Video Conversation Survey Questions

Video Conversation runs a live, AI-moderated interview with an on-screen avatar: the respondent sees an interviewer's face, talks out loud, and the avatar listens and answers in real time, the way a video call would. Respondents can interrupt it, and it replies as soon as they finish a thought. It suits qualitative probes and interview-like prompts where a face-to-face exchange fits the study better than a voice-only or typed one — the video counterpart of Audio Conversation. It is a beta question type, available to beta testers on a Standard plan or above.

The avatar runs on Anam (anam.ai), using your own Anam account and API key. More providers are planned.

• Interviewer instructions tell the avatar what to ask about and how to probe, like Text Chat's system prompt. They can include piped answers and variables: type {{ to pick from earlier questions and survey variables. They are piped on Domandata's server and never reach the respondent's browser.
• An optional opening line is said word for word when the call starts. Leave it blank to let the interviewer open on its own.
• Avatar, voice, and model choose the face, the voice, and the language model behind the interviewer. Leave them blank for Anam's stock avatar (Cara), its voice, and Anam's recommended model, or paste ids from Anam Lab, where you can browse stock avatars and voices or build your own. Avatar model picks the rendering model (blank uses the avatar's own default), and Speech language optionally pins speech recognition to one language with an ISO 639 code such as en, es, or yue.
• Max length (5 minutes unless you change it; 30 seconds to 30 minutes) caps each respondent's call. The cap is carried inside the provider's session, so Anam itself ends the call at the limit — a tampered-with browser can't run it longer. Pauses don't reset it: a resumed call gets only the time that's left. To keep your Anam bill bounded even if someone tampers with their browser, each network (one IP address) can use up to ten full-length calls' worth of time per question per hour; a classroom or office sharing one connection shares that allowance.
• Min turns is how many times a respondent must reply before a required question lets them end the call and continue. If the call's time runs out first, they can continue anyway, since nothing more can be said in it.
• No transcript on screen. Respondents see and hear the avatar but never a running transcript, so the call feels like talking to someone. The transcript is still saved with each response and appears in exports.
• Label the avatar as AI (on by default) overlays Anam's "AI avatar" disclosure on the video.
• Require camera (off by default) makes respondents turn their camera on to start the call, shows them a small self-view, and pauses the call if the camera goes off. The avatar doesn't see the camera — Anam only hears the microphone — so this confirms the respondent is present rather than changing the conversation. It is checked in the respondent's browser.
• Save respondent video (off by default) records the respondent's camera and voice and saves the video with their response. It always turns on Require camera. With it off, the respondent's video is never recorded and never leaves their browser.
• A typed-reply alternative can be turned on per question — a low-key "type instead" control a respondent can use for any reply, without justification, if they can't speak it. The typed-reply box has its own Character limit, Disable pasting, and Track pasting settings. Instrument Review flags a required Video Conversation without it.

1. Step 1: Open Block Builder. Open the survey and select the block where the video interview belongs.

   Step 1: Open Block Builder

2. Step 2: Add a Video Conversation question. Choose Video Conversation from the question type selector.

   Step 2: Choose Video Conversation question type

3. Step 3: Add your Anam key. Choose a saved Video Avatar key, or paste an Anam API key. A pasted key is saved to Settings → Integrations → API Keys as soon as you leave the field — surveys never store raw keys.

4. Step 4: Write the prompts. Add the interviewer instructions, an optional opening line, and any respondent instructions shown above the video (for example, "Find a quiet place and allow microphone access").

5. Step 5: Choose the avatar. Leave the avatar, voice, and model blank for Anam's defaults, or paste ids from Anam Lab.

6. Step 6: Set length, turns, camera, privacy, and accessibility. In the question's options, set the max length and min turns, whether the avatar is labeled as AI, whether the camera is required and the respondent's video saved, whether Anam may keep its recording, and whether to allow a typed-reply alternative.

7. Step 7: Preview it. Use Preview to hold a real conversation before publishing — Preview calls run on your Anam account like live ones. Try brief, off-topic, and interrupted answers.

   Step 7: Preview it

The respondent presses Start conversation and allows microphone access; the avatar appears and starts talking. Unless the question requires the camera, only the microphone is used. With Require camera on, they're told before starting that the camera is needed — and, when the video is saved, that their video and voice are recorded — and they allow camera and microphone together. A countdown shows the time left. They can Pause (which hangs up) and Resume, which starts a new call that carries the conversation so far without greeting them again; if the connection drops, Resume works the same way. Once they have replied at least the minimum number of times, End conversation finishes the call; it also ends by itself at the max length, and they can then continue even if they replied fewer times than the minimum. If they move on to the next page without ending the call, it hangs up and everything said so far is kept.

Video Conversation needs a browser with WebRTC and microphone support — current Chrome, Edge, Firefox, and Safari all qualify — and a reasonably steady connection.

In Settings → Integrations → API Keys, use Add Key under Video Avatar to save an Anam API key (create one in Anam Lab). The key stays on Domandata's server. When a respondent starts a call, Domandata uses it to ask Anam for a short-lived session that can start exactly one call, expires within two minutes if unused, and is capped at the question's remaining time — only that session, never your key, reaches the respondent's browser. A key saved for a team can be used by the team's surveys only.

Domandata stores the conversation as text: each respondent reply and each interviewer reply, in order. No audio or video of the respondent is stored unless the question saves respondent video. A reply the respondent talked over is marked as interrupted, since its stored words may run past what they heard, and a typed reply is marked as typed. The response also records the total time the avatar was live and Anam's session id for each connection, so you can find a call in your Anam account.

Let Anam keep its recording is off by default: Domandata asks Anam not to retain a recording or transcript of the call. Turn it on only if your protocol covers a third party storing respondents' voices, and say so in your consent text; Instrument Review reminds you when it is on. If your Anam plan doesn't allow turning retention off, calls fail with a message saying so — turn the setting on, or ask Anam about your plan.

Saved respondent video. With Save respondent video on, each connection of the call (a paused and resumed call has several) is recorded from the moment the avatar appears and uploaded straight from the respondent's browser to private storage when that connection ends; the respondent waits briefly for the upload when they end the conversation. If they move on without ending the call, the recording is still kept with their answer and finishes uploading in the background (it is lost only if they close the tab before it finishes). Recordings count toward your workspace's storage — roughly 3 MB per minute. Exports get a Respondent Video column of links that open only for people signed in with access to the survey's responses (not translators), each through a link that expires after five minutes, so a shared spreadsheet never exposes a video by itself. A recording is deleted when its response is deleted, by you or by a response retention period; a recording from a respondent who never submitted is deleted about a week after their last activity — unless the survey submits unfinished responses for respondents (after some days, or at its close deadline), in which case it is kept until that response is submitted. Turning Save respondent video off later stops new recordings but keeps the ones already made, and exports keep the Respondent Video column (it is empty for questions that never saved video). Turn this on only if your protocol and consent cover storing identifiable video; Instrument Review reminds you when it is on.

Exports include a Transcript column (both sides of the conversation), Respondent Turns, Duration (s), Provider Session IDs, and, when video is saved, Respondent Video. The transcript comes from automatic speech recognition — treat it as unverified, as with any AI transcript, because noise, accents, or a quiet voice can produce transcription errors that look like bad answers. Transcripts are treated as open-ended text for PII redaction and Personal Info review.