Skip to main content

An Aggie's Guide to Your First Copilot Studio Agent

Audience: IT staff, developers, content creators new to Copilot Studio

Purpose: Build your first custom AI agent step-by-step

Quick Reference​

TL;DR
  1. Create agent — Use conversational setup or manual configuration; focus on the Instructions field
  2. Add knowledge — Upload documents or point to public websites
  3. Test and tune — Refine knowledge sources and create specific Topics

Overview​

Copilot Studio is a toolbox for building conversational agents (copilots) without coding expertise. These agents can:

  • Answer questions based on provided information
  • Perform tasks through integrations
  • Interact with users in natural language

Understanding Environments​

Everything in Copilot Studio lives in a Power Platform Environment:

Environment TypeUse Case
DefaultInitial learning and experimentation
DevelopmentBuilding and testing agents
ProductionLive deployment for users

Agents can interact with data sources and Power Automate flows within the same environment.

Prerequisites​

RequirementDetails
LicenseCopilot for M365 or Copilot Studio license
Environment AccessPermission to create agents (may require admin approval)
Knowledge SourcesPublic websites or documents (Word, PDF, TXT) ready

Procedure​

Step 1: Create Your Copilot Agent
  1. Open Copilot Studio in your browser
  2. Verify you're in the correct environment (top-right corner)
  3. Click Create in the left navigation
  4. Click + New agent

Choose Your Setup Path​

Option A: Conversational Setup (Default)

Chat with the assistant to describe:

  • What your agent should do
  • Its purpose and tone
  • Initial website URLs

Option B: Manual Configuration

Click Skip to configure and fill in:

FieldDescription
NameClear name for your agent
InstructionsCritical — Define behavior, rules, persona, limitations
KnowledgeInitial URLs or documents
LanguageConfirm language setting

Create the Agent​

  1. (Optional) Click ... → Edit advanced settings for Solution and Schema Name
  2. Click Create
  3. You'll land on the Agent Overview page
For Beginners

Advanced settings can be ignored — defaults work fine for getting started.

Step 2: Add Knowledge Sources

Knowledge sources power your agent's ability to answer questions.

Access Knowledge Settings​

  1. On the Agent Overview page, find the Knowledge section
  2. Click + Add knowledge

Adding Websites​

  1. Select Public websites
  2. Enter the URL of a publicly accessible website
URL Crawling Rules
  • Prefer directory URLs (e.g., .../storage/) over file URLs (e.g., .../storage-intro.html)
  • Depth limit — Crawler reads pages up to 2 levels deep from your URL
  • No authentication — Cannot access login-protected pages
  • Redirects excluded — If URL redirects to different site, content won't be included
  1. Add descriptive Name and Description
  2. Click Add

Uploading Documents​

  1. Select Upload a document
  2. Add your files

Supported formats:

  • Word (.docx)
  • Excel (.xlsx)
  • PowerPoint (.pptx)
  • PDF (.pdf)
  • Text (.txt, .md)
  1. Add descriptive Name and Description
  2. Click Add
Technical Requirement

Using uploaded files requires Dataverse search enabled in your environment.

Step 3: Understand and Use Topics

Topics are conversation scripts that provide controlled responses.

  1. Click Topics in the top navigation
  2. View pre-built topics like Greeting and Escalate

Key Topics​

TopicPurpose
GreetingInitial user interaction
EscalateHand-off to human support
Conversational BoostingUses Knowledge Sources when no topic matches

When to Create Custom Topics​

  • Common questions needing specific responses
  • Guided workflows with multiple steps
  • Actions requiring specific data collection
  • Integration triggers (Power Automate, etc.)
Step 4: Test Your Agent

Open Test Panel​

  1. Click Test your copilot (usually bottom-right)
  2. The test panel opens alongside your editor

Testing Best Practices​

PracticeDescription
Test frequentlyAfter each knowledge or topic change
Vary questionsUse different phrasings
Check citationsVerify sources are being used correctly
Note failuresIdentify gaps in knowledge

Refine Based on Testing​

  • Poor answers → Improve knowledge source quality or coverage
  • Wrong topic triggered → Adjust topic trigger phrases
  • Missing information → Add more knowledge sources
Step 5: Tune Agent Performance

Refine Instructions​

The Instructions field is critical. Include:

  • Agent's persona and tone
  • What it should and shouldn't do
  • Limitations and boundaries
  • How to handle unknown questions

Adjust Generative AI Settings​

Access through agent settings:

  • Response length and detail level
  • Citation preferences
  • Fallback behavior

Create Specific Topics​

For common questions or workflows:

  1. Create a new topic
  2. Add trigger phrases
  3. Design the conversation flow
  4. Test thoroughly

Tips for Success​

TipDescription
Start SimpleBegin with minimal knowledge, add gradually
Quality Over QuantityBetter knowledge sources = better answers
Clear InstructionsBe specific about agent behavior and limitations
Test Like a UserAsk questions the way real users would
Iterate ContinuouslyImprovement is ongoing

Common Issues​

IssueResolution
Agent gives wrong answersCheck knowledge source accuracy; add specific topic
Agent says "I don't know"Add more relevant knowledge sources
Agent hallucinatesImprove Instructions with explicit limitations
Website content missingVerify URL is accessible; check crawl depth