Power Platform Community Forum Thread Details

I am experiencing an issue with a chatbot created in Copilot Studio that retrieves links and documents from a PDF file hosted in SharePoint. This PDF contains portal names and their corresponding links, serving as the chatbot's knowledge base. The expected behavior is for the chatbot to accurately process user queries and provide the correct link or document from the PDF. However, the chatbot often provides irrelevant or incorrect links in response to user queries.

For example, when a user asks for the link to the "HR Portal," the chatbot might return a link to a completely unrelated portal or provide an incorrect response altogether. It appears that the knowledge base from the PDF is not being properly utilized or indexed to match user queries accurately.

The chatbot was configured as follows:

A PDF containing the relevant portal names and links was uploaded to SharePoint.
This PDF was linked as a data source in Copilot Studio.
The Generative Answers node was set up to dynamically respond to user queries based on the data in the PDF.

I have taken steps to resolve the issue by verifying that the PDF is properly linked, ensuring that it contains clear labels and corresponding links, and testing various user prompts. Despite these efforts, the chatbot continues to provide inaccurate responses.

I would like to know how I can enhance the chatbot's intelligence and accuracy within Copilot Studio, without involving external flows like Power Automate. Specifically:

Are there best practices for formatting the PDF data source (e.g., structured tables or metadata) to improve accuracy?
How can I make the chatbot to give the relevant links upon getting the user query? Any workarounds that can be made within the copilot studio?
Can Copilot Studio's Generative Answers node be optimized to better match user queries with the data source?
What are the limitations of Copilot Studio in this context, and how can they be mitigated?
Additionally, may I know what is the purpose of conversational boosting and how & when it is used? Give me an example to understand better.

It would be very helpful if you could provide real-world examples of how to address similar issues and optimize the chatbot's ability to deliver accurate responses using a PDF as the primary knowledge base.

Categories:

General topics