About this tool
A small tool for practicing interview answers out loud and a plain account of how its feedback is produced and where it stops being reliable.
Who runs it
ZoeVera Interview Prep is run by ZoeVera as a small independent service. There is no team of recruiters behind it and no private hiring data. It is software that transcribes a practice answer and gives feedback on it.
The company profile is on LinkedIn. We do not name an individual on the site. A company profile is something you can check. For questions or corrections write to support@zoevera.com.
How a report is made
- Transcription. Your voice is sent to OpenAI Whisper, which returns the words with their timing. The audio is not stored afterwards. The video never leaves your device.
- Content feedback. The transcript, the question and the role go to an Anthropic language model, which scores the answer and writes the coaching points against a rubric.
- Delivery numbers. These are calculated from the word timings, not by the model. Pace is words per minute, with 100 to 160 treated as in range. A long pause is 2 seconds or more between two words. Filler words are counted from a fixed list and the words that are also ordinary words (“like”, “so”, “right”) count only when the transcript sets them off with a comma.
The model is told how to read a score:
- 90 to 100
- Would pass a Bar Raiser
- 70 to 89
- Solid but probing would expose gaps
- 50 to 69
- Hand-waving detected
- Below 50
- Significant structural or depth failure
What the score does not claim
- It is not a prediction of whether you will get an interview or a job. It has not been checked against hiring outcomes.
- It is not how any employer or hiring system would rate you. No employer shares its criteria with us.
- It is a language model's judgment of your words and two runs on the same answer can differ a little. Treat the quoted, specific points as suggestions to test on your next take.
- The filler count depends on the punctuation the transcription adds, so some fillers will be missed or miscounted.
- Recording and paid features are described on the pricing page. Trial answers carry a score and three coaching points and a plain coaching prompt, so trial and paid scores are not comparable with each other.
Where the studies come from
The two studies use public sources only: employers' public job postings and policy pages for the AI in job interviews study and the U.S. government's O*NET database for the role speaking index. Each publishes its data as a CSV under a CC BY 4.0 license, states its method and limits on the page and has a way to report an error. Nothing from anyone's practice sessions is used in them.
See all research, the privacy policy for how practice data is handled and the practice guides.