Hi everyone, I wanted to highlight an issue I encountered with ChatGPT’s handling of comments in Word documents and share a workaround I’ve been using.
The Problem
When I asked ChatGPT whether it could see and respond to comments in a Word document, it initially stated (quite confidently) that it could:
- View the comments in the context of the attached text.
- Identify the text, section, or phrase the comments referred to.
- Provide suggestions or updates based on those comments.
Encouraged by this, I uploaded a Word document with comments. ChatGPT acknowledged the comments, saying it would respond to them systematically—only to later backtrack, stating that Word comments were not visible within its capabilities. This left me scratching my head, wondering why it initially claimed the functionality and acknowledged the comments if it couldn’t actually process them.
The Contradiction
The conflicting responses suggest that:
- ChatGPT might assume it can handle certain tasks without verifying first.
- There’s a lack of clarity in its communication when capabilities fail mid-process.
- The parsing process for Word documents doesn’t account for comments as separate entities, despite initial claims.
This behavior creates confusion and undermines trust when users expect functionality based on ChatGPT’s confident statements.
My Workaround
Since ChatGPT can’t directly access Word comments, here’s what I’ve been doing instead:
- Printing the commented Word document in a “Save as PDF” manner, which – if you’re “showing” the comments in Word at the time, shows both the text and the comments as well as which portion of the text the comment is attached to.
- Uploading the document as reference material for context.
- Confirming that ChatGPT now can see the comments and what they’re attached to, and respond to each comment accordingly.
While this works, it adds extra steps and isn’t ideal when working with large, heavily annotated documents.
Questions for OpenAI
- Why does ChatGPT assume it can perform a task before verifying its capabilities?
- Could Word document parsing be improved to include comments as visible data?
- If functionality is limited, can ChatGPT communicate this upfront to manage user expectations better?