Skip to main content

Microsoft Copilot Unpacked: Web-Grounded / Work-Grounded

WEB-GROUNDED in Microsoft Copilot

AI models like GPT-4o are trained on large datasets, but their knowledge only goes up to a certain point (for example, mid-2024). This means they cannot automatically know real-time events or the latest updates.

Web-grounded capability solves this limitation by connecting Copilot to the internet through Bing Search. When you ask a question that requires current information, Copilot:

  1. Searches the web using Bing to find the most relevant and up-to-date content
  2. Analyzes and filters that information for accuracy and trustworthiness
  3. Combines the fresh data with the model existing knowledge
  4. Generates a response that is both smart and current

Why It Matters

Without web grounding, Copilot would only rely on static training data. With this feature, you get answers that reflect todays reality, such as:

  • Latest product updates
  • Current market trends
  • Breaking news or recent events

WORK-GROUNDED in Microsoft Copilot

While web grounding brings in public information, work grounding focuses on your organizations internal data. Copilot uses Microsoft Graph to securely access:

  • Emails
  • Files
  • Chats
  • Meetings
  • Other enterprise resources

This allows Copilot to:

  1. Understand your work context (projects, documents, conversations)
  2. Combine internal data with model knowledge for personalized answers
  3. Provide relevant, secure, and compliant responses tailored to your organization

Why It Matters

Work grounding ensures Copilot is not just smart, but context-aware. Examples:

  • Summarizing your recent meetings
  • Drafting emails based on your companys tone and policies
  • Answering questions using internal documents and resources

Summary

TypeDescription
Web-GroundedBrings in current public knowledge from the internet
Work-GroundedUses your organizations data for personalized, secure answers

Together, these capabilities make Microsoft Copilot powerful, accurate, and context-aware.