What is Amazon Textract? A plain-English explainer

What it does

Point Textract at a document — a scanned form, an invoice, an ID, a multi-page PDF — and it returns the text it finds, plus structure: tables broken down into rows and columns, and forms broken down into key-value pairs, such as recognising that "Invoice Number" is a label and the digits next to it are its value. It also handles handwriting, not just printed text, which matters for a lot of real-world paperwork that was never typed in the first place.

How it differs from plain OCR

Traditional optical character recognition answers one question: what characters are on this page? That’s useful, but it hands back an unstructured wall of text — you still have to work out, yourself, which words form a table row, or which label a number belongs to. For anything beyond a simple document, that reconstruction work is often harder than the character recognition itself.

Textract answers a richer question: what characters are on this page, and how are they organised? By returning tables and key-value pairs directly, it removes the reconstruction step that plain OCR leaves for you to solve — which is exactly why it’s described as OCR plus an understanding of document structure, not just character recognition.

Common uses

That structural understanding is what makes a handful of real automation use cases practical:

• Automating forms, invoices and receipts — pulling the relevant fields straight into a database or accounting system without manual data entry.
• Identity document processing — extracting fields from IDs or application forms as one step in an onboarding flow.
• Document-processing pipelines — feeding scanned paperwork through an automated sequence of steps (extraction, validation, storage) rather than a person retyping each document by hand (see our explainer on what a data pipeline is).
• Digitising archives — converting large volumes of historical paper records into searchable, structured digital data.

How it pairs with other AI services

Textract typically isn’t the last step in a workflow — it’s the first. Once it has extracted the raw text and structure from a document, that text is often handed to a natural-language-processing service to actually understand it: pulling out sentiment from extracted customer complaints, or entities like names and dates from an extracted contract, using a service such as Amazon Comprehend. Extraction and understanding are genuinely different jobs done by different tools, and treating them as one step is a common design mistake.

Pricing model

Textract is billed on a pay-for-use basis, typically tied to the number of pages processed and which analysis features (tables, forms, plain text) are requested — the exact rates change over time, so check AWS’s current pricing documentation rather than a fixed number quoted here.

Cert-study tie-in

Textract is the standard AWS example for document-understanding AI, and certification exams test the conceptual distinction covered here — extraction versus plain OCR, and where structured document data flows to next in a pipeline — rather than API syntax. It shows up in both the AWS Certified AI Practitioner and Developer – Associate exams. Our /revision lessons cover the full syllabus depth for each.