Live Conversation: New experience - AI summaries (beta)

On this page:

• What it is
• How summaries are generated
• What's in a session summary
• What's in a study summary
• How the summaries stay accurate
• What you can do with a summary
• Status and troubleshooting
• FAQ

What it is

Live Conversation AI Summaries automatically turn a session transcript into a structured, readable summary so you can understand what happened without watching the full recording. Summaries are generated at two levels:

• Session summary — one summary per interview/recording, covering what happened in that single session.
• Study summary — an aggregated summary across all sessions in a study, surfacing cross-session insights.

Summaries are generated automatically — there is no button to click and nothing to schedule.

How summaries are generated

When they run

There is no on-demand "generate now" or scheduled option — generation is fully event-driven.

What the AI reads

The summary is built from the session transcript only — not the video or audio. The transcript is organized into speaker turns (Moderator vs. Participant), and the AI references specific lines to build the summary, so highlights and sections link back to real moments in the conversation with accurate timestamps.

What's in a session summary

A session summary contains:

• Executive summary — a short (under ~150 words) overview of the session.
• Session goal — what the session set out to learn.
• Highlights — typically 3–7 key moments, each with its own short summary and a start/end timestamp so you can jump straight to that point in the recording.
• Sections — the session broken into labeled parts (e.g., introduction, instructions, background, task, wrap-up), each with a title, a summary, and timestamps.
• Section questions & answers — for task-focused sections, a breakdown of the questions explored and a summary of how the participant responded.

What's in a study summary

A study summary rolls up all sessions and contains:

• Combined executive summary — an overview across the whole study.

• Combined study goal — the study's overall objective.

• Study insights — cross-session findings, each tagged with one of the following categories and linked back to the source session highlights that support it:

Need · Pain point · Expectation · Process · Opportunity · Ease of use · Language · Trust · Satisfaction · Suggestion · Confusion

How the summaries stay accurate

Several safeguards are built in so summaries reflect what actually happened:

• Grounded in the transcript — the AI is instructed to use only content present in the transcript and not to invent quotes.
• Hallucination filtering — any highlight, section, or question that references a line not found in the transcript is automatically dropped before the summary is saved.
• Focused on what's being evaluated — the summary concentrates on the participant's interaction with the product/design/concept being tested, and de-emphasizes logistics, setup, and background chatter.
• Privacy-conscious phrasing — summaries avoid personal names and pronouns.

What you can do with a summary

Summaries are read-only. You view the summary and click any highlight or section to jump to that moment in the recording.

If you have feedback on study summaries, you can leave thumbs-up / thumbs-down feedback with an optional comment that we use to improve our models that generate summaries.

Status and troubleshooting

Each summary reports a status so the interface can show progress:

• Processing — the summary is being generated.
• Ready — the summary is complete and viewable.
• Error — generation failed.

Common reasons a summary may not appear:

• Empty or near-empty transcript — if a session has no usable dialogue, no summary is generated.
• Transient processing errors — occasional throttling or timeouts can cause a failure; failed jobs are captured and can be reprocessed by the team.
• Content flagged by safety guardrails — content that trips built-in safety checks may block generation.

If a summary is stuck in Error, contact support so the session can be reprocessed.

FAQ

Do I have to turn this on? No. Summaries are generated automatically for eligible studies. Gen-AI opt out customer, however, will not see summaries.

How long does it take? Summaries typically appear shortly after a session's transcript finishes processing (session level) or after the study completes (study level).

Can I trust the highlights' timestamps? Yes — highlights and sections are anchored to specific transcript lines, and timestamps are mapped from the actual transcript rather than estimated by the AI.

Why does my summary look thin? Summaries are only as rich as the transcript. Short sessions, sessions with little participant dialogue, or heavy setup/logistics with little product interaction will produce shorter summaries.

Related content