Announcing Amazon Textract Insights: Verifiable generative AI for documents

What happened
Amazon has introduced Amazon Textract Insights, a new generative AI feature available through Amazon Textract. The feature allows users to extract structured data from complex, multi-document files using a single API call, without the need for model training. Users can specify the fields they want in natural language, and the system returns results with confidence scores and source citations. The service is now generally available and targets industries that handle diverse document types, such as finance and healthcare.
Key details
- Amazon Textract Insights processes up to 3,000-page files from Amazon S3.
- Users define extraction requirements using blueprints, which are JSON Schemas with natural language field descriptions.
- Each extracted value includes a confidence score and page-level grounding for traceability.
- The system can split mixed document packets into logical sub-documents and match them to relevant blueprints.
- Amazon Textract Core now supports 40 languages and is available in additional AWS regions, including Tokyo and Zurich.
Why it matters
Organizations in sectors like finance, insurance, healthcare, and the public sector can now automate document processing tasks that previously required custom-built pipelines or manual review. This reduces the need for technical integration work, lowers operational complexity, and allows teams to focus on reviewing only those fields where the AI's confidence is low. The new feature enables faster, more auditable processing of large, mixed-document files, improving efficiency for businesses that manage high volumes of paperwork.
