How to Set Up and Configure the SharePoint Knowledge Agent Step by Step

The SharePoint Knowledge Agent is one of the most powerful additions to Microsoft 365 in recent years. Built on the same Copilot foundation that powers Microsoft's broader AI strategy, the Knowledge Agent transforms ordinary SharePoint sites into intelligent, conversational knowledge hubs. Instead of forcing users to dig through nested document libraries, click through pages, or guess at search keywords, the Knowledge Agent allows them to simply ask questions in natural language and receive grounded, source-cited answers drawn directly from the content they already have permission to see.

For administrators and IT professionals, however, the magic only happens after a careful setup. The agent has to be licensed correctly, enabled at the tenant level, scoped to the right sites, and tuned so that it surfaces accurate, trustworthy information without leaking sensitive data. This tutorial walks you through the entire process, from prerequisites and licensing to advanced customisation and governance, so you can confidently roll the Knowledge Agent out across your organisation.

Understanding What the Knowledge Agent Actually Does

Before touching any settings, it helps to be clear about what you are configuring. The Knowledge Agent is a site-level conversational assistant that indexes the pages, documents, lists, and metadata within a SharePoint site (or collection of sites) and uses a large language model to answer user questions grounded in that content. It respects existing SharePoint permissions, which means a user querying the agent will only ever receive answers based on files and pages they are already entitled to read. It cites its sources, so users can verify the original document, and it supports follow-up questions, summarisation, comparison, and extraction tasks across multiple documents at once.

This makes the agent fundamentally different from classic SharePoint search. Search returns links; the agent returns reasoned answers. That distinction also means the quality of your configuration, especially around content cleanliness and permissions, directly determines the quality of the answers users receive.

Prerequisites and Licensing

Setting up the Knowledge Agent begins outside of SharePoint itself. You will need a Microsoft 365 tenant with SharePoint Online active, and at minimum one Microsoft 365 Copilot license assigned to each user who will interact with the agent. Administrative actions require a Global Administrator or SharePoint Administrator role, and customisation of agents is typically performed by site owners with Copilot licenses of their own.

You should also confirm that your tenant is running a supported region and that Microsoft 365 Copilot has been provisioned. If your organisation uses sensitivity labels, Microsoft Purview, or Data Loss Prevention policies, these should already be in place before the agent is enabled, because the agent honours them automatically and you do not want to surface content that has not yet been properly classified.

Finally, ensure that the SharePoint sites you intend to target are healthy. Broken permissions, orphaned pages, outdated documents, and duplicated files all degrade the quality of the agent's responses. A short content audit before rollout will save you considerable troubleshooting later.

Enabling the Knowledge Agent at the Tenant Level

Sign in to the Microsoft 365 admin center using a Global Administrator account and navigate to the Copilot settings area. From there, open the section dedicated to agents and Copilot extensibility. You will see a control that governs whether agents can be created and used across SharePoint sites in your tenant. By default in most tenants this is enabled, but in regulated environments it may be turned off pending review.

Switch the setting to allow agents tenant-wide, then move to the SharePoint admin center. Under the settings menu, locate the Copilot and agent configuration page. Here you can decide whether all users, no users, or a specific security group can create and edit agents. For an initial pilot, restrict creation rights to a controlled security group containing your champions, knowledge managers, and a handful of site owners. This prevents an immediate sprawl of half-configured agents while you learn how your users want to work with the technology.

While you are in the SharePoint admin center, review the Intelligent Services settings as well. The Knowledge Agent depends on the underlying semantic index and on Microsoft Graph connectors if you plan to extend its reach beyond native SharePoint content. Make sure both are enabled and that the semantic index has had time to build, which can take from a few hours to a couple of days depending on the size of your tenant.

Deploying the Default Agent to a Site

Every SharePoint site that meets the prerequisites receives a default Knowledge Agent automatically once the tenant setting is on. To verify this, browse to any modern SharePoint site as a site owner. In the top right of the site, you will see a Copilot icon that opens the agent pane. The default agent is scoped to the current site, uses the site's existing content as its knowledge source, and inherits the site's permission model.

Try a few sample questions to confirm the agent is responsive. Ask it to summarise the most recent news posts, list the policies stored in a particular library, or compare two documents. If you receive answers with citations, the agent is functioning correctly. If you receive a message that the agent is still indexing, give it more time, especially on larger sites.

At this stage, communicate to your pilot users that the default agent is available, but resist the temptation to roll it out broadly. The real value comes from customised agents tailored to specific business scenarios.

Creating a Custom Knowledge Agent

To create a tailored agent, open the SharePoint site where the agent will live. In the Copilot pane, choose the option to create a new agent, or from the site's pages library, create a new agent file directly. SharePoint will open the agent builder, a no-code interface where you can define the agent's identity and behaviour.

Give the agent a clear, descriptive name that reflects its purpose, such as HR Policy Assistant or Engineering Onboarding Guide. Write a short description so users understand what kinds of questions it can answer. The description also serves as a hint to the underlying model, so be specific. Phrases like "Answers questions about employee benefits, leave policies, and onboarding procedures based on documents in the HR Policies library" produce better results than generic statements.

Next, define the knowledge sources. This is the most important configuration step. You can point the agent at the entire current site, at specific document libraries, at individual folders, at lists, or even at particular files. You can also include other SharePoint sites if your user account has access to them. The narrower and more focused the scope, the more accurate the answers will be. An agent pointed at one curated policy library will outperform an agent pointed at an entire intranet, every time.

