The AI interviewer, live and in writing
How the AI interviewer runs the live call and the text interview: the same core questions, timekeeping in code, fixed rules, and no video.
The AI interviewer runs Part 2 of a founder interview: the questions after the pitch. It works the same way on the live call and in the text interview, from the template version the invitation was sent with. See Interview templates and questions.
What the interviewer does
| Live call | Text interview | |
|---|---|---|
| Part 1, the pitch | Listens silently while the founder pitches with screen and camera | The founder writes a pitch summary, about 120 words per pitch minute |
| Core questions | Spoken in your template's words, in your order | Written in your template's words, in your order |
| Follow-ups | At most one per question, up to the template's limit | The same |
| Time | Closing line one minute before the cap, hard stop at the cap | No timer; the same time plan decides how many questions are asked |
| What it receives | The founder's microphone only | The founder's messages |
The interviewer never chooses or rewrites core questions. After each answer it may ask one follow-up tied to a criterion or an open claim, usually for a number, an example or a source. A follow-up is only possible while the most important question left still fits in the time. When the plan has no more questions, it closes with: "Thank you. That's all the questions we have. The team at your fund will read your answers and let you know the next step."
Fixed rules
These rules are part of every interview and can't be changed in a template:
- Neutral and polite. No flattery, jokes, arguments, enthusiasm or doubt about the company.
- One question at a time, short.
- No advice, feedback or opinions about the company, the market or the answers.
- No comments on how the founder speaks, writes or looks.
- No questions about age, gender, race or ethnicity, nationality or immigration status, religion, disability or health, pregnancy or family plans, marital status, sexual orientation, or any other protected characteristic.
- Never reveals or discusses scores, ratings or how answers are judged.
A follow-up that would break a rule is never sent: the interviewer asks the next core question instead. If the AI is unavailable, the interview carries on with the core questions only.
Timekeeping
The platform's clock decides the time, not the AI. Each core question is given about 90 seconds and each follow-up about 60 seconds. On the call, the closing line comes one minute before your cap and the session ends at the cap, even mid-answer. In the text interview, the same time plan decides how many questions fit, so a written interview gets the same questions a spoken one would.
What you see
On the deal's Interview tab, a written interview is labelled Text interview. Its transcript is the founder's own words, so it goes straight to the assessment. Live calls are transcribed again from the recording after the call. Founders never see scores or your team's ratings.
Frequently asked questions
Does the AI see the founder's face or screen?
No. The interviewer subscribes to the founder's microphone only. The camera and screen recordings are for your team's review.
Can a founder change the interviewer's behaviour by what they say?
No. The founder's words are treated as data. Instructions inside an answer, such as asking the interviewer to skip questions or reveal anything, are ignored.
Is the text interview a weaker signal?
It is assessed on the same criteria from what the founder says. Delivery, writing style, spelling and grammar are never assessed in either format.
What is logged about the AI?
For each model and speech call: the model, prompt version, tokens or audio seconds, cost and latency. Transcripts and answers are never written to the logs.
Something unclear or missing? Tell us