SMART WORLDJOURNALSubmit Manuscript
← All technology news

Artificial Intelligence

Amazon Textract Insights Launches Verifiable AI Document Processing

Amazon unveils Textract Insights, a generative AI enhancement for its Textract service, enabling organizations to extract structured, confidence-scored data from complex, multi-document files via natural language queries without custom model training.

NEWS

Amazon has announced the general availability of Amazon Textract Insights, an advanced generative AI feature integrated within Amazon Textract. This service allows organizations to extract data from scanned files and paper forms by describing desired fields in natural language, returning structured data complete with confidence scores and page-level grounding, all accessible through a single API call.

Traditional document processing often struggles with mixed packets containing various document types and layouts, such as mortgage applications or insurance claims that bundle forms, payslips, handwritten notes, and other materials into single files. Existing approaches require either complex custom pipelines integrating OCR, classification, and language models, or rely on general-purpose large language models which lack traceable source citations and confidence metrics.

Amazon Textract Insights simplifies this by enabling users to define extraction blueprints—JSON schemas with natural language descriptions of target fields and their inference types. Users can combine public blueprints for common documents or author custom ones and submit multi-document files up to 3,000 pages. The service automatically splits the packet, matches documents to blueprints, extracts specified fields, and returns each value with a confidence score and precise page and bounding box references.

The confidence scoring and source grounding features enhance auditability and facilitate selective automation. Reviewers can verify extracted data against original documents, ensuring that automation extends to higher-risk documents while retaining human oversight where needed. This approach addresses the trade-off between automation and auditability prevalent in intelligent document processing.

Amazon also updated the Textract Core reading engine to improve accuracy, handwriting recognition, and expand language support to 40 languages, benefits inherent to Textract Insights. The new Insights feature supports multiple input formats including PDF, PNG, JPEG, and TIFF and currently processes English, French, German, Italian, Portuguese, and Spanish.

Security measures include encryption of projects and blueprints with AWS-owned or customer-managed keys at rest, secure transmission over AWS’s network, and transient storage of intermediate job data. Additionally, Textract Insights intelligently routes inference requests to the optimal AWS region within the user’s geographic area to maximize performance and data residency compliance.

Limitations include Textract Insights not yet being available in all AWS regions where the core Textract engine operates, restricting geographic deployment options. Furthermore, while the service supports multiple languages, its coverage is less comprehensive than the broader Textract Core engine’s 40 languages.

Amazon Textract Insights represents a significant step in verifiable, generative AI-powered document processing, combining automation efficiency with traceable, confidence-based output to improve data extraction workflows across sectors like financial services, healthcare, insurance, and public sector organizations.

Source & further reading

Announcing Amazon Textract Insights: Verifiable generative AI for documents ↗

Source published 2026-11-20. Claims and limitations should be checked against the original announcement or research.