Define the study
Set the research goal, qualifying intake, and script segments so the interviewer knows when to talk, observe, and debrief.
How the feedback loop works
From an agent-run interview to evidence-backed work.
Your agent starts with a research question. UserTold conducts the interview inside your product, preserves what the user said and did, and returns source-linked evidence for review. Only verified evidence moves into GitHub or Linear.
Real usersSource-linked evidenceReviewed work
Interview to delivery
Interview evidence, review status, and linked work
Interview signal
"I expected billing to live under account settings, not workspace settings."
/settings/billing00:42
Evidence packet
Billing settings findability issue, supported by source quotes and observed task behavior.
TypeStruggling moment
DecisionNeeds review
After shipping
Linked evidence resolves when Linear completes, then matching future evidence can resurface.
Five steps
From study setup through post-release review
Each stage retains its source context. Evidence remains linked to the interview, the delivery decision, and the completed Linear issue.
Run interviews in-product
Participants enter through your product. UserTold captures voice, transcript, screen context, and task behavior in the same session.
Extract evidence
Raw sessions become structured struggling moments, desired outcomes, and workarounds with quotes, confidence, and source context.
Review packets
Related evidence clusters into review packets that humans or agents verify before anything becomes delivery work.
Resolve and watch
When linked Linear work completes, current evidence resolves and future interviews are watched for possible recurrence.
Evidence before output
Review what happened before deciding what to ship
UserTold.ai treats evidence packets as triage bundles, not automatic orders. A project-aware human or agent verifies the source evidence before routing work.
Review packet
"I tried this flow three times and still cannot find where to change billing settings."
Each packet keeps the source quote, transcript or playback context, project goal, confidence, and delivery status visible for review.
Source quotePlayback contextConfidenceWork link
Source context
Quotes, transcript moments, page paths, and playback context stay attached so interpretation remains reviewable.
Conservative grouping
Packets summarize related evidence without pretending every cluster is automatically a delivery order.
Delivery handoff
Verified work can be pushed to GitHub or Linear with the evidence that explains why the issue exists.
Three surfaces
Use the dashboard, widget, CLI, or MCP
The public website, dashboard, embeddable interviewer, and agent-facing tools all point at the same research-to-delivery workflow.
Dashboard
Review studies, interviews, evidence, packets, and routed work from the researcher workspace.
- Study setup
- Evidence review
- Packet triage
Widget
Embed interviews where the product behavior happens, with planned conversation and silent observation segments.
- Voice interviews
- Screen context
- Task debriefs
Agent tools
Use CLI, MCP, and REST surfaces when agents need JSON-first context for research and delivery workflows.
- CLI output
- MCP resources
- Tracker routing
Need implementation details? The docs cover study setup, the widget embed, CLI commands, and MCP integration.
Start with a real conversation
Give your agent evidence from the people using your product.
Define a study, capture a real session, inspect the evidence packet, and route only the work that survives review.