Hi
@ZN-26022205-0
DOCX is supported by Copilot Studio, so I wouldn't expect you to have to convert the Word document to PDF.
The first thing I'd check is how the DOCX is being provided to the agent, because there are two different scenarios.
1. If you're uploading the DOCX directly into the chat
Make sure File uploads is enabled under the agent's settings.
DOCX is one of the supported file types for user file input, with a maximum individual file size of 15 MB. (learn.microsoft.com)
Also test with a very simple Word document containing only normal paragraphs, for example:
Test Document
Customer: Contoso
Order: 12345
Status: Approved
Then ask:
What is the customer name and order status from the attached document?
If PDF works but this simple DOCX doesn't, that points more toward a file-processing issue than your prompt.
2. If the DOCX is an agent knowledge source
DOC/DOCX are also explicitly supported as uploaded knowledge sources. (learn.microsoft.com)
After uploading the file, check its status under Knowledge. A newly uploaded file can initially show In progress and needs to reach Ready before the agent can use it. Microsoft notes that this processing can take a few minutes. (learn.microsoft.com)
If it remains stuck in In progress, I would remove the file and upload a fresh copy rather than repeatedly testing the conversation.
3. Check the contents of the DOCX
This is an easy one to miss.
For user-uploaded Word documents, Copilot Studio extracts text from normal paragraphs and tables. It may not extract content that exists only in:
- Text boxes
- Shapes/WordArt
- SmartArt
- Headers/footers
- Embedded or linked objects such as Excel files
So a DOCX can technically be supported while still appearing to contain little/no usable text to the agent. (learn.microsoft.com)
4. Check for protection/encryption
Password-protected or encrypted Office files aren't supported for the document knowledge-source scenario. (learn.microsoft.com)
I'd test with a completely new, unprotected .docx.
What I'd test
I'd create a brand-new Word file with 2–3 plain paragraphs, upload that instead of the existing document, and ask the agent a question whose answer is explicitly in that file.
If:
PDF → works
Simple DOCX → doesn't work
then I'd suspect a current DOCX ingestion/attachment regression rather than a configuration issue, especially since you mentioned DOCX worked a few days ago.
Also, if the agent keeps answering from an existing knowledge source instead of the attached file, make the test question very specific to the new document. This helps determine whether the problem is file ingestion or simply knowledge-source selection/grounding.
I would therefore not disable/rebuild the whole agent yet. First isolate it with a tiny, unprotected DOCX. DOCX is currently a supported input and knowledge-source format. (learn.microsoft.com)
