Optimize Knowledge Content for Large Language Models
Learning Objectives
After completing this unit, you’ll be able to:
- Describe grounding in generative AI.
- Explain writing and process guidelines for knowledge bases.
What Is AI Grounding?
Grounding connects your AI model to trusted information sources to improve the accuracy and relevancy of your AI features. AI models are born generalists: They train on large language model (LLM) datasets that provide them with a broad base of knowledge. But this broad knowledge base means they don’t have the specialized information needed to perform specific tasks or answer technical questions.
To get smarter responses and the ability to complete complex tasks, you need to provide your AI model with the right tools and information. Grounding with retrieval-augmented generation (RAG) intelligently connects your knowledge base to your AI features. This helps ensure that AI-generated content is more accurately tailored to your organization and use cases.
For customers, accuracy is everything. Give a customer the wrong answer, and it could discourage them from future purchases. Provide agents with incorrect information, and they might frustrate customers rather than support them. By grounding your AI model in verified sources of information like your knowledge base and uploaded files, the LLM can accurately return responses to customer inquiries, suggest replies to agents, provide search summaries, and more. To get the best results for your AI-powered features, prepare your knowledge base for grounding.
Prepare Your Knowledge Base
Your Salesforce knowledge base likely stores a wealth of information specific to your organization, from formal policies and processes to troubleshooting guides or special announcements. Your Salesforce knowledge base also contains metadata to go along with the articles themselves, which helps the LLM sort through data to find the most up-to-date and relevant information.
For additional knowledge outside Salesforce—such as in Google Docs, Microsoft SharePoint, Atlassian Confluence, or elsewhere—you can import and sync it with Unified Knowledge. Unified Knowledge integrates all your third-party knowledge in Salesforce. Once all your knowledge articles are in Salesforce, agents and customers can find the information they need from one platform.
To improve grounding results, consider these guidelines for writing knowledge articles.
Explain Thoroughly and Give Examples
Generative AI works best with comprehensive and detailed information. It stores, organizes, and strategically searches through this data to deliver insights at various levels of detail for users. In your articles, always lean toward including more information, especially technical details, so it has enough information to return the right answers as well as any relevant context. Think about common, real-world situations users face and write about them with specific examples in your knowledge content.
Structure Content
AI prefers structured content. Write sentences with logical relationships and break up content into paragraphs or lists. Use heading tags (like H1 to H6) to show information hierarchy. Use correct grammar and a straightforward sentence structure when writing, and separate your internal and customer-facing information into different fields or article types.
Annotate Media
If you use visual content such as videos, screenshots, or animated gifs in articles, annotate them with text descriptions. AI requires captions and alt-text to understand visual content.
Answer Common Questions
When adding content to your knowledge base, consider the common questions or requests from users. For frequent requests, make sure you have enough information to wholly address them.
Create Focused Content
Write articles that cover a single topic. For example, instead of covering ten issues in one article, break up the content into ten separate articles. Focused writing also helps you manage security and visibility, as certain topics are appropriate for some audiences but not others. Also, focused content helps AI generate more relevant output for a given situation or use case.
Manage Your Knowledge Base
Next, to ensure that your knowledge base remains a reliable and effective resource for generating AI output, consider the processes needed to manage it. To get the best grounding results, consider using the following process guidelines when developing knowledge articles.
Follow Knowledge-Centered Service (KCS) Principles
Knowledge-Centered Service (KCS) Principles are a set of industry-standard best practices on knowledge management. They help you create accurate and consistent articles, which in turn improves knowledge grounding. We’ve listed out some key practices here, but you can learn more at the Consortium for Service Innovation.
Audit Your Knowledge Base
Regularly review and update your articles so that your AI-generated content does not include incorrect and outdated information. Use Salesforce Flow to build logic on articles’ Next Review Date field to identify content for review.
Include Sources
To build trust and help users dive deeper into information, add citations and link to the sources your knowledge is grounded in. Proper references to information give credibility to generated content.
Remember Humans
Prepare content that’s both consumable by AI, and useful and understandable to humans. Remember, AI is here to assist humans and work with us in our day-to-day tasks.
Conclusion
Now that you’ve learned about AI grounding and best practices for knowledge management, you’re ready to get the best results from your AI-generated content and Agentforce. Get started today by organizing an audit of your knowledge base, and using Unified Knowledge to bring in your trusted content from other sources!