You ask Copilot in Word to draft a report, a proposal, or an email, and it stops mid-sentence or mid-paragraph. The generated text cuts off abruptly, leaving you with an incomplete draft. This truncation happens because Copilot has a fixed output token limit per generation, and it also respects the context window of the current document. This article explains the technical limits that cause truncation, shows you how to extend the generated output, and lists other scenarios where Copilot stops writing early.
Key Takeaways: Preventing Copilot Truncation in Word
- Copilot output token limit: Copilot stops generating after about 2,000 tokens per request, which equals roughly 1,500 words of plain text.
- Prompt splitting: Break your request into smaller, sequential prompts to get longer drafts without hitting the token cap.
- Document context window: Copilot reads only the last 8,000 tokens of the document, so very long files cause the model to lose earlier context and stop early.
Why Copilot in Word Stops Generating Mid-Draft
Copilot is built on a large language model that processes text in units called tokens. A token is roughly four characters in English, so one word equals about 1.3 tokens. Each Copilot generation request has a maximum output token limit. In Word, that limit is approximately 2,000 tokens. Once the model reaches that boundary, it stops generating and returns whatever it has produced so far.
Token Limit Per Request
The 2,000-token limit applies to each individual prompt you send. If you ask Copilot to write a 5,000-word chapter, the model will stop at roughly 1,500 words. This is not a bug. It is a design constraint that keeps response times low and prevents the model from generating overly long outputs that are hard to edit. The limit is the same in Word, Excel, PowerPoint, and Outlook.
Context Window of the Document
Copilot also reads the content of your current document to understand the context. The model can only process a limited number of tokens from the document itself. In Word, the context window is about 8,000 tokens. If your document already contains 7,000 tokens of text, Copilot has only 1,000 tokens left for the generated response. This causes truncation even when the output limit has not been reached.
Prompt Length Impact
Your own prompt consumes tokens from the context window. A long, detailed prompt with sample text, bullet points, or formatting instructions reduces the space available for the generated draft. For example, a 500-token prompt leaves only 1,500 tokens for the output. The effective output limit shrinks as your prompt grows.
Steps to Get Longer Drafts From Copilot in Word
Method 1: Split Your Request Into Sequential Prompts
Instead of asking for the entire document at once, break the request into logical sections. This keeps each prompt under the token limit and lets you build the full draft piece by piece.
- Prompt for the first section
Type a request for the first part of the document. For example: “Write the executive summary for a quarterly sales report. Include revenue totals and key growth drivers.” Copilot generates up to 2,000 tokens for this section. - Review and accept the output
Read the generated text. Click the Keep button to insert it into the document. The text you keep becomes part of the document context for the next prompt. - Prompt for the next section
Place your cursor after the inserted text. Type a new prompt for the following section. Example: “Now write the regional breakdown for North America and Europe. Use bullet points for each region.” Each sequential prompt resets the output token counter, so you can generate another 2,000 tokens. - Repeat until the draft is complete
Continue with prompts for each subsequent section. This method produces a full draft without hitting the per-request limit.
Method 2: Reduce the Prompt Length
Shorten your prompt to free up tokens for the generated output. Remove examples, sample text, and formatting instructions from your prompt. Keep only the essential instructions.
- Remove sample text from the prompt
Instead of saying “Write a three-paragraph introduction like this: [sample text]”, say “Write a three-paragraph introduction.” The model does not need the sample to understand the request. - Avoid listing formatting rules
Do not include instructions like “Use bold headers, italic for quotes, and numbered lists.” These consume tokens without improving the draft. Add formatting manually after generation. - Use single-sentence prompts
Keep each prompt to one or two sentences. Longer prompts reduce the output budget.
Method 3: Start With a Blank Document
If your current document is long, Copilot may truncate because the context window is nearly full. Start a new blank document for the Copilot generation, then copy the result into your original file.
- Open a new blank document
Press Ctrl+N to create a new empty document in Word. - Prompt Copilot in the blank document
Type your request. Because the document has zero tokens, Copilot can use the full 8,000-token context window for your prompt and the generated output. - Copy the generated text
After Copilot finishes, select the text and press Ctrl+C. Switch back to your original document and paste it with Ctrl+V.
If Copilot Still Truncates After the Main Fix
Copilot Returns Only One Sentence or a Few Words
This usually happens when the prompt is too long and consumes most of the context window. Check your prompt length. If your prompt is longer than 500 tokens, reduce it to 100 tokens or fewer. Also verify that your document is not spilling over the 8,000-token context window. Use a blank document to test.
Copilot Generates a Draft That Ends With an Incomplete Sentence
The model may stop mid-sentence when it hits the exact token limit. This is normal behavior. To fix it, place your cursor at the end of the truncated text and type a new prompt that says “Continue from where you left off.” Copilot will generate the next 2,000 tokens, completing the sentence and adding more content.
Copilot Ignores the “Write a Full Page” Instruction
Copilot cannot follow length instructions that exceed its token limit. Telling it to write 3,000 words will not work. Instead, use the split-prompt method described above. Generate each section separately and combine them. The model will produce a total draft that matches your requested length, but only across multiple prompts.
Truncation Happens in the Middle of a Table or List
Tables and lists consume more tokens per line because of formatting characters. A table with five columns and ten rows can use 500 tokens or more. If Copilot truncates inside a table, reduce the table size in your prompt or ask for plain text instead. Format the table manually after generation.
Copilot Truncation vs Manual Drafting: Output Comparison
| Item | Single Prompt | Sequential Prompts |
|---|---|---|
| Maximum output per request | ~2,000 tokens (1,500 words) | Unlimited across multiple requests |
| Context window used | 8,000 tokens total including document and prompt | Each prompt resets context, so full 8,000 tokens per request |
| Prompt length impact | Long prompts reduce available output tokens | Short prompts per section maximize output per request |
| Document length impact | Long documents reduce available context for generation | Blank document or short document gives full context window |
| Typical use case | Short paragraphs, emails, summaries | Long reports, proposals, chapters |
Copilot truncation in Word is not a failure of the tool. It is a predictable constraint caused by the token limit and the context window. By splitting your request into separate prompts, reducing prompt length, and starting with a blank document, you can generate full-length drafts without interruption. When truncation still occurs, use the Continue prompt to extend the output. These techniques let you work within Copilot’s limits and produce complete documents.