You can use AI to write better work emails by treating it as a drafting and editing partner: give it your goal, the reader and the key facts, let it produce a first draft, then review and send it yourself. It is good at structure, tone and summarising long threads. It is not responsible for your facts, your commitments or your company's confidential information — you are.
Where AI helps most with email
Email is a good fit for AI because most messages follow a pattern: a purpose, some context, a request and a next step. Common uses:
- First drafts. Turn a few bullet points into a clear message, so you start from something instead of a blank screen.
- Tone adjustment. Make a blunt message more polite, or a long-winded one more direct. This helps when writing to a senior stakeholder, a customer, or someone whose first language is not English.
- Shortening. Cut a rambling draft down to the points the reader needs.
- Summarising threads. Condense a long back-and-forth into decisions made, open questions and who owns what.
- Translation and rewording. Draft in the language you are most comfortable in and ask for a version in the language your reader prefers, then have someone fluent check it.
- Replies. Paste the message you received (see the data warning below) and ask for two or three reply options with different levels of firmness.
How to ask: a simple prompt pattern
Vague requests give generic emails. Give the AI the same briefing you would give a capable colleague:
- Who is the reader? A new customer, your manager, a supplier who is late.
- What is the goal? Get approval, confirm a date, say no politely.
- What facts must be included? Dates, quantities, names, the exact request.
- What tone and length? Warm but brief; formal; three short paragraphs.
- What should it avoid? Promising anything you have not agreed, or adding facts you did not supply.
For a deeper look at writing instructions that work, see how to write a good prompt for Claude. If an answer is not right, reply with a specific correction ("shorter, remove the apology, keep the deadline") rather than starting again.
What never to paste into an AI tool
This is the part people skip. Before pasting an email thread, check what is in it. Do not paste material your company would not want leaving its controlled systems unless your organisation has approved the tool and the use:
- Customer personal data such as IC numbers, addresses, phone numbers or account details.
- Passwords, access links, API keys or anything that grants access.
- Unannounced financial results, pricing under negotiation, or contract terms marked confidential.
- Employee records, performance reviews or disciplinary matters.
A practical habit is to replace names and figures with placeholders ("Customer A", "[amount]") before pasting, then fill in the real details yourself afterwards. Our article on whether you can put customer data into Claude covers this in more depth. Your company's own AI-use policy comes first; if you are unsure, ask your IT, security or compliance team.
A review checklist before you hit send
AI drafts read smoothly, which makes mistakes easy to miss. Before sending, check:
- Facts. Every date, number, name and spelling is correct. AI can add details you never gave it.
- Commitments. The email does not promise a discount, deadline or outcome you have not approved.
- Tone. It sounds like you. Stiff, over-polished wording can feel odd to colleagues who know how you write.
- The ask. The reader can tell in the first few lines what you want them to do and by when.
- Recipients. The right people are in To, Cc and Bcc, especially after pasting in a thread.
Build a habit, not a one-off trick
The teams that get the most from AI do not just try it once. They keep a few reusable prompts for recurring emails — follow-ups, meeting recaps, polite refusals, status updates — and agree simple rules on what may and may not be pasted in. That turns scattered experimentation into a consistent, safe practice. If you want your team to build that skill with guidance, Agmo Studio, a Select Partner of the Anthropic Claude Partner Network, runs hands-on training; see AI training in Malaysia for the options.