---
title: "Comment by Vaibhav Srivastava on \"Get started with AWS Textract\" | Hashnode"
description: "A useful implementation detail to add after extraction: Textract's blocks describe geometry and relationships, but they do not automatically become a stable edi"
url: "https://hashnode.com/posts/get-started-with-aws-textract/658dcace3baff33db738181d/comment/6ab8dbc8f8f7f3693d5971c6"
---

[Discussion](https://hashnode.com/posts/get-started-with-aws-textract/658dcace3baff33db738181d)

Comment [Get started with AWS Textract](https://hashnode.com/posts/get-started-with-aws-textract/658dcace3baff33db738181d)

[Vaibhav Srivastava](https://hashnode.com/@vaibhavjeet19vaibhav)

[Sep 27](https://hashnode.com/posts/get-started-with-aws-textract/658dcace3baff33db738181d/comment/6ab8dbc8f8f7f3693d5971c6)

A useful implementation detail to add after extraction: Textract's blocks describe geometry and relationships, but they do not automatically become a stable editable document. For multi-column pages, sort LINE blocks within detected layout regions instead of globally by Y coordinate, or copied text can interleave columns. For tables, follow CHILD relationships into CELL blocks and preserve row/column spans before exporting; flattening everything into strings loses the structure users expect in Word or Excel. I also recommend storing confidence beside every critical field and routing low-confidence names, dates, totals, and handwriting for review rather than silently accepting them. A successful API response proves extraction completed, not that the result is correct.
