BW-301f · Module 3
Drafting Correspondence with AI
4 min read
AI-assisted drafting is genuinely useful for professional correspondence. The draft that takes AI thirty seconds would take a human five minutes — and the AI draft is often structurally sound, well-organized, and free of the throat-clearing that afflicts first drafts. The professional who uses AI to generate a first draft and then edits it with judgment is not cutting corners. They are allocating their effort to where it adds the most value: the judgment layer, not the generation layer.
- Provide context, not just a task The quality of an AI-drafted email depends entirely on the quality of the input. "Draft an email declining a vendor proposal" produces generic boilerplate. "Draft an email declining a software vendor proposal — we appreciate the relationship, the pricing was competitive, but we selected a different vendor for reasons related to integration complexity. Keep the tone warm and leave the door open for future consideration. Audience: VP of Sales at a firm we may want to work with in two years" produces something worth editing. Context is the investment. The draft is the return.
- Use AI for structure, your voice for the edit AI-generated correspondence often has a voice that is competent but slightly generic — it sounds like a professional, not like you. The first pass through the draft should tune the voice: replace phrases that you would never use with your actual phrasing, add the specific context only you know, and remove the hedges and softeners that AI often inserts to avoid sounding too direct. The edited draft should read as if you wrote it. It just did not take as long.
- Always read the full draft before sending AI drafts contain errors — of fact, of tone, of interpretation — that appear in otherwise well-structured text. The professional who sends an AI draft unread has delegated a judgment function to a system that cannot exercise judgment. Read the full draft. Not skim — read. The error you miss because you skimmed the AI output is your error, not the AI's.