Then craft the instructions, sometimes called the system prompt. This is where you tell the agent how to behave, what tone to use, what to avoid, and how to handle ambiguity. A useful instruction might read: "You are an assistant for the HR team. Always answer in a friendly but professional tone. Cite the source document for every factual claim. If you cannot find a clear answer in the provided documents, say so and suggest the user contact HR directly. Do not speculate about legal questions."

Finally, configure the starter prompts. These are the suggested questions that appear when a user opens the agent for the first time. Good starter prompts demonstrate the agent's strengths and steer users towards questions it can answer well. Three to five well-chosen prompts are usually enough.

Save the agent and test it thoroughly with a variety of questions, including edge cases and questions the agent should refuse to answer. Iterate on the instructions and the knowledge scope until the behaviour is reliable.

Sharing and Permissions

Once the agent works as intended, share it with the people who need it. Sharing an agent in SharePoint works much like sharing any other file. You can grant access to specific users, security groups, or the entire site. Remember, however, that the agent itself does not bypass content permissions. If you share the agent with a user who does not have access to the underlying documents, the agent will simply not return answers based on those documents for that user.

This is a feature, not a limitation. It means you can confidently share a single agent with a broad audience knowing that each person will only see answers based on their own access rights. It also means that if users are getting incomplete answers, the fix may lie in adjusting SharePoint permissions rather than in changing the agent.

For organisations using Microsoft Entra ID groups, prefer group-based sharing over individual sharing. It scales better, integrates with joiner-mover-leaver processes, and makes audits straightforward.

Customising Behaviour with Advanced Instructions

Beyond the basics, the agent builder allows for considerably deeper customisation. You can instruct the agent to follow a particular response format, such as always producing bullet-pointed summaries, always including a section labelled Next Steps, or always responding in a specific language regardless of the input. You can also instruct it to handle particular phrases in particular ways, for example treating questions about salary as requests to contact the HR business partner rather than as factual queries.

A particularly powerful pattern is to define personas. You can tell the agent to behave as a friendly onboarding buddy for new hires, a precise technical reference for engineers, or a cautious legal assistant that always recommends professional review. Combined with carefully scoped knowledge sources, personas make a single SharePoint tenant feel like a portfolio of distinct, purpose-built assistants.

For multilingual organisations, you can instruct the agent to detect the user's language and respond in kind, while still drawing from English-language source documents. This works remarkably well for global rollouts and reduces the need to translate every document up front.

Governance, Monitoring, and Lifecycle Management

A Knowledge Agent is not a deploy-and-forget asset. The content it draws from changes constantly, and the agent's quality changes with it. Establish a governance routine from the start. Site owners should review their agents at least quarterly, checking that the knowledge sources are still relevant, that outdated documents have been archived, and that the instructions still reflect business reality.

From the Microsoft 365 admin center and the SharePoint admin center, you can monitor agent usage through the Copilot dashboard and the usage reports. These show how often agents are invoked, who is using them, and which sites generate the most queries. Pair this with user feedback, both informal and via short surveys, to identify agents that are underperforming or that need to be retired.

Pay attention to sensitive content. Even though the agent respects permissions, a misclassified document in an over-shared library can become an embarrassment when it suddenly surfaces in an answer. Use Microsoft Purview sensitivity labels, DLP policies, and the SharePoint Advanced Management features to keep your content posture clean. Consider running a Restricted SharePoint Search configuration during pilots so that the agent only sees a curated allowlist of sites until you are confident in your overall content hygiene.

When an agent is no longer needed, delete it. Stale agents pointing at stale content are worse than no agent at all because they erode user trust in the technology.

Troubleshooting Common Issues

If users report that the agent returns no answers, the most common cause is that the underlying content has not yet been indexed, or that the user lacks permission to the relevant documents. Check the semantic index status and the user's effective permissions before assuming the agent itself is at fault.

If answers are inaccurate or hallucinated, the cause is almost always in the knowledge scope or the instructions. Narrow the scope, remove duplicate or contradictory documents, and tighten the instruction to require source citations and discourage speculation.

If the agent is slow, verify that you are not pointing it at an enormous, sprawling content surface. Agents perform best on focused, well-curated knowledge sources. Splitting one giant agent into several smaller, purpose-built agents almost always improves both speed and accuracy.

If creation of new agents fails for some users, verify their Copilot license assignment and confirm that the SharePoint admin center setting allows their security group to create agents.

Closing Thoughts

The SharePoint Knowledge Agent is one of those rare features where the technology is genuinely impressive but the value it delivers depends almost entirely on the discipline of the people configuring it. A thoughtful administrator who curates knowledge sources, writes precise instructions, manages permissions carefully, and reviews agents on a regular cadence will give their users a tool that feels close to magical. A careless rollout, by contrast, will produce confusing, inconsistent answers that quickly erode trust.

Take the time to start small, pilot with a handful of well-scoped agents, gather feedback, and expand deliberately. Treat your agents as living products rather than one-off deployments. Do that, and the Knowledge Agent will become one of the most valuable pieces of the Microsoft 365 stack your organisation has ever rolled out.

Source: How to Set Up and Configure the SharePoint Knowledge Agent Step by Step