You ask Copilot a question and get a decent answer, but you know it could be better. The first response might lack depth, skip important details, or contain minor errors. Self-critique prompting is a technique where you ask Copilot to review and improve its own output before you use it. This article explains what self-critique prompting is, how to set it up in your prompts, and how it produces more accurate and thorough final answers.
Key Takeaways: Self-Critique Prompting in Copilot
- Prompt suffix “Review your answer for errors and suggest improvements before finalizing”: Triggers Copilot to self-check its own response.
- Follow-up prompt “List three weaknesses in your previous answer and rewrite it”: Forces Copilot to identify and correct specific gaps.
- Copilot pane > Compose box > Add critique instruction before pressing Enter: The exact action to apply the technique in real time.
What Self-Critique Prompting Does and Why It Works
Self-critique prompting is a method where you instruct Copilot to evaluate its own output and then produce a revised version. The technique works because large language models can generate multiple candidate responses internally. When you ask for a critique, Copilot simulates a second pass over its first answer, checking for logical consistency, completeness, and factual accuracy.
The root cause of weak first answers is that Copilot optimizes for speed and relevance, not depth. The model produces the most probable token sequence given your prompt. Without a self-critique instruction, Copilot stops after the first pass. Adding a critique step forces a second reasoning cycle, which often catches missing context, ambiguous phrasing, or unsupported claims.
Prerequisites for this technique are minimal. You need Copilot in any Microsoft 365 app: Word, PowerPoint, Excel, Teams, or the Copilot chat on Edge. No special permissions or plugins are required. The method works with Copilot Pro and Copilot for Microsoft 365 licenses. You do not need to enable any experimental features.
How to Write a Self-Critique Prompt
- Ask your original question
Type your main request into the Copilot compose box. For example: “Write a one-page summary of the Q3 sales results for the North America region.” Do not add the critique instruction yet. Press Enter and review the first answer. - Insert a critique instruction in the same conversation
In the compose box, type: “Review your previous answer for errors, omissions, and unclear language. List the issues you find. Then rewrite the answer with those issues fixed.” Press Enter. Copilot will produce a critique and then a new version. - Refine with a targeted weakness prompt
If the second answer still feels shallow, type: “Identify three specific weaknesses in your last answer. For each weakness, explain why it matters. Then produce a final answer that addresses all three.” This forces Copilot to prioritize the most impactful improvements. - Use the rewritten answer as the final version
Copy the final answer into your document or email. You can also continue the conversation by asking for a specific section to be expanded. For example: “Now expand the section on regional trends with more data points.”
You do not need to repeat the critique instruction in every message. Once Copilot is in a critique loop, it will continue to apply self-checking if you ask follow-up questions that reference the critique. To exit the loop, simply start a new conversation.
Common Problems When Using Self-Critique Prompting
Copilot Produces a Critique But Does Not Rewrite the Answer
This happens when the prompt asks for a critique without an explicit rewrite instruction. Copilot treats the critique as the final output. Fix this by adding the phrase “and then rewrite the answer” to your critique prompt. Example: “Review your previous answer for errors and then rewrite it with corrections.”
The Rewritten Answer Is Identical to the First Version
Copilot may judge that no changes are needed. This can occur if the original answer is already well-structured. To force a deeper review, ask for specific improvements: “Rewrite the answer to include at least two data sources and a comparison with the previous quarter.” This gives Copilot a concrete target for revision.
Copilot Introduces New Errors in the Rewritten Version
Self-critique can sometimes cause the model to over-correct. For example, it might remove a correct fact because it misjudges it as ambiguous. If you spot a new error, type: “In your last answer, the statement about X is incorrect. Please correct it and explain why the previous version was wrong.” This directs Copilot to fix only the specific error without reworking the entire answer.
The Critique Loop Consumes Too Many Messages
If you have a message limit per conversation, a self-critique loop can use up your quota quickly. To avoid this, combine the critique and rewrite into a single instruction: “Answer this question, then critique your answer, and finally provide a corrected version — all in one response.” Copilot will output all three parts in a single message.
| Item | Single-Pass Prompt | Self-Critique Prompt |
|---|---|---|
| Number of Copilot passes | One | Two or more |
| Typical response length | Short to medium | Medium to long |
| Factual accuracy | Moderate — first pass only | Higher — second pass catches errors |
| Completeness of details | Often missing context | More thorough with cited gaps |
| Time to generate | Faster | Slightly slower due to extra processing |
| Best use case | Quick drafts, simple lists | Reports, analysis, decision support |
Self-critique prompting is a practical way to improve Copilot’s output without changing your Microsoft 365 settings or switching to a different tool. Start your next conversation with a clear question, then add a critique instruction after the first answer. You will get final answers that are more accurate, better organized, and closer to what you actually need. For advanced use, combine self-critique with a system prompt that defines the tone and format before you begin.