How to Write a Letter With AI: Emails, Cover Letters and Formal Requests

A language model can turn a handful of bullet points into a tidy, well-structured letter in seconds. What it cannot do is know your facts, your relationship with the reader or what you are prepared to promise. Writing a letter with AI therefore works best as a three-part job: you supply the substance, the tool drafts, and you check and finish.

Step by step

1. Decide the outcome. Write one sentence on what the reader should do after reading: pay, reply, approve, attend, accept a new date.
2. Gather the facts. Names, dates, amounts, reference numbers and deadlines. Give these to the tool yourself; never let it guess.
3. Pick the format and tone. Formal letter or email, warm or firm, brief or detailed.
4. Write the prompt. Include the reader, the purpose, the facts and the format. The framework in how to write AI prompts fits letters well.
5. Review the draft. Check every fact against your notes and every sentence against what you are willing to stand behind.
6. Make it yours. Swap generic phrases for your own wording and add one specific detail only you would know.
7. Proofread the final version. Names, titles, attachments and the subject line deserve a last look.

A sample prompt for a routine reminder might read: "Draft a polite but firm payment reminder to a client. Invoice [number] for [amount] was due on [date] and is now three weeks overdue. Mention that bank details are on the invoice, ask for payment or a reply within seven days and keep it under 150 words. Use placeholders for anything I have not given you."

What to tell the tool for common letters

For payment reminders, the reference details should match the original document exactly; our guide on how to write an invoice lists the fields worth repeating.

Using AI for a cover letter without sounding generic

Cover letters are where AI drafts most often fall flat, because the tool defaults to safe, interchangeable phrases. A few habits help:

• Paste the job advertisement and a short list of your genuine achievements, then ask the tool to match three requirements to three pieces of your experience.
• Ask for plain verbs and no stock adjectives such as "passionate" or "dynamic".
• Never let it invent results, job titles, qualifications or numbers. If it adds something you did not provide, delete it.
• Read the final text aloud. If it does not sound like you speaking at your best, rewrite those lines.

Beyond letters: everyday writing with AI

The same approach helps with most routine writing at work:

• Turning rough notes into a first outline for a report or proposal
• Shortening a long email while keeping the key request intact
• Rewriting technical text for a reader without a specialist background
• Offering several subject lines or opening sentences to choose from
• Producing a rough translation to check before a fluent speaker reviews it

What to keep out of the prompt

Letters often involve personal data, account details or confidential terms. Before pasting, check what your organization allows and replace identifying details with placeholders where you can. The comparison in offline AI chatbots explained shows when a local tool is the safer choice.

Mistakes to avoid

• Sending the draft unedited. Readers notice stiff, generic wording, and any error carries your name, not the tool's.
• Letting it fill gaps. Ask it to use placeholders such as [date] for anything you have not supplied.
• Over-formal language. Ask for the tone you would use in person with that reader.
• Relying on it for legal letters. Anything with legal or financial consequences, such as a formal dispute, a contract change or a dismissal, should be checked by a qualified professional before it is sent.

Questions

Is it acceptable to send a letter drafted with AI?

In most workplaces, yes, provided you have checked it and it says what you mean. Some settings, such as certain job applications, schools and regulated industries, have their own rules on AI assistance, so look them up before relying on it.

Will the reader be able to tell?

Sometimes, from the tone rather than any technical test. Detection software is less dependable than many people assume, and how AI text detection works explains why. Editing the draft into your own voice matters more than any tool.

Which tone should I ask for?

Describe the relationship rather than an abstract style: "a supplier we have worked with for years" or "a hiring manager I have never met" gives the model more to work with than "professional".