This month’s SharePoint Showcase highlights a breakthrough for organizations embracing AI: Knowledge Agent in SharePoint. As more teams rely on Copilot and agents to drive productivity, the difference between generic and truly valuable responses comes down to how your content is structured and enriched.
In this blog, we’ll explore how metadata unlocks smarter, more deterministic responses from Copilot and agents, and how Knowledge Agent makes it effortless to embrace metadata.
Metadata and why it matters for AI, Copilot, and agents
Metadata is the descriptive information attached to files (think tags, categories, dates, and custom labels). It’s what transforms a simple document library into a rich, searchable knowledge base.
Knowledge Agent extracted the candidate name, most recent title, and professional summary as the metadata associated with each file in this ‘Software Engineer Resumes’ library.
In September, Microsoft rolled out metadata understanding to agents, including Knowledge Agent, enabling them to deliver more precise, context-aware answers by leveraging file metadata in SharePoint. Not only does Knowledge Agent understand metadata to provide more accurate Q&A, it also automates metadata population, among a number of other skills to keep your content and sites updated and AI-ready.
Coming in November, this same metadata understanding is arriving for Microsoft 365 Copilot, when you reference files, folders or libraries. This means Copilot, Knowledge Agent, and agents grounded on SharePoint content with associated metadata will soon be able to reason over tags, categories, and business context, not just the text inside your documents. This ability to reason over SharePoint metadata is an advantage only available to AI built and used within the Microsoft ecosystem – Copilot and its associated agents.
This evolution marks a major step forward: with metadata, organizations can now rely on AI to deliver answers that are more precisely anchored in the structure and business context of their content. For many of our customers, deploying AI that produces accurate, audit-ready, deterministic answers is not optional – it’s business critical. Metadata is the key to providing the guardrails that unlock trustworthy, business-ready AI.
The challenge: LLMs are probabilistic. How can metadata make AI output more deterministic?
Large Language Models (LLMs), which Copilot and agents rely on, are probabilistic in nature. They generate responses based on patterns in data, which makes them creative and adaptable. This flexibility is great for open-ended questions and dynamic scenarios.
However, industries like law, finance, engineering, and healthcare, need deterministic answers: consistent, repeatable, and grounded in facts. Deterministic systems always produce the same output for the same input, which is crucial for compliance and regulatory needs.
The agent uses metadata to deliver precise answers from structured data.
Making AI fully deterministic would sacrifice the creativity and nuance that make LLMs valuable, reverting to rigid, rule-based chatbots. Instead, the solution is to create deterministic environments for AI to operate safely. By enriching SharePoint content with metadata, organizations set clear boundaries, and metadata acts as the anchor, helping Copilot and agents deliver reliable, business-ready answers while retaining their creative strengths.
LLMs will always be probabilistic at their core, but with structured, metadata-rich environments, AI can be much more precise when business needs require it.
Real world scenario: The difference metadata makes in AI responses
Let’s look at how metadata features make a tangible difference for organizations:
Scenario 1: Without Metadata
Imagine Carole, a finance manager, maintains a SharePoint library of expense invoices and transaction documents, but the files lack structured metadata.
Carole submits her prompt to the agent:
“List all expense transactions over $10,000 posted between July 1 and September 30, 2025, tagged as cap ex and include the associated cost center, vendor name, and approval status. Sort by transaction date in ascending order.”
The agent returns a single transaction in response. While this appears helpful at first glance, what Carole doesn’t realize yet is that there are actually three transactions in the library that meet these criteria. The agent is missing key details within the document contents and fails to surface the complete set of results.
Scenario 2: With Metadata
When Carole leverages Knowledge Agent to organize her library, each file is tagged with extracted metadata like transaction date, amount, expense type, and more details relevant to Carole’s business context.
You can specify the metadata columns you need or let the agent do it for you.Now, when she asks the same question, the agent instantly returns all three matching transactions, each with the correct cost center, vendor, and approval status—precisely sorted by transaction date. The answer is complete, accurate, and ready for audit or reporting.
After metadata, the agent's response is complete and accurate.
