You want Copilot to produce logical, accurate answers, not random or incomplete responses. Chain-of-thought prompting is a technique where you break a complex request into a sequence of reasoning steps. This forces Copilot to process each part before moving to the next, reducing errors and improving output quality. This article explains how chain-of-thought prompting works, shows step-by-step examples for Microsoft 365 apps, and highlights common mistakes to avoid.
Key Takeaways: Chain-of-Thought Prompting with Copilot
- Copilot chat pane > Text prompt with numbered steps: Forces Copilot to process one reasoning step at a time instead of guessing the final answer.
- Word > Copilot > Draft with numbered instructions: Produces structured outlines, reports, or data analyses that follow your logical sequence.
- Excel > Copilot > Step-by-step formula builder: Reduces formula errors by having Copilot explain each calculation stage before writing the final formula.
What Chain-of-Thought Prompting Is and Why It Works
Chain-of-thought prompting is a method where you write your request as a series of logical steps rather than a single command. Instead of asking Copilot to write a summary directly, you ask it to first list the key points, then group them by theme, then write a concluding paragraph. This sequential approach mirrors how humans solve complex problems.
The technique works because Copilot has a limited context window. When you give a vague or compound request, Copilot may skip intermediate reasoning and produce an answer that misses details or contradicts itself. By structuring your prompt into distinct steps, you keep Copilot focused on each sub-task. Each step builds on the previous output, creating a coherent final result.
Chain-of-thought prompting is especially useful in three scenarios: data analysis in Excel, document drafting in Word, and multi-condition logic in Copilot chat. In each case, the step-by-step format reduces ambiguity and increases the accuracy of the output.
Prerequisites for Using Chain-of-Thought Prompting
You need a Copilot license that includes Copilot for Microsoft 365. This feature works in Copilot chat, Word, Excel, PowerPoint, and Outlook. No additional plugins or settings are required. The technique relies entirely on how you phrase your prompt.
Step-by-Step Examples for Copilot Chat
The following examples show how to apply chain-of-thought prompting in Copilot chat. Each example includes a weak prompt and a strong chain-of-thought prompt. Use the strong version to get better results.
Example 1: Summarize a Long Email Thread
- Weak prompt
“Summarize this email thread.” Copilot may produce a generic summary that misses deadlines or action items. - Chain-of-thought prompt
“Step 1: Identify all questions asked in this email thread. Step 2: List all deadlines mentioned. Step 3: List all decisions made. Step 4: Write a summary that includes only the questions, deadlines, and decisions.” - Result
Copilot outputs a structured list first, then combines them into a concise summary. No action items are lost.
Example 2: Compare Two Product Options
- Weak prompt
“Compare Product A and Product B.” Copilot may produce a short paragraph without clear categories. - Chain-of-thought prompt
“Step 1: List the price of Product A and Product B. Step 2: List the features of Product A and Product B. Step 3: List the customer ratings for both. Step 4: Create a table with columns Price, Features, and Ratings. Step 5: Write a recommendation based on the table.” - Result
Copilot generates a table first, then writes a recommendation. The output is organized and verifiable.
Step-by-Step Examples for Word
In Word, you can use chain-of-thought prompting in the Copilot draft pane. This method works well for creating structured reports, proposals, or meeting notes.
Example 3: Draft a Project Status Report
- Open Copilot in Word
Click the Copilot icon in the ribbon. The draft pane opens on the right side. - Write the chain-of-thought prompt
Type: “Step 1: List the project name and current phase. Step 2: List three completed milestones this month. Step 3: List two risks and their mitigation plans. Step 4: Write a one-paragraph status update that includes all items from steps 1 through 3.” - Review and refine
Copilot generates a draft. If a section is missing, ask Copilot to expand only that step. For example, “Expand step 3 with more detail on the first risk.”
Step-by-Step Examples for Excel
In Excel, chain-of-thought prompting helps build complex formulas and data summaries. The step-by-step approach reduces formula errors.
Example 4: Build a Conditional Sum Formula
- Open Copilot in Excel
Click the Copilot icon in the ribbon. The pane opens on the right. - Describe the data structure
Type: “My data has columns A (Region), B (Product), C (Sales). Rows 2 to 100 contain data.” - Write the chain-of-thought prompt
Type: “Step 1: Identify all unique regions in column A. Step 2: For each region, sum the sales in column C where the product in column B equals ‘Widget’. Step 3: Write a SUMIFS formula that performs the calculation described in step 2. Step 4: Explain what each argument in the formula does.” - Result
Copilot returns a SUMIFS formula and a plain-English explanation. You can copy the formula directly into your worksheet.
Common Mistakes and How to Avoid Them
Copilot Ignores the Step Order
If Copilot skips a step or reorders them, your prompt may be too long or contain contradictory instructions. Keep each step to one sentence. Do not combine two actions in one step. For example, do not write “List the features and then compare them.” Split that into two steps.
Copilot Returns a Single Paragraph Instead of Steps
Copilot may ignore your step structure and output a continuous paragraph. This happens when the prompt does not explicitly request separate sections. Add a final instruction: “Output each step as a separate numbered section.”
Copilot Repeats the Same Information Across Steps
If Copilot repeats data, your steps may overlap in scope. Review your steps and ensure each one covers a unique sub-task. For example, do not have step 1 list all features and step 2 list features again. Merge or remove the duplicate step.
Copilot Produces an Incomplete Result
A very long chain-of-thought prompt may exceed Copilot’s context limit. If Copilot stops mid-output, split your request into two separate conversations. Use the first conversation to generate the intermediate data, then start a new conversation with that data as input.
Chain-of-Thought Prompting vs Standard Prompting: Key Differences
| Item | Standard Prompting | Chain-of-Thought Prompting |
|---|---|---|
| Prompt structure | Single request or command | Numbered list of logical steps |
| Output format | Single paragraph or list | Separate sections following the step order |
| Error rate for complex tasks | High — missing details or contradictions | Low — each step is processed independently |
| Best use case | Simple questions or one-line commands | Multi-condition analysis, reports, formulas |
| Time to write the prompt | Less than 10 seconds | 30 to 60 seconds |
| Reusability | Low — each prompt is unique | High — you can save the step template for similar tasks |
You can now use chain-of-thought prompting in Copilot chat, Word, and Excel to get structured and accurate outputs. Start by saving a template of steps for your most common task, such as report drafting or data analysis. For advanced use, combine chain-of-thought prompting with Copilot’s “Explain this” feature to verify each step before proceeding to the next.