A Guth Labs publication ↗
Back to News

AWS launches Textract Insights for document extraction with source references

Blank cream sheets pass through a green scanner beneath a gold light band.
AI-generated editorial illustration

Amazon Web Services made Textract Insights generally available on November 20. The new API lets teams describe the fields they want in plain language and receive structured results, with a confidence score and a source location for each value.

The service accepts files from a customer’s Amazon S3 bucket, including mixed packets of up to 3,000 pages in a single file. It can split a packet into separate documents and apply different field definitions to each one. Results distinguish values read directly from the page from values inferred from the document. Both include the page and location used to produce them.

AWS says Insights supports six languages and bills per page with no minimum commitment. It is not yet available in Tokyo or Zurich, even though the existing Textract reading service is available in both. The source references let reviewers check an extracted or inferred value against the document instead of relying on the confidence score alone.

Sources and citations

Revision history

Published .