Nov 18, 2026: AI_EXTRACT citations (Preview)

Snowflake Cortex AI_EXTRACT now supports citations that connect extracted values to supporting content in source documents and text. This addition makes it easier to verify extraction results, review information in context, and build applications that display source evidence alongside extracted fields.

Key capabilities include:

  • Document citations: Identify supporting content using page numbers and bounding boxes, enabling applications to highlight the relevant regions in source documents.
  • Text citations: Locate supporting passages using start and end offsets in the input text, enabling applications to display the source text associated with an extracted value.
  • Citations with extraction scores: Request citations alongside extraction scores to help prioritize results for review and inspect the supporting source content.
  • Integration with existing SQL workflows: Enable citations by setting citation => TRUE in an AI_EXTRACT call. Citations are returned in a citations object alongside the extracted values, with no separate function call required.

For more information, see Extract information with citations in the AI_EXTRACT user guide and Citations in the AI_EXTRACT SQL reference.