Agent Takeover (Human Moderation)
Agent Takeover
Overview
Agent takeover allows a human moderator to step into any active WhatsApp conversation and communicate directly with a participant. The AI pauses, the moderator takes control, and the participant sees no difference — messages continue in the same chat, from the same number.
This gives you the flexibility to combine AI scale with human depth — let the AI handle the bulk of conversations, and intervene personally when it matters most.
How It Works
Taking Over a Conversation
Open the Interview Transcripts tab.
Select the participant whose conversation you want to join.
Click Agent Takeover to activate.
The AI moderator immediately pauses.
You type and send messages directly to the participant.
The participant receives your messages in the same WhatsApp chat — there is no visible transition.
When you're finished, hand back to the AI, which resumes from where it left off.
What the Participant Experiences
From the participant's perspective, nothing changes. They continue receiving messages in the same conversation, from the same WhatsApp number. They have no way of knowing whether they're speaking to the AI or a human moderator — the transition is seamless in both directions.
When to Use Agent Takeover
During an Active Study
A participant shares something particularly interesting — Probe deeper with human intuition and follow-up questions the AI might not think to ask.
The AI isn't probing in the direction you want — Steer the conversation toward a specific topic or angle.
A participant is confused — Clarify instructions, question meaning, or study expectations with a human touch.
You want to ask a very specific unscripted question — Insert a one-off question that isn't part of the configured interview.
A participant mentions something sensitive — Handle with human empathy and judgment rather than relying on AI.
Your editorial or research team wants a live interview — Conduct a full human-moderated WhatsApp interview with a selected participant.
After Reviewing Transcripts
A common and powerful workflow is to let the AI conduct interviews with all participants, review transcripts as they come in, identify the 5–10 most interesting participants, then use agent takeover to follow up with those individuals for deeper exploration.
This is the ideal hybrid approach: AI handles scale, humans handle depth. You get the breadth of 50+ AI interviews with the richness of hand-picked human follow-ups.
What You Can Do in Agent Takeover
Once you've taken over a conversation, you can send text messages directly to the participant, ask follow-up questions probing on anything they've said so far, request media (photos, voice notes, videos, or location), provide clarification on questions, study instructions, or next steps, and have a full extended back-and-forth interview if needed.
Reading the Full Context
Before you start messaging, you can see the entire conversation history — all survey responses, AI interview exchanges, media files, and transcriptions. This means you can reference specific things the participant said earlier, creating a natural and informed follow-up.
For example: "Earlier you mentioned that you stopped using the app because it kept crashing. Can you tell me more about that? How often was it happening and what were you trying to do when it crashed?"
The participant experiences this as a seamless continuation of the conversation, not a cold outreach.
Handing Back to the AI
When you've finished your human moderation, click to end agent takeover. The AI moderator resumes the conversation from where it left off. The AI has full context of both the original conversation and your moderator messages. The participant continues as normal with no visible transition.
Transcript Recording
Everything that happens during agent takeover is captured in the transcript. All moderator messages are recorded alongside AI messages and participant responses. The full conversation — AI and human — appears in the transcript view and in data exports. Moderator messages are identifiable in the data so you can distinguish between AI-generated and human-generated questions during analysis.
Best Practices
Preparation
Read the full transcript first — Understand what the participant has already said before jumping in.
Have a clear purpose — Know what you want to explore before you start typing.
Match the tone — The participant has been speaking to an AI configured with a specific style. Maintain a similar register so the transition feels natural.
During the Conversation
Be responsive — Once you've taken over, the participant expects human-speed responses. Don't leave them waiting.
Keep it conversational — The participant has been in a chat environment. Formal interview language will feel jarring.
Reference earlier responses — This demonstrates that you've been paying attention and makes the participant feel heard.
Know when to hand back — Once you've got what you need, hand back to the AI cleanly rather than letting the conversation trail off.
After the Conversation
Review the combined transcript — Check that the transition was smooth and the data is coherent.
Tag or flag the participant — Note which participants received human follow-up for analysis purposes.
Share key quotes — Agent takeover conversations often produce the most powerful verbatims in a study.
Use Cases Beyond Research
Customer support — If a participant raises a complaint or issue during a study, a moderator can address it directly.
Participant management — Clarify incentive details, answer logistical questions, or resolve technical issues.
Editorial interviews — Your editorial or content team can conduct full WhatsApp interviews with selected participants, using the AI study as a warm-up and screening mechanism.
Known Limitations
24-hour response window — WhatsApp's Business API requires that you respond within 24 hours of the participant's last message. If more than 24 hours have passed, you'll need to send a broadcast template message to re-open the conversation window before using agent takeover.
One moderator at a time — Only one moderator can control a conversation at any given time.
Participant may respond while you're typing — Be prepared for overlapping messages, especially if the participant is a fast responder.
AI context after handback — The AI resumes with full context, but very long moderator conversations may push earlier context further back in the AI's memory. Keep moderator interventions focused.
Last updated