Write Emails With AI: Draft, Check, and Keep Your Voice

To write an email with AI, explain who will read it, what you want them to do, and which facts the message must preserve. Ask for a draft, check its meaning and details, then send the version you approve from your email app.

You can keep reusable email prompts and approved communication rules in fenbs, so each connected assistant starts with the same instructions. The message still needs your judgment about the recipient and the situation.

What are the steps from notes to a sendable email?

1. Write one sentence describing the purpose: confirm a meeting, ask for an answer, explain a delay, or share a decision.
2. List the verified facts, including dates, names, what has already happened, and any commitment you are authorized to make.
3. Tell the assistant the recipient’s relationship to you, the desired tone, and the approximate length.
4. Ask for a subject line and draft, with unanswered questions listed separately.
5. Review the facts, tone, recipient, links, and attachments before sending the approved message yourself.

Say you need a client to choose between two finish samples. The purpose is to obtain that choice. Include the sample names and the real response deadline; leave out an unrelated history of the project unless it helps the client act.

If you are still choosing a drafting tool, see AI email tools for work. Here the job is to make the writing and review process dependable in whichever approved tool you use.

How much background should you share?

Supply the context that changes the message. A longtime client who already knows the project needs different wording from a new contact. An internal reminder may need only a task name and date, while an explanation of a delay needs the confirmed cause and what happens next.

Use information your organization permits in the chosen service. You can often replace a person’s name with a role or summarize the relevant exchange instead of pasting a whole thread. Keep private details out when they do not affect the draft.

Mark uncertainty directly. “The replacement delivery date is unconfirmed” helps the assistant preserve the gap. “Make this reassuring” on its own can lead to a confident promise that nobody has agreed to.

The same approach applies when writing a task for an AI assistant: specify the result, provide the facts, and explain what needs a person’s answer.

What does a useful email prompt look like?

Draft an email to an existing client asking them to choose a cabinet finish.
Purpose: get one selection so the project can continue.
Confirmed facts: the options are Natural Oak and Painted White; the sample images will be attached; a reply is requested by Thursday.
Unknown: the installation date is not confirmed.
Tone: friendly, concise, and direct.
Return a subject line and a message under 120 words.
Ask for one clear action. Do not invent an installation promise or imply the attachments have already been sent.
List any missing information outside the draft.

The exact wording is an example, so replace the details with your own confirmed information. The useful pattern is purpose, facts, uncertainty, tone, and requested output. Reusing that structure saves you from reconstructing the assignment every time.

OpenAI’s prompting guidance↗ (opens in a new tab) recommends supplying relevant context and setting boundaries, including keeping a message as a draft. Those are useful instructions whenever you want writing help before a person takes the final action.

Store the version that works in Skills and prompts in fenbs, with a note identifying the facts you must replace for each new recipient.

How do you keep the email sounding like you?

Give a short example of your own writing that you have permission to share. Explain what to preserve: a direct opening, short paragraphs, a warm closing, or ordinary words. Avoid broad instructions such as “make this impressive,” which do not tell the assistant how you communicate.

Request revisions with a specific purpose. “Put the question first and remove the repeated background” is easier to evaluate than “improve this.” If the draft becomes too abrupt, explain what consideration the recipient needs and ask for a warmer version that keeps the request clear.

The National Archives’ plain-language principles↗ (opens in a new tab) emphasize the reader, the main point, and concise wording. Apply that to email by checking whether the recipient can quickly see why you wrote and what happens next.

For a delayed delivery, for example, explain the confirmed situation and the next update you can actually provide. A long opening full of apologies can obscure the useful information. A short acknowledgment followed by a practical next step often says more.

Which mistakes should you catch before sending?

Ask the assistant to list assumptions, then inspect the draft yourself. An assistant can miss the same invented detail during a second pass. Your review should compare important statements with the actual notes, thread, or decision.

Sending is a separate action with consequences for the recipient. Our guide to whether AI assistants can send emails covers that question; this workflow ends with a draft a person checks and sends.

How to do this with fenbs

1. Create a free account and add the relevant work project. fenbs is free for one person, no credit card.
2. Save your reusable email assignment in Skills and prompts. Describe its purpose and the inputs to replace each time.
3. Record the communication rules a person approved in Decisions and rules. Keep unresolved promises or dates as questions awaiting an answer.
4. Connect a supported assistant through the integration guides, using MCP, a standard way for an AI app to use an outside tool. Ask it to read the project context before drafting.
5. Review and send the message in your email app. Save the task’s progress and next step in fenbs when the exchange changes the work.

fenbs does not send emails, manage your inbox, or import your email history. If you only need a quick rewrite, a writing assistant may be enough. fenbs helps when project rules and reusable instructions need to stay available over several sessions.

What should you save after the reply arrives?

Record the work outcome: the chosen finish, the approved date, or the question still waiting on someone. Keep only the context needed to continue and link to the source where appropriate. A personal exchange should not become a shared project note just because AI helped draft it.

When a repeated correction improves the prompt, update the saved version and explain the change. The next session should begin with that improvement. Browse the AI at work library for other small writing tasks you can approach in the same way.