How to Edit Test Cases in Microsoft Copilot Studio

In Microsoft Copilot Studio, you can edit questions, multi-turn conversations, reference responses, keywords, and expected tool use inside a test set. Save each case with Apply, then save the whole test set or run Evaluate. Use reviewed, realistic cases so a passing score reflects behavior that matters to users.

Edit a single-response test case

1. Open the agent and go to Evaluation.
2. Select the relevant test set.
3. Select the test case you want to change.
4. Edit the text in the Question field.
5. Complete any fields required by the test methods, such as expected answer, keywords, pass score, or expected capability.
6. Select Apply.
7. When all cases are ready, select Save or Evaluate.

Configure the expected outcome

Each evaluation method needs different case data. Do not paste the same expected answer into every field.

Write reference answers around the facts and actions that must be present. Avoid unnecessary wording that could penalize a correct paraphrase. Reserve exact match for deterministic responses such as a fixed code or tightly formatted value.

Edit a conversation test case

Conversation evaluations test behavior across multiple turns. They are useful for clarification, context retention, follow-up questions, and workflows where the agent must gather information before acting.

1. Open the test set from the Evaluation page and select the conversation.
2. In the right pane, select the edit icon.
3. Select a user message or agent reference response to change it.
4. Optionally select Add a user question to extend the scenario.
5. Select Done to close the conversation editor.
6. Complete any fields required by the selected test methods.
7. Select Apply, then save the test set or run the evaluation.

Keep each turn purposeful. If a reference response assumes information the user has not supplied, add a turn in which the agent asks for it. Also include a case where the user changes a detail or corrects an earlier statement.

Add more cases to an existing test set

1. Open the test set from the agent's Evaluation page.
2. Select Add questions for single-response tests or Add conversations for multi-turn tests.
3. Choose an available source: write a case manually, generate additional cases, import supported data, or use another option shown in the current interface.
4. Open each new case and add the required expected outcomes.
5. Apply the case changes and save the test set.

What makes a useful test case?

• One clear purpose: each case should reveal a specific success or failure.
• Real language: preserve spelling, abbreviations, and ambiguity seen in actual user requests.
• Verified expectation: a subject-matter owner should approve important reference answers.
• Permission awareness: the expected result must match the selected user profile's access.
• Stable scoring: do not require exact wording when meaning is what matters.
• Negative coverage: include unsupported, unsafe, and out-of-scope requests that should be declined.

Avoid common test-set mistakes

• Editing a case but forgetting to select Apply.
• Applying case changes but forgetting to save the test set.
• Using AI-generated cases without checking their facts or expected answers.
• Adding many near-duplicate questions that inflate the pass rate.
• Changing both the agent and test expectations merely to make a failure disappear.
• Comparing runs made with different accounts, connections, or test cases as if they were equivalent.

When an agent changes, keep a stable regression set and add new cases for the failure that motivated the change. Run the same saved set before and after the update. Microsoft's current test-case editing guide documents the interface and supported methods. For set-level settings, see our guide to editing a Copilot Studio test set.