Hello Haritha,
Thank you for your detailed question about retrieving comprehensive results from external knowledge sources in Copilot Studio. The limitation you're experiencing with only getting top 3 results is due to default response limits and search configurations. Here's how to address this:
Understanding the Limitation
Copilot Studio's Generative Answers node has built-in result limits to:
- Ensure response quality and relevance
- Prevent token limit exhaustion
- Maintain reasonable response times
Solutions to Retrieve More Results
1. Custom Topic with Enhanced Search Configuration
Create a custom topic with these settings:
Topic: Comprehensive Q&A Search
Trigger Phrases: "all questions about", "complete list", "comprehensive search"
Variables:
- Keyword (Text)
- Subject (Text)
2. Modify Generative Answers Node Settings
In your Generative Answers node:
Data Source Configuration:
- Set "Number of sources" to maximum (typically 20)
- Enable "Search across all content"
Search for ALL questions and answers from the knowledge source where the keyword appears in the question text. Return a comprehensive list including:
1. Question number/identifier
2. Complete question text
3. Full solution/answer
4. Organize by relevance but include all matches
3. Advanced Prompt Engineering
Use specific prompts that request comprehensive results:
"Retrieve every question from [subject] where '[keyword]' appears in the question.
Format as: Question [number]: [full question] | Answer: [complete solution].
Continue until all matching questions are listed."
4. Iterative Search Approach
Create a multi-turn conversation flow:
- Initial search returns first batch
- Follow-up prompts: "Show me more results" or "Continue the search"
- Use conversation memory to track what's already been shown
5. Knowledge Source Optimization
Ensure your external website data is properly indexed:
- URL Configuration: Verify all relevant pages are included
- Content Depth: Set crawling depth to capture all Q&A pages
- Refresh Frequency: Regular updates to maintain current content
6. Alternative Technical Approaches
Option A: Power Automate Integration
- Create a flow that scrapes the website comprehensively
- Store results in SharePoint or Dataverse
- Reference this structured data in Copilot
Option B: Custom Connector
- Build a custom connector to the website's API (if available)
- Implement pagination to retrieve all results
- Return structured JSON with all matching Q&As
Implementation Example
Custom Topic: "Complete Physics Q&A Search"
Node 1: Question (Keyword Input)
"What keyword would you like to search for in physics questions?"
Node 2: Generative Answers
Data Source: External Website
Instructions: "Search the entire knowledge base for ALL questions containing the keyword '{Keyword}'. Return every match with question number, full question text, and complete solution. Do not limit results."
Node 3: Follow-up Options
"Would you like to:
1. Search for another keyword
2. Filter by specific chapter
3. Get additional results if available"
Best Practices
- Chunk Large Responses: If hitting token limits, implement pagination
- Structured Output: Request specific formatting for easier parsing
- Conversation Memory: Track previous searches to avoid duplicates
- User Control: Let users specify how many results they want
Limitations to Consider
- Token limits may still restrict extremely large result sets
- External website structure affects search quality
- Response time increases with result volume
Thanks,
Karan Shewale.
*************************************************************************
If the response is helpful, please click "Accept Answer" and upvote it. You can share your feedback via Microsoft Teams Developer Feedback link. Click here to escalate.