ISSUE

If using Auto Extract Locator on documents that have handwritten text, you may notice inconsistent results,

such as sometimes it's recognizing handwritten text and other times it doesn't, regardless of what prompt you input for the locator.

SOLUTION

Auto Extract Locator is designed for semantic extraction of OCR/ICR results and is not intended for handwriting detection. The locator operates on the text output generated by OCR/ICR rather than the visual characteristics of the document itself. As a result, the LLM cannot determine whether extracted text originated from handwritten or printed content.

Based on this, handwriting detection is not a supported use case for Auto Extract Locator and the inconsistent results observed are expected. It was also confirmed previously that there are currently no planned enhancements in this area.

It seems that trying with different prompts may improve the results to a degree, but it will still be inconsistent.