Skip to main content
Calls are completed interview sessions. Each call record carries the transcript, recording URL, AI analysis, and quality signals. 6 tools.
ToolDescription
list_callsList interview calls, filterable by study/participant/status/quality
get_callFetch one call (includes transcript, messages, recording_url, analysis)
update_callUpdate mutable fields (e.g. is_visible, success_evaluation)
delete_callSoft-delete a call
get_call_usage_statsAggregate call counts/durations over a date range
generate_reportTrigger study-level report generation

​
Response shaping

Response shaping. get_call and list_calls return lean summaries by default:
  • view — "summary" (default) or "full" (raw record).
  • include — array of heavy fields to expand, e.g. ["transcript"], ["messages"], ["screener_questions"].
  • fields — exact allowlist of top-level fields (id always returned); overrides view/include.
By default large screener option lists are truncated and transcripts are returned as a short excerpt. Pass include or view: "full" to fetch the omitted data. Default page_size is 5 for list_calls (max 100). Using include on a list multiplies payload by page_size — prefer get_call for full heavy fields on a single record.

​
How User Intuition handles this

Every interview User Intuition runs is stored as a searchable, analyzable call record. generate_report triggers cross-study synthesis — the same engine powering the in-app dashboard reports. See /platform/agentic-research/ for the full agent-driven research workflow.

​
AI participant detection

Each scored call is checked for whether the participant (not the AI moderator) was likely an AI or bot rather than a genuine human — for example, someone relaying the questions through a chatbot and reading the answers back. The result is stored on the call’s success_evaluation_details JSON under the ai_participant_check key: 
{
  "ai_participant_check": {
    "ai_participant_suspected": true,
    "confidence": 0.9,
    "signals": ["AI self-disclosure", "essay-like enumerated answers"],
    "reasoning": "Short explanation of why the call was flagged."
  }
}
This check is flag-only: it never changes the call’s success_evaluation score. A flagged call is surfaced to admins for review and can be moderated (excluded from results via is_visible, have its pending panel reward voided, or be re-scored). confidence is a 0–1 suspicion score and is 0.0 when not suspected; the detector is biased toward false negatives so genuine, articulate participants are not flagged.