Extract data from documents
A Document understanding action — structured extraction — reads each source document and pulls out structured fields, each backed by a source citation (a link to the exact page and text, or the point in an audio transcript, it came from). This is called grounding: every value can be traced back to where it was found.
You configure whether an extraction binds to a record type (“Data object”) or stays as Schema only in the action’s configuration. To launch a run, see Run an action.
The results grid
When an extraction run finishes, the run page shows a “Results” grid — one row per extracted result, with the output fields as columns and a “Source” column frozen on the left. The header summarizes the run, for example “{n} extracted results” or “{n} extracted results · {n} failed.”
Each row carries a status badge:
Cells show a confidence score and a citation-count chip.
The toolbar above the grid gives you:
- “Export CSV” — download every result.
- “Approve all clean (
{n})” — approves the results that still need review and have no warnings. - “Approve selected (
{n})” — select rows first, then approve just those. - “Retry failed (
{n})” — re-run the documents that failed. - “Clear” — clear the current selection.
The review workspace
Click a row to open the full-screen review workspace. The source document sits on the left with the selected citation revealed in context — a highlighted bounding box on the right PDF page, or the audio transcript seeked to the cited moment. The fields sit on the right, each listing its citations (“Page {n}”, “Transcript”, or “Source”), a confidence score, and any warnings.
To correct a value, edit it inline and “Save changes.”
Editing a value drops its evidence: “Saving replaces the extracted value and removes its source citation.” Only edit when you’ve verified the value against the source yourself.
Every run has an editable window. Once it closes, the grid shows “Editing is closed for this run. Results stay available.” — you can still read and export results, but not change them.
When a result is right, use “Approve & import this result” in the review workspace, or approve in bulk from the grid.
Importing into Data
Approving results promotes them into the space’s Data tab as records. An import banner (“Add to data”) walks you through it, previews the results, and offers “Add {n} results to data” and a “View in data” link once they’re in.
If the run wasn’t bound to a record type (“Schema only”), a “Record type” picker appears first — choose the type to import into. Any extracted fields that aren’t part of that record type are ignored.
Reviewing inside a playbook
When the AI assistant runs an extraction as a step in a playbook, the same review happens through a post-action “Output review” card on the playbook run. A “Review” button opens the extraction review surface full-screen; saving there only saves your table edits. When you’re done, “Approve & resume agent” — approving adds the extracted records to Data with every field marked approved, and the agent continues. See Approvals and steering.