Metadata transforms AI responses from “best guess” to business-ready. With Knowledge Agent, Copilot and agents can filter, group, and sort results by any metadata column, making it effortless to get the right answer, every time.
MVP Insights: Expert strategies and quick wins with the Knowledge Agent
Microsoft MVPs (Most Valuable Professionals) are recognized experts and community leaders who share their deep knowledge and real-world experience with Microsoft technologies.
This month, Femke Cornelissen (MVP) shares some practical solutions using Knowledge Agent in SharePoint!
Explore the Knowledge Agents in SharePoint cheatsheet here.
Stay tuned each month to hear more from our community for fresh perspectives and actionable guidance from experts.
Best practices: Using Knowledge Agent’s content management skills
Even before generative AI, adding metadata to SharePoint content has always been important, particularly for organizations with regulatory and compliance content requirements. But adding metadata has often been a manual, time-consuming process. Knowledge Agent solves this by automating metadata suggestions, tagging, and organization, making it easy to keep your content Copilot-ready and compliant with minimal effort.
Below are a set of best practices for using Knowledge Agent based on feedback from our customers in preview, our community, and our own usage at Microsoft.
Folders vs. Flat Lists:
Use folders to represent natural boundaries like teams or projects. Flat lists work best for content with shared characteristics, such as all invoices or resumes for a particular role. Knowledge Agent works best with flat lists, so choose the structure that fits your needs.
Create Columns:
Select “Organize this library,” and then let Knowledge Agent suggest columns by selecting “create columns,” or ask the agent to create your own. Think about how you want to filter and group your documents. Use the classification option for a simple way to categorize files by document type. This is great for basic sorting and filtering. For example, a loan officer might ask the agent to create columns for loan type, loan amount, and summary.
Automatic Metadata Suggestions:
The agent auto-fills columns with metadata based on your files, including those that are images or scans.
Thanks to built-in Optical Character Recognition (OCR), Knowledge Agent reads text from PDFs and images, extracting key details like dates, names, and document types. This ensures even non-text files are fully searchable and ready for Copilot and agents to reason over.
For example, take this library of scanned Microsoft employee newsletters from the past. Using OCR, Knowledge Agent extracts the publication date, as well as products and employee benefits mentioned in each PDF.
A document library with Microsoft newsletters from the past. Using OCR, Knowledge Agent extracts the publication date, as well as products and employee benefits mentioned.As new files are added to the document library, Knowledge Agent automatically populates metadata for those documents – no need to re-run the agent.
Editing column prompts:
If you want to adjust the output, simply select “Edit” to revise the agent’s autofill prompt and then test before applying changes. Once metadata is applied, you can save different views to see your data in a variety of ways.
Know when to apply manual columns:
Apply manual columns when you need to add information to your content that only you know – and that AI can’t discern without your input.
Managing Views:
When you organize documents with Knowledge Agent, changes apply to the whole library. To avoid altering the default view, save your changes as a “new view.” This lets everyone access your view as a selectable pill at the top, without overriding the default. You can also ask Knowledge Agent to sort, group, and filter data, and then save that as a separate view.
_________________________________________________________________________________
Get started with Knowledge Agent in SharePoint today with a Microsoft 365 Copilot license in Public Preview. Learn more about Knowledge Agent.
IT administrators can opt in the Knowledge Agent Public Preview by following the steps outlined in this article. Starting in November, individual sites will also be able to opt into the preview, enabling a more scoped and flexible rollout for your organization.
SharePoint at Ignite
Join Us at Microsoft Ignite
Don’t forget to register for Microsoft Ignite! Join our sessions on SharePoint AI and OneDrive, or stop by the SharePoint and OneDrive booth to:
- See Knowledge Agent & Copilot in OneDrive in action.
- Learn how to scale agents across your organization.
- Get hands-on with new SharePoint and Copilot features.
- Engage with our product experts
Save the Date: Microsoft Ignite, November 18, 2025
Stay tuned for more updates and don’t forget to check out the SharePoint Showcase blog for the latest insights and guidance.
Microsoft