We introduced citations and bounding boxes for Box Preview to help users locate, review, and validate extracted metadata back to the data source on the original file. This new capability accelerates human-in-the-loop review by providing a more intuitive and seamless way to locate and identify sources for extracted data.
When reviewing extracted metadata, users cam simply click on any extracted metadata value in the Box Preview panel and instantly be guided to the exact page and highlighted data source with a bounding box for easy visibility. Reviewers no longer need to manually scroll through pages of text to verify whether an AI-extracted value is correct. The bounding box provides immediate, visual evidence of where the data came from. To maintain absolute data integrity, if a reviewer manually edits or deletes an extracted metadata value, the bounding box will automatically be removed from the visible field. This ensures the UI clearly reflects that the final value is now user-confirmed rather than AI-generated.
By eliminating manual document searching, process owners and reviewers can streamline data validation and review to ensure extracted data is trustworthy before it triggers downstream workflows.
Please note that the ability to leverage bounding boxes incurs an additional cost in the form of AI unit consumption. Bounding boxes are toggled 'off' by default, however, users can enable bounding boxes via a dedicated enablement setting when building or customizing Custom Extract Agents in Box Extract. Admins can enable or disable the ability to leverage bounding boxes when using the Box Extract agents in Box Preview via the "Autofill" capability.
To learn more about this release, please see: