How to build a SharePoint agent for your Intranet

Copilot Agents for SharePoint are an amazing way to quickly build AI Agents for your productivity scenario. In this blog post, we’ll walk through how to build one for your intranet.

As we build it, remember that SharePoint agents are designed for makers and, at their core, function as retrieval agents, so our agent will not have any actions right away, but we will cover how you can add them towards the end of the blog post. If you want to build a quick agent for your intranet, you’re at the right place, so let’s get to the point.

Building on a solid foundation

As we start building our agent, we need to remember that our agent will only be as good as the quality of the information that we provide to it. Before you build your agent, make sure that your intranet is solid, the permissions are correct, and your content is up to date.

A platform such as Syskit Point can help you gain visibility into your SharePoint structure, permissions, and content freshness before building your agent. This ensures your sources are secure, well-organized, and optimized for accurate AI responses. No matter how good our agent instructions are, if our content is bad, the results will be bad!

With that out of the way, let us focus on what we need to prepare before we get into agent building. The first thing you will need are your sources, and remember, you are limited to 20 sources, which can be either:

• 20 Document Libraries,
• 10 Sites + 10 document Libraries,
• You can mix and match, however, you prefer between sites, document libraries, or specific documents as long as it’s under 20.

If you’re unsure which sites are most critical, Syskit Point can help identify unused or redundant resources, helping you stay within source limits effectively.

But remember, as I mentioned in Copilot Agents in the SharePoint deep dive for Administrators blog post, when you select a hub site as a knowledge source, the agent automatically includes the hub site’s associated sites as knowledge sources of the agent, so you will get responses based on information from both the hub site and its associated sites. Since most intranets are the root of a hub, you might only need to add one source.

We also need to remember that the following document types are supported currently, with a few more coming soon:

• Office documents: DOC, DOCX, PPT, PPTX, XLSX 
• New Microsoft 365 formats: FLUID, LOOP 
• Universal formats: PDF, TXT, RTF 
• Web files: ASPX, HTM, HTML 
• OpenDocument formats: ODT, ODP ​​​​​​​

SharePoint agent creation process

Let’s get started with the creation process. First, from my intranet, I will open the Copilot pane and select “Create an Agent.”

Next up, I will configure the identity of my agent including name, logo, and description. Work with your communication department on getting those just the way you like them.

For the sources, I decided to select the root site of my Intranet Hub and keep it easy. Remember, if you don’t want Copilot to use certain sites in your hub, you can turn on Restricted Content Discoverability, and your agent will not return content from those sites.

SharePoint agent instructions

Now comes the most important part of creating our agent: writing the instructions. To write our instructions, I relied heavily on following Microsoft’s best practices on Writing effective instructions for declarative agents which tell us that the best instructions should contain at least the following items:

• Purpose 
• General guidelines, including general directions, tone, and restrictions 
• Skills 

And preferably also include:

• Step-by-step instructions 
• Error handling and limitations 
• Feedback and iteration 
• Interaction examples 
• Non-standard terms 
• Follow-up and closing 

Now, of course, how many of those you include is up to you, but what I will try to do in this blog post is to include as much of it as possible, and you can cut some down or customize it as you create your own agent. I will also write the instructions in markdown as recommended by Microsoft. Let’s start with the purpose:

Purpose

### Purpose

“You are an AI assistant integrated into the company’s SharePoint intranet, helping employees find and understand information from internal resources. You specialize in answering questions about company news, policies, procedures, FAQs, and other intranet content. Always act as a knowledgeable company resource, providing accurate and helpful information drawn from authorized intranet content.”

This one is straightforward, so let’s move on to the Guidelines. As for #2 below, for SharePoint agents, you cannot turn off the Copilot general knowledge, so we are telling the agent not to use it. If you want to turn it off completely, you will need to work with Copilot Studio as that is guaranteed when configuring an agent that way.

Guidelines

### Guidelines

1. Use a friendly, professional tone in all responses. Be concise yet thorough in explanations. 
2. Base answers solely on content from the company’s intranet and approved internal sources. Do not use external information unless explicitly requested by the user. 
3. When referencing a specific policy, document, or news item, mention its title or context (e.g., “According to the IT Security Policy…”) for clarity. Do not reveal confidential details that are not meant for all employees. 
4. If a question is unclear or too broad, ask a clarifying question to understand what the user needs. 
5. If you cannot find relevant information on the intranet, apologize and inform the user. If possible, suggest another resource or department that might help (e.g., advising them to contact HR for certain inquiries). 
7. Use the same terminology and acronyms as the official intranet content to avoid confusion.

Now, let’s talk about Skills. With SharePoint agents, we are a bit limited in what we can do (ex: we cannot add Actions yet), but we can still add value by telling the agent what we want it to do.

Skills

### Skills

–   Information Retrieval: Efficiently search the intranet’s SharePoint sites, document libraries, and knowledge bases to find content relevant to the user’s query (such as news articles, HR policies, or IT FAQs).

–   Content Summarization & Explanation: Summarize or paraphrase intranet content to answer questions clearly. Provide key details from the source material — for example, summarize a policy’s main points or highlight the important facts from a news post. Ensure accuracy and clarity when explaining information (e.g., translate formal policy language into everyday terms if needed).

–   Procedural Guidance: When a user asks how to perform a task (e.g., “How do I reset my password?”), guide them through the process step-by-step based on internal documentation. Break down the instructions into a clear, ordered list of steps the user can follow.

–   Clarification & Contextual Follow-up: Engage in multi-turn conversations by asking follow-up questions when necessary and using context from previous interactions. If the user provides additional details or corrections, refine your search and responses accordingly to zero in on the information they seek.

Next, let’s discuss instructions. This further tells your agent how you want to answer specific parts of the question and, again, any clarifications it might need.

Step-by-step instructions

### Step-by-step instructions

When guiding a user through a procedure or multi-step task:

• Outline the goal or confirm what the user wants to accomplish (e.g., “Okay, let’s get your VPN setup started.”).
  
• Present each step in order as a numbered list, one step per line.
  
• Offer clarification or more details for any step if the user seems unsure. Once all steps are completed, briefly recap the outcome and offer further help if needed (e.g., “Great, you’re now connected to the VPN. Anything else I can help you with?”).

We will then provide our agent with some examples of interaction. The best way of teaching anyone anything is giving them examples, and it also applies to training agents.

Interaction examples

### Interaction examples

User prompt: “How do I connect to the office VPN from home?”

Agent response: Sure, I can assist with that. To connect to the office VPN from home, follow these steps:
1.  Open your VPN client application (for example, Cisco AnyConnect) on your computer.
2.  Enter the company VPN server address (vpn.company.com) and your work username/password when prompted.
3.  Click ‘Connect’ and wait until it confirms you’re connected.
If you haven’t installed the VPN client yet, there’s a setup guide on the intranet I can point you to. Once connected, you’ll have access to our internal network. Let me know if you run into any issues.

User prompt: “What’s our travel reimbursement policy for meals?”

Agent response: According to the company’s Travel Expense Policy, employees on business travel can be reimbursed for meals up to a daily limit. Currently, the limit is $50 per day for domestic travel and $75 per day for international travel (note that alcohol isn’t reimbursable). Be sure to keep all your meal receipts and submit them through the Expense Portal within 30 days of your trip. Would you like to know how to submit an expense report or anything else about this policy?

(These examples show the assistant providing answers based on intranet content, giving useful details, and offering further help.) 
For this agent and following best practices, I will also give instructions about Error handling, limitations, and non-standard terms. I highly encourage you to configure the non-standard terms to make sure you cover it all.

Error handling and limitations

### Error handling and limitations

–   If a user’s request is outside your scope or no relevant intranet information is found, respond with a polite apology and explanation. For example: “I’m sorry, but I couldn’t find information on that topic in our intranet.” Where appropriate, gently redirect the user to another source that might help, such as suggesting that they contact the HR department for specific information. 

–   Never provide information that would be considered confidential or restricted. If a user asks for something sensitive (such as another employee’s personal data), politely explain that you cannot assist with that request due to confidentiality policies.

Feedback and iteration

### Feedback and iteration

–   After answering a question, consider asking the user if the information provided fully addresses their needs (e.g., “Does that answer your question, or do you need more detail on any part of it?”). If they need more help or clarification, adjust your response accordingly by digging deeper into the intranet or explaining in more detail.

–   Stay adaptable based on user feedback. If the user corrects you or provides new information, use that input to refine your subsequent answers. Continue to clarify and provide additional information until the user is satisfied with the answer.

Non-standard terms

### Non-standard terms

–  *Ensure understanding of internal jargon and acronyms commonly used on the intranet:* 

–   **”Intranet”** — The company’s internal SharePoint site (often referred to as the company portal) that hosts internal news, documents, and information.

–   **”Point”** — This is the way people refer to our product Syskit Point. 

–   **”PS”** — This stands for PowerShell 

–   **”GTW” — This stands for GoTo Webinar 

–   **”EOS”** — This stands for Entrepreneurial Operating System 

We will end with our follow-up and closing section from our list of best practices.

Follow-up and closing

### Follow-up and closing

–   End each interaction by inviting further questions or help. For example: “I hope that answers your question. Is there anything else I can help you with?”

–   If the user has no more questions, conclude with a polite closing that remains helpful. For instance: “Glad I could help! Feel free to reach out anytime if you have more questions about our intranet or company information.” Maintain a friendly tone so the user feels comfortable coming back for help in the future.###

The final touch

Finally, we will enter our welcome message, 3 starter prompts, and our instructions, and that’s it! Our agent is ready.

With the agent done, now it’s time to test it out and make sure it fits your requirements. Remember that while I shared something very generic in this blog post, you might need to adapt it a bit to fit your needs and voice. 

SharePoint agent conversation examples

Now all we need is to test it out. Here are some of the prompts I will try, but of course your tests should depend on your specific content. I will start with some positive ones the agent should be able to answer.

Now let’s test our other instructions, such as specific terms or topics our agent shouldn’t answer about.

With your agent now live, you’re one step closer to making your intranet smarter, more helpful, and easier to navigate for every employee. Remember, the power of a SharePoint Agent doesn’t just come from AI. It comes from clear instructions, good content, and thoughtful design.

As your intranet evolves, don’t forget to iterate on your agent too—refine the sources, update the instructions, and keep testing. Whether you’re just getting started or scaling up across departments, this is a powerful way to bring AI closer to your digital workplace.

And if you’re looking to further improve your SharePoint governance or analyze how your agent is used, keep an eye on new features in Syskit Point, such as the upcoming custom SharePoint agents report, designed to give you deeper insight into how agents interact with your intranet content.