Quickstart

Run your first in-product AI interview from the dashboard. You do not need an API key to start: platform inference uses your prepaid credits, including the small allowance on a new account.

If you want an agent to perform setup, use the focused MCP setup or CLI setup instead.

1. Create a project

Sign in to the dashboard, then create a project for the product or product area you want to research.

You can optionally add your own OpenAI API key in Project Settings. Keys are encrypted at rest and scoped to the project. When you use your own key, inference is billed directly to your OpenAI account instead of UserTold credits.

2. Create a study

Open Studies and create a study around one research question. Add specific goals and a short sequence of interview segments:

  • talk for conversation and follow-up questions
  • speak for scripted instructions and transitions
  • observe for silent capture while the participant completes a task

Start small and test the interview yourself. The Study Design Guide owns the script schema, patterns, and complete examples.

Open Intakes, create qualification questions, and link the intake to your study. Activate both the intake and study so qualified participants can enter the interview.

4. Embed the widget

Open the intake detail page and copy its Embed Code snippet into your product:

<script
  async
  src="https://usertold.ai/v1/widget.js"
  data-project-key="ut_pub_YOUR_KEY"
  data-screener-id="your-intake-handle"
></script>

Use https://usertold.ai for the widget script. The data-screener-id attribute takes the intake handle; the embed API retains the older screener name for compatibility.

The loader waits for the host page to finish loading before fetching the widget. When a visitor opens it, they complete the intake and, if qualified, begin the linked study without leaving your product.

5. Test one complete interview

Visit the embedded product as a participant and confirm:

  1. The widget opens and shows the intake.
  2. A qualifying response enters the intended study.
  3. The interview reaches its final segment and completes.
  4. The interview appears in Interviews with its recording, transcript, or processing state.
  5. Evidence appears after processing, or the result clearly says that no evidence was found.

If the widget or interview does not behave as expected, use Widget Integration and Study Runtime to diagnose that path.

6. Review the result

Open the interview and inspect the source recording, transcript, timestamps, and page context. Review packets group related evidence for triage, but they are not automatic delivery instructions.

Before pushing work to Linear or GitHub:

  1. Verify the linked source context.
  2. Confirm the grouping and current product area.
  3. Promote only the packet that represents work worth shipping.

When a linked Linear issue completes, UserTold can resolve the current evidence and continue watching future interviews for possible recurrence.

Continue