Your feedback
How your answers are marked
Your spoken answer is turned into text. Then an AI model marks it against what a strong answer to that type of question includes. Pace, filler words and pauses come from the timings, and eye contact, framing and light are checked on your device. The feedback is practice guidance, not a prediction of any hiring decision.
From your voice to written feedback
- You answer out loud. In a video session your answer is recorded in your browser, and the recording stays on your device. In a live interview you can speak or type.
- The sound is turned into text. Only the audio of your answer is sent, never the video. It goes to a transcription service, which sends back the words and when each was said. Then the audio is deleted. Our privacy notice names the services.
- An AI model marks the words. It reads the question, your transcript and a few measures of your delivery, and writes feedback for that type of question: for a “Tell me about a time…” question, whether you covered the situation, your task, what you did yourself and the result. Each question in our bank also has its own notes on what a strong answer includes, which the model uses to sharpen its feedback.
- You get a report. Scores for structure, content and delivery, an assessor’s view of the evidence, what went well, what to change, a follow-up question an interviewer might ask next, and a short outline of a stronger answer built on your own example.
What’s measured, and where
| What | How | Where it’s worked out |
|---|---|---|
| What you said | The transcript, marked by an AI model | On our AI service, from the text only |
| Pace, filler words and pauses | From the transcript’s word timings | On our server, from the timings |
| Time used | From the length of your answer | On our server |
| Eye contact, framing and light | A face-tracking model that runs in your browser | On your device: only the totals leave it |
You can switch off delivery feedback, or eye-contact feedback, in your Account. With eye-contact feedback off, it isn’t measured at all.
Standard and detailed marking
Every answer is marked by our standard AI model. On Pro and the season pass (and for Deeplink students), answers are marked by a larger model when it’s available, which usually gives more specific feedback. When it’s busy, we mark with the standard model and say so on the report, and you can ask again later. Each account also has a daily share of our AI, far more than a day of normal practice needs: past it, later answers that day get simpler feedback worked out by our own rules, and the report says so. The models are open-weight models run on Cloudflare Workers AI, and neither the transcription service nor Cloudflare uses what we send to train AI models.
What the feedback never does
- It doesn’t predict a hiring decision. It’s practice guidance, and it isn’t linked to any employer. We never share your results with employers.
- It doesn’t judge who you are. We don’t score your accent, your appearance or your emotions, and the camera check doesn’t recognise who you are.
- It isn’t always right. AI can misread an answer, and a transcript can mishear a word. If feedback doesn’t match what you said, trust your own judgement, and check the transcript on your report.
Want to see it for yourself?
Start free, with full written feedback on every answer and no card needed. Or read our free answer guides to interview questions first.