Enhancing Your Copilot with Generative Answers

The power of Generative AI has revolutionized how copilots can respond to user queries, making conversations more natural and dynamic. In Copilot Studio, generative answers allow your copilot to pull information from multiple knowledge sources, ensuring that it can answer user questions even when predefined topics fall short.

This blog takes you through how to leverage Generative Answers effectively in Copilot Studio, manage various knowledge sources, and configure your copilot to provide comprehensive, real-time responses to users. Whether you’re using external resources like Bing or internal systems such as SharePoint or Dataverse, generative answers offer versatility and depth to your conversational AI.

What are Generative Answers?

Generative answers in Copilot Studio enable your copilot to source information directly from various knowledge bases, both internal and external, and respond to user queries without relying solely on predefined topics. Essentially, this feature allows copilots to dynamically generate responses based on external or internal data, expanding their capabilities significantly.

For instance, if a user asks a question that doesn’t align with any predefined topics, generative answers act as a fallback, pulling data from configured knowledge sources such as a website, OneDrive, or even a public URL indexed by Bing.

This ensures that your copilot always has an answer, even for unforeseen queries.

Supported Knowledge Sources

Generative answers can utilize both external and internal resources. Some common sources include:

• Public Websites: Pull data from public websites using Bing’s web search, enabling real-time and dynamic answers based on publicly available information.
• AI General Knowledge: Powered by AI models, your copilot can access general knowledge from the web to answer queries.
• Documents in Dataverse: Internal documents stored in Dataverse can be searched and used to provide responses to user queries.
• SharePoint & OneDrive: These internal sources can be tapped using GraphSearch, allowing copilots to access and retrieve relevant documents securely.
• Custom Data: If you have custom data, such as a Power Automate Flow or custom-built sources, you can configure your copilot to retrieve and generate answers based on that data.

Generative Answers as a Fallback

Generative answers also work as a fallback system when no predefined topic or intent matches the user’s query. In this case, the copilot automatically triggers generative answers, searching through your configured knowledge sources to present an appropriate response.

For example, let’s say your copilot has predefined topics for answering questions about office hours or store locations. However, if the user asks something outside of these topics, generative answers will step in to find and present the best possible response from your connected knowledge sources.

If the copilot still can’t find an answer, it can escalate the query using system topics.

Adding and Managing Knowledge Sources

You can incorporate knowledge sources into your copilot at different stages of its development:

1. During Creation: When you first create a copilot, you can add URLs or connect internal knowledge bases like SharePoint or OneDrive.
2. After Creation: You can always add or modify the knowledge sources after your copilot has been created to expand its capabilities.
3. In a Generative Answer Topic Node: A topic node can be configured to use a specific knowledge source for generating answers.

To get the best results from Bing Search, make sure your organization owns the websites being searched through Copilot Studio. This enhances the accuracy and relevance of the responses your copilot generates.

Source Authentication

When using internal knowledge sources like SharePoint or OneDrive, additional authentication steps might be required to ensure your copilot has access to the right data. For example, your copilot users will need Microsoft Entra ID authentication to retrieve information from these internal systems.

Make sure that your URLs and sources are correctly configured for security and access to avoid any disruptions in data retrieval.

Increasing Your Copilot’s Reach

Ready to expand your copilot’s capabilities? Here’s how you can set up generative answers to extend your copilot’s reach:

1. Create a Copilot: Start by giving your copilot a name, selecting the language, and providing a relevant website that can be used to generate answers.
2. Adjust Content Moderation: Use the Generative AI settings to determine how strict or loose the content moderation should be. You can choose between Low (more creative but less precise), Medium (balanced), or High (more relevant but stricter).
3. Test and Refine: Test your copilot by asking questions related to the configured knowledge sources and evaluate the answers. If the results aren’t ideal, you may want to tweak your knowledge sources or content moderation settings.

URL Considerations

When setting up generative answers, it’s important to provide valid URLs that are structured properly:

• The URL can have up to two levels of depth, indicated by forward slashes (e.g., http://www.example.com/product/category).
• If a URL redirects to a different top-level domain, the content won’t be included in the copilot’s results.
• Authentication-required URLs, such as internal wikis or private SharePoint sites, cannot be used for public generative answers.

Be mindful of the structure of your URLs to ensure your copilot is using the correct sources of information.

Testing and Improving Generative Answers

Once your copilot is set up with knowledge sources, test it extensively:

• Test Different Query Types: Ask your copilot various types of questions to see how well it retrieves data from knowledge sources.
• Use Specific Questions: Test with specific and general queries to understand how generative answers work with your copilot’s knowledge base.
• Edge Case Testing: Test extreme cases, such as complicated or very general questions, to see how your copilot responds.

In the case of irrelevant or nonsensical answers, provide feedback through the “thumbs up” or “thumbs down” icons in the chat window to improve the AI’s response accuracy.

By leveraging generative answers in Copilot Studio, you can build a truly dynamic and responsive copilot. With access to multiple knowledge sources, your copilot can provide accurate and timely responses, no matter the complexity of the query. Whether you’re connecting to public websites, internal documents, or AI general knowledge, generative answers offer an expansive and flexible approach to building conversational AI.

Start experimenting with generative answers today and give your copilot the power to handle a wider range of user queries with ease!