AI, replying live to what you say | No |
| The interviewer’s voice | A synthetic, computer-generated voice. It is not a recording of a real person | No |
| Turning your speech into text | AI speech recognition. It can mishear names, accents and quiet speech | No |
| Feedback and scores | AI, marking the text of your answer against the station’s criteria | No, unless you ask us to look |
| Practice stations | AI, written in advance and checked by separate AI passes | No. A person looks at every problem report |
| Support messages | AI reads each incoming message and sorts it. If you are signed in and ask a common question (how something works, how to get a refund, how to delete your account), AI may write and send the first reply, following our written policy | Everything else is answered by a person, including every complaint, anything the AI is unsure about and anything about safety. Reply to any email from us to reach a person |
Which AI companies we use and what they receive
| Company | Used for | What we send | Training on your data |
|---|---|---|---|
| Anthropic (United States) | The interviewer’s replies, feedback and scores; sorting and answering support messages | The text of what you said or typed in the station, or of your support message. Not your email address, name or payment details, unless you type them yourself | None. Under our commercial terms, Anthropic does not use what we send to train its models |
| Deepgram (United States) | Turning your speech into text; making the interviewer’s synthetic voice | Your audio while you speak, live, and the text of each interviewer line. We never save your audio | None requested. Every request we send opts out of Deepgram’s model improvement program |
We use current Claude models from Anthropic: a faster model for live replies and a larger one where more care is needed, such as marking. Speech uses Deepgram’s current speech-to-text and text-to-speech models. We change model versions from time to time, so the same answer may not get exactly the same reply twice. We test a new version on our own checks before we switch. How long each company keeps data is in our privacy notice.
Before your first spoken station, your browser or phone asks for permission to use the microphone. You can choose typing instead: then no audio leaves your device.
What people do
People on our team read problem reports, support messages that need a person, refund requests and the note made when a station stops for a safety reason. Nobody on our team reads your practice answers as a matter of routine, and nobody checks AI feedback or scores before you see them.
If you think your feedback or score is wrong
Tap Report a problem on the result, or email hello+mockseat@medservicesuk.com. A person looks at it, usually within 5 working days, and tells you what they found. If the content was wrong we fix or remove it. We cannot re-mark a station by hand like a tutor would, but if the AI misheard you or missed something you said we will say so, and where that spoiled a station we will consider giving it back or a goodwill refund. Scores are for your own practice only. We do not send them to any school, university, employer, regulator or exam body.
Changes
- Version 2.0, 11 November 2026. Added where AI is used, the companies we use and what each receives, synthetic voices, what people do, and how to question a score.
- Version 1.0, 10 November 2026. First version.
Med Services Ltd, a company registered in England and Wales. Company number: 14915054.
Registered office: 52a Cherwell Street, Oxford OX4 1BG.
ICO registration: ZC278740.
Contact us, including about your data, at hello+mockseat@medservicesuk.com or through the Help form.