The short version: AI can draft almost any letter, resignation, complaint, reference, cover, condolence, in under two minutes, but only if you feed it specifics first and edit it hard afterwards. The mistake most people make is asking for the letter before giving the facts, which is how you end up with three paragraphs of confident nothing. Do it in the right order and you’ll cut a 45-minute writing job down to about ten.
Why people get this wrong before they even open ChatGPT
I get asked this most weeks by clients: “can you just get AI to write this letter for me?” The answer is yes, and also, most of them are doing it backwards. They open the chat box, type “write me a letter of complaint to my landlord” and hit enter, then wonder why the result reads like it was written by nobody about nothing.
The tool isn’t the problem. The order of operations is. AI doesn’t know your landlord ignored three emails about the boiler since November, or that you’ve lived there four years and never missed a payment. It doesn’t know that unless you tell it. So the first rule of getting a good letter out of AI is simple and unglamorous: dump the facts in first, ask for the letter second.
The exact process, step by step
Here’s what I do, using a real example from last month. I had to write to a supplier we’d used for two years to end the contract, without burning the relationship, because we might need them again in six months.
- Step 1: Write the facts as a messy list, not prose. I typed: “Ending contract with [supplier]. Reason: pricing went up 40% in January without notice. Been with them 2 years, no complaints before this. Need 30 days notice per contract clause 6. Want to leave door open, might work with them again. Tone: firm but friendly, not cold.”
- Step 2: Ask for a draft, not a final letter. I typed “give me a first draft” not “write my final letter”. That one word, draft, changes the tone AI produces. It writes looser and more human when it thinks it’s a starting point rather than a finished product.
- Step 3: Set a length. I asked for under 150 words. Left open, AI will happily produce 400 words of padding because more feels more helpful to the model. It isn’t. A landlord, a supplier, an ex-boss, none of them want a wall of text.
- Step 4: Read it out loud. This is the step almost everyone skips. If a sentence makes you wince when you say it, it needs cutting. AI-drafted letters often have a phrase or two that sound like a corporate memo, “I trust this finds you well” is the classic one. Cut it.
- Step 5: Change one thing that makes it yours. A specific date, a specific name, a small detail only you would know. This is what stops the letter reading like it came from a template, because in a sense it did.
Total time from blank page to sent email: seven minutes. The letter itself, once I stripped the AI padding, ran to 118 words.
What AI is brilliant at
Reference letters. This is the one I recommend to almost every business owner I coach. You know the facts, dates, role, achievements, but staring at a blank page trying to find the right adjectives eats an hour you don’t have. Feed AI the bullet points, ask for a draft, and you’ll have something usable in ninety seconds that you then tighten for ten minutes.
Formal complaint letters also work well, because the structure is predictable: state the issue, state what you want, state the deadline. AI is good at structure. It’s your job to add the temperature, the bit that makes it clear you’re a real person who’s had enough, not a bot filling in a form.
Cover letters and application letters are the trickiest, and I’ll be honest with you about why, because most advice on this topic skips it. Recruiters now run applications through AI detectors, and some hiring teams have started penalising letters that read as obviously machine-written, not because AI use is banned but because a letter that sounds identical to five hundred others tells them nothing about you. LinkedIn’s own 2024 data showed a sharp rise in AI-assisted applications, and recruiters have noticed the sameness. If you’re using AI for a cover letter, the draft is the easy 60%. The other 40%, the specific project you led, the number you hit, the reason you want this job, has to come from you or the letter does you no favours at all.
The part nobody wants to say out loud
Here’s the uncomfortable bit. Every AI model is trained on roughly the same slice of the internet, which means every AI-drafted letter starts from the same well of phrasing. “I hope this email finds you well.” “I am writing to formally request.” “Please do not hesitate to reach out.” If you’ve read one AI-drafted resignation letter, you’ve read most of them, because the model reaches for the safest, most average phrasing every single time. That’s what it’s built to do.
This matters more than people admit. A reference letter that sounds exactly like every other reference letter doesn’t help the person you’re writing it for. A resignation letter that reads like a template tells your manager you couldn’t be bothered to write it yourself, even if you spent twenty minutes on the prompt. The fix isn’t to avoid AI. It’s to treat the AI draft as scaffolding, not the finished building. I’ve started asking clients a blunt question when they show me an AI-drafted letter: “would your closest colleague know this was you if I removed the name?” If the answer is no, it needs more editing, not a different AI.
A short story about getting this wrong
Two years ago I asked AI to draft a letter declining a speaking invitation, one I wanted to keep the relationship warm on because the organiser had booked me twice before. I pasted in barely any context, just “write me a polite decline to a speaking invitation.” What came back was technically fine and completely forgettable, three paragraphs of “unfortunately,” “regretfully,” and “I do hope you understand,” with nothing in it that told the organiser I valued the relationship. I sent it anyway because I was rushing between calls.
She replied a week later, polite but noticeably cooler than usual. I hadn’t mentioned the two events we’d done before, hadn’t referenced anything specific, and the letter read like a form rejection because that’s essentially what I’d let it become. I rewrote the next one, one line about the event we’d run in 2023, one honest sentence about workload, and a clear “ask me again next year.” She booked me again eight months later. Same tool, same task, completely different outcome, because the second time I gave it something to work with.
Prompts that get you somewhere
Skip “write me a letter about X.” Use this shape instead:
- Who it’s to and your relationship with them (new client, old friend, former employer)
- What happened, in plain facts, no adjectives
- What you want as the outcome
- The tone (firm, warm, formal, apologetic)
- A word limit
Then, once you’ve got a draft, ask it to “make paragraph two shorter” or “cut the last sentence, it’s too formal” rather than regenerating the whole thing. Editing in small pieces gets you a better letter faster than starting over each time you don’t like a word.
Where this connects to everything else you write
If you’re drafting letters this way, you’re probably drafting other things the same way without noticing, your newsletter intro, your LinkedIn posts, your client emails. The same rule applies across all of it: AI gives you the structure, you give it the specifics that make it sound like a person wrote it. I’ve written before about what a good newsletter should look like, and the failure pattern is identical, competent but forgettable copy because nobody added the one detail only they knew. If you’ve never sent one and you’re wondering whether you even can start with AI’s help, this piece on whether you need an existing list to start a newsletter is worth ten minutes.
I went further with this last year and built an AI system to draft my own newsletter for me, and the first version failed badly for the exact reason letters fail: it had structure and no specifics. I wrote up what happened, mistakes included, in I Built an AI Agent to Run My Newsletter Without Me, and most of the fixes apply directly to letter-writing too.
When you shouldn’t let AI near it
Employment tribunal letters, anything with legal weight attached, formal grievances, contract terminations with financial penalties attached. Use AI to get your thoughts in order, then get a solicitor or your HR advisor to check the actual wording before it goes anywhere. A badly worded AI-drafted letter with legal consequences has cost people real money. This isn’t a reason to avoid AI for letters generally, it’s a reason to know where the line sits.
Condolence letters sit in an odd middle ground. AI can help you find the words when your brain has gone blank with grief, and that’s a legitimate use. What it can’t do is know the story about the person, so use it for structure and put the memory in yourself. A condolence letter with zero specific detail about the person who died reads as cold no matter how well it’s phrased.
The numbers that matter here
From tracking my own time on this over the last few months: a letter I’d normally spend 30 to 45 minutes drafting from scratch takes about 4 minutes to get a workable AI draft, then 10 to 15 minutes of editing to make it sound like me. That’s roughly a 60% time saving, and the final result, if I’ve done the editing, is often better than what I’d have written unaided, because I’m editing rather than staring at a blank page fighting for the first sentence.
Frequently asked questions
Can AI write a legally binding letter?
AI can draft the wording, but it can’t confirm the letter meets the legal requirements for your situation or jurisdiction. For anything with legal consequences, a resignation with notice disputes, a formal grievance, a contract termination with penalty clauses, get a solicitor or HR professional to check the final wording before you send it.
How do I stop an AI-drafted letter sounding generