AI for Work: Using AI for writing and rewriting

Using AI for Writing and Rewriting

What This Is

AI for writing and rewriting means using large language models (LLMs) to draft, edit, refine, or adapt text—from emails and reports to marketing copy and technical documentation. It matters in everyday work because it saves time, reduces writer’s block, and improves clarity without sacrificing professionalism. Example: A product manager uses AI to rewrite a dense internal memo into a concise, actionable update for executives, cutting 500 words to 200 while keeping key insights.

Key Facts & Principles

• Prompt engineering: The art of crafting clear, specific instructions to guide the AI. A vague prompt (“Write a report”) yields generic output; a precise one (“Draft a 300-word quarterly update for investors, focusing on revenue growth and risks, in a confident but cautious tone”) delivers usable results.
• Tone matching: AI can mimic formal, casual, technical, or persuasive styles. Example: Ask the AI to rewrite a customer email in a “friendly but professional” tone to match your brand voice.
• Iterative refinement: AI output is a starting point, not a final product. Use follow-up prompts (“Make this more concise”, “Add a call to action”) to polish the text.
• Context window: The amount of text an AI can process at once (e.g., 4K–128K tokens). Exceeding it causes the AI to “forget” earlier parts of the conversation. Workaround: Break long documents into chunks or summarize key points first.
• Bias and sensitivity: AI may default to stereotypes or insensitive language. Mitigate: Explicitly instruct it to avoid bias (“Write this job description to appeal to diverse candidates”) and review output critically.
• Plagiarism risk: AI-generated text can resemble existing content. Solution: Use AI for ideas and structure, then rewrite in your own words or cite sources if needed.
• Fact-checking: AI can hallucinate details (e.g., incorrect stats, fake citations). Rule: Always verify facts, especially for external-facing content.
• Version control: AI edits can drift from your original intent. Tip: Save drafts with clear labels (e.g., v1_rough_draft, v2_client_ready) to track changes.

Step-by-Step Application

1. Define the goal
2. Ask: What’s the purpose of this text? (Inform? Persuade? Summarize?)

3. Example: “I need a 1-page proposal to convince my boss to approve a new tool. The goal is to highlight cost savings and efficiency gains.”

4. Gather context

5. Provide the AI with:
   • Audience (e.g., executives, customers, team members).
   • Key points to include (e.g., data, examples, constraints).
   • Tone (e.g., “professional but enthusiastic”).

6. Prompt template: “Write a [type of document] for [audience] about [topic]. Include [key points]. Use a [tone] tone. Keep it under [word count].”

7. Generate a first draft

8. Use a simple prompt to get a baseline: “Draft a proposal for a new project management tool. Focus on how it will save 10 hours/week for the team and reduce costs by 15%. Write for my manager, who cares about ROI and team morale. Keep it to 300 words.”

9. Pro tip: For complex tasks, break it into smaller prompts (e.g., “First, list 3 key benefits. Then, draft a paragraph for each.”).

10. Refine with follow-up prompts

11. Edit the output by asking for specific changes:

    • “Make the introduction more compelling.”
    • “Add a bullet-point summary of the ROI.”
    • “Shorten this by 20%.”
    • “Rewrite this in a more conversational tone.”

12. Fact-check and humanize

13. Verify all data, names, and claims.
14. Replace generic phrases with your voice (e.g., “We’re excited to…”-“I’m confident this will…”).

15. Example: If the AI writes “Studies show…”, replace it with “According to a 2023 Gartner report…” (and link the source).

16. Test and iterate

17. Share the draft with a colleague for feedback.
18. Use AI to generate alternatives (e.g., “Give me 3 versions of the conclusion”) and pick the best one.

Common Mistakes

• Mistake: Treating AI output as final. Correction: Always review and edit. AI is a co-writer, not a replacement. Why: It may miss nuance, tone, or context (e.g., inside jokes, company culture).

• Mistake: Overloading the prompt with too much detail. Correction: Start simple, then refine. Why: Long prompts can confuse the AI or lead to repetitive output. Example: Instead of “Write a 500-word email about Q3 goals, including revenue targets, team updates, and risks, in a formal tone…”, try “Draft a Q3 goals email. Focus on revenue targets and risks. Formal tone.”

• Mistake: Ignoring tone for the audience. Correction: Specify tone in the prompt. Why: An email to a client should sound different from one to your team. Example: “Rewrite this for a customer who’s frustrated. Be empathetic and solution-focused.”

• Mistake: Not setting length constraints. Correction: Always include word/page limits. Why: AI tends to be verbose. Example: “Summarize this 2,000-word report in 300 words.”

• Mistake: Assuming AI understands implicit context. Correction: Provide background. Why: AI doesn’t know your company’s history or unspoken rules. Example: “This is for our internal wiki. Our team values brevity and bullet points.”

Practical Tips

• Use templates for common tasks:
• Keep a folder of prompt templates for emails, reports, and social media posts. Example: “Write a LinkedIn post announcing [product]. Highlight [key benefit]. Keep it under 150 words. Use a [tone] tone.”
• Combine AI with other tools:
• Use AI for drafting, then tools like Grammarly or Hemingway for polishing.
• Batch similar tasks:
• Generate multiple versions of a document (e.g., 3 email subject lines) in one session to save time.
• Track what works:
• Note which prompts yield the best results for future reference (e.g., “Prompt X worked well for executive summaries”).

Quick Practice Scenario

Scenario: You’re a sales rep who needs to follow up with a prospect who went dark after a demo. You want to re-engage them without being pushy. Write a short email using AI.

Question: What prompt would you use to generate this email?

Answer: “Draft a follow-up email to a prospect who didn’t respond after a demo. Remind them of the key benefit (saving 10 hours/week on reporting) and ask if they’d like to discuss next steps. Keep it short, friendly, and low-pressure. Use a subject line that stands out.”

Explanation: The prompt specifies the goal, key details, tone, and constraints to avoid generic output.

Last-Minute Cram Sheet

1. Prompt engineering = Clear instructions + context + constraints.
2. Tone matching = Specify audience and style (e.g., “casual but professional”).
3. Iterative refinement = Use follow-up prompts to edit (e.g., “Make this shorter”).
4. Context window = AI “forgets” if input is too long; chunk long docs.
5. Hallucination = AI makes up facts; always verify.
6. Plagiarism risk = Rewrite AI output in your own words.
7. Bias = Explicitly instruct AI to avoid stereotypes.
8. Version control = Save drafts with labels (e.g., v1_rough).
9. Mistake: Treating AI output as final. Fix: Always edit.
10. Mistake: Overloading prompts. Fix: Start simple, then refine.