Interview summaries are useful, but product teams also need evidence behind a claim. The first visual scanner version can ground answers in sampled frames from a demonstration or shared screen; it does not perform speech-only interview analysis.
Separate visual extraction from transcript analysis
Use an extractor for facts shown on screen, a classifier for a bounded visible state, a summarizer for the demonstrated workflow, or find_moments when visual timing matters. For spoken objections or answers, supply a separate transcript-aware system instead of implying that visual sampling heard the interview.
Example visual scanner set
- Workflow demonstration: the product or workaround visibly shown on screen.
- Visible blocked state: a displayed error, repeated action, or stalled step.
- Artifact extraction: requested fields visible in a form, table, or dashboard.
- Reaction window: a timestamped visible reaction that a researcher can review in context.
Preserve consent and ownership boundaries
Only process recordings your organization is authorized to use. Keep raw links out of tickets and downstream logs. The API returns source metadata and evidence without exposing the submitted source URL in normal control responses.
Build a research repository
Attach interview, segment, study, and participant pseudonym metadata to each run. Query observations across runs, then store links back to the run detail and scanner version. ChatClipThat does not currently replace consent management, research recruitment, or a full repository UI.