Configure the interview
How the AI moderator works
Strella runs every interview with an AI moderator that talks to the participant by voice, follows your guide, and asks follow-ups in real time.
The AI moderator is the voice that conducts each interview. It opens with a built-in greeting, walks through every item in your guide, asks dynamic follow-ups where you've enabled them, and closes the interview with a built-in goodbye, all over voice.
The interview lifecycle
When a participant joins, the moderator follows a fixed lifecycle:
- Built-in greeting. The moderator opens with a fixed line: "Hi, I'm [moderator name], welcome to the interview." You can rename the moderator from the default "Strella," but you can't edit or remove the greeting itself. If you want scripted framing in your own words, add a Statement as the first item in your guide — the moderator reads it right after greeting the participant.
- Walks through your guide top to bottom. Each item plays out in its own way:
- Open questions and tasks: asked conversationally; Ask question verbatim forces exact wording.
- Statements: read aloud exactly as written, then move on. No response collected.
- Survey questions: appear on the participant's screen; the interview advances only when the participant answers. The moderator can't answer for them or skip past a survey question on its own.
- Media blocks: the media appears on screen, stays for the whole block, and the moderator works through any questions you attached.
- Asks follow-ups. Each Ask-group item has a Follow-up depth setting (Off / Low / Medium / High ≈ 0 / 1 / 2 / 3 follow-ups). Follow-up guidance tells the moderator what to probe for ("dig into pricing objections"). The moderator composes follow-up wording live, so you can't script the exact text.
- Built-in closing. After the last item, the moderator automatically thanks the participant and lets them know they can hang up. You can't customize or remove this. If you want a thank-you in your own words, add a Statement as the final item, and it plays just before the automatic goodbye.
Returning participants
If a participant leaves and comes back later, the moderator welcomes them back, the interview resumes where they left off, and any randomization or section-targeting assignments stay the same: they see the same variant and the same conditional sections they were on track to see.
How participants interact
- The moderator listens continuously while it's the participant's turn, with no push-to-talk by default, though you can enable it.
- During tasks, the moderator stays out of the way while the participant works, then checks in. With Adaptive pause detection on, it waits in silence rather than filling gaps.
- The moderator can't see what's on the participant's screen by default, so it asks them to confirm they can see media before moving on. Agent vision (beta) lets it periodically look at the participant's camera and shared screen while media is up.
The two behavior modes
| Mode | What it does | Use it when |
|---|---|---|
| Adaptive | Rephrases questions naturally, references prior answers, asks follow-ups | Most studies. Conversational. Default. |
| Strict | Reads each question verbatim from the guide, minimal improvisation | Quant-leaning studies where exact wording matters across people. |
Things to know
- The moderator doesn't pretend to be human. If a participant asks, it confirms it's an AI.
- It uses the participant's first name where appropriate, unless Name anonymization is on in Recording modes.
- It handles multiple people on a call. Interviews are usually one-on-one, but the moderator can detect and work with more than one voice in real time, which is useful for studies that need a parent and child, two teammates, etc. Not the standard use case, so test it before relying on it.
- It supports many languages. Pick one for the whole study, several for participant choice, or all of them. See Supported languages. It can also understand individual words or proper nouns interjected in another language (e.g. brand names) even when conducting the rest of the interview in the study's main language.
- It translates your guide live. Write your guide in any language; the moderator translates as it speaks based on the participant's chosen language. Custom translation instructions (in moderator settings) let you steer style, formality, and which terms to keep untranslated.
- No total length cap. Interviews aren't cut off for running long; the duration the participant sees on the join screen is an estimate. The only time limit is the Session timeout (30 sec to 15 min) for inactivity. See Recording modes.