Amazon Web Services has announced the general availability of Amazon Textract Insights, a generative AI-powered document extraction service designed to answer one of enterprise AI's most persistent complaints: models that produce plausible results with no way to verify where they came from. The launch, detailed in an AWS blog post on November 20, pairs large language model extraction with page-level citations and confidence scores for every value returned.
Textract has long been AWS's engine for turning scanned pages into text and tables. Textract Insights adds a new API on top of that foundation: users describe, in plain language, the fields they want extracted, upload a document, and receive structured results in return — with no model training required. AWS is pitching the service at banks, insurers, healthcare organizations, and government agencies, the paperwork-heavy industries where verification is not optional. For more tool launches like this, check out our coverage of the latest AI developments.
How Textract Insights Works
Setup is organized around what AWS calls blueprints: a JSON Schema in which each field carries a plain-English description of what to find. Each field is marked as either read verbatim from the page or inferred from the document's content — a distinction that lets reviewers know exactly which values came straight off the paper and which were derived by the model.
AWS provides ready-made public blueprints for common form types, and customers can write their own for business-specific documents. The result, according to AWS, is an extraction pipeline that a reviewer or auditor can check against the original document, field by field.
Built for Messy, Real-World Paperwork
Real document flows rarely arrive tidy. A mortgage application might arrive as a single scan containing a W-2, payslips, bank statements, and handwritten notes in no particular order. Textract Insights is built for exactly that case: it accepts a single file of up to 3,000 pages, splits it into separate documents, matches each one to the right blueprint, and extracts the fields — all as one asynchronous request through an API named StartDocumentInsights.
Results land in the customer's own Amazon S3 bucket. Every extracted value includes the page number and bounding box it was drawn from, so a human reviewer can trace any figure back to its exact source location in seconds.
AWS shared a statement from Symbeo, a CorVel company, which said plain-language field descriptions help because the medical bills it processes arrive in many different layouts — rules built for one sender often fail on another.
Why Verifiability Matters for Enterprise AI
The core pitch is trust. AWS argues that a general-purpose chatbot can give fluent answers but often cannot say which page a number came from or how confident it is. Textract Insights is designed so that automated workflows can act on high-confidence values automatically while routing uncertain ones to a human reviewer — a pattern compliance teams in finance and healthcare have been asking for since generative AI extraction first appeared.
Pricing follows a per-page model rather than per-token, which AWS says should make costs more predictable than typical language model APIs. Specific rates were not listed in the launch announcement; AWS points customers to the Textract pricing page.
On security, projects and blueprints are encrypted with AWS-owned keys by default, with the option to bring your own key through AWS KMS. Intermediate job data is temporary and deleted when a job finishes, and AWS says requests are routed to compute within the same geography.
Textract Core Gets an Upgrade Too
Alongside the Insights launch, AWS updated Amazon Textract Core, the reading engine underneath the service. The company says accuracy has improved on dense tables and handwriting, and language coverage now reaches 40 languages for the standard text and analysis APIs. Existing customers get those gains without changing any code.
Core is also now available in Tokyo and Zurich, bringing the service to 18 AWS regions in total. Textract Insights itself is not yet offered in those two new regions, and AI Parabellum's review of the launch notes that Insights currently supports a smaller set of languages than Core.
Where It Fits in AWS's AI Push
The launch is part of a steady drumbeat of AWS AI releases aimed at enterprise workflows rather than consumer chatbots. By wrapping generative AI in citations, confidence scores, and audit-friendly primitives, AWS is betting that verifiability — not raw model quality — is what unblocks AI adoption in regulated industries.
For teams that handle loan files, insurance claims, or patient forms, the practical value proposition is consolidating what used to be a chain of OCR, splitting, classification, and model prompting into a single API call. Whether that simplification holds up on the messiest real-world documents will be the measure of the service's success.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →