Feedback summary

A feedback summary is an AI-written prose summary of everything a survey has collected. It sits on the survey’s detail page, above the response list.

The summary is written as prose with headings, not as a fixed set of fields. What it contains depends on what the responses contain.

Where it comes from

Survey type	Summary behaviour
Agent Survey	A results brief is produced for you.
Static Survey	No summary by default. You generate one explicitly, and doing so requires a plan that includes Agent Surveys.

Retrospectives are separate. A closed retro has its own summary with statistics and top-voted cards — see Closing a retro.

Generating one

Choose Generate summary on the card. Generation runs in the background, so the card shows a generating state rather than blocking the page.

A summary is in one of three states:

State	Meaning
`generating`	The brief is being written. Regenerating is refused until it finishes.
`complete`	A summary exists and is displayed with the date it was generated.
`failed`	Generation did not complete. Try again from the same button.

Generating a summary does one other thing worth knowing: it runs signal extraction across the responses in a batch, which is what lights up themes and citable evidence for that survey. So generating a summary is also what makes a Static Survey’s responses queryable as evidence.

When it goes stale

The summary records which responses it was built from — how many there were, and the submission time of the most recent one.

Responses that arrive afterwards are not folded in automatically. A survey is treated as having new material when either the number of responses has grown or any response was submitted later than the newest one the summary covers. Regenerate to take them in.

This is why the generated date on the card matters: it tells you the evidence cut-off, not just when someone last pressed the button.

Regenerating

Regenerate rewrites the summary from the current responses, replacing the previous one. There’s no history of earlier summaries.

Regenerating while a summary is already generating is refused rather than queued. Wait for the current run to finish.

Copying

Copy summary puts the summary on your clipboard as text, for pasting into a doc, a ticket, or a message.

If you’re looking for the old Feedback Summary

The v2.5 Feedback Summary was generated automatically at the end of every feedback session and always showed the same four things: the session name, an Engagement Score, sentiment analysis, and issues found. v3 works differently:

It’s generated on request, not automatically on completion — except for Agent Surveys, which produce a brief for you.
The Engagement Score no longer exists. Nothing in the platform computes it.
Sentiment and detected issues moved to Signals, which keeps the evidence behind each one rather than reporting a bare score.
The “what the team needs to know” and “what leadership needs to know” split is gone. Ask a question scoped to the audience you care about in Ask instead.

Responses and transcripts — the evidence a summary is derived from
Agent Surveys — how the results brief fits the Agent flow
Ask — for questions a summary doesn’t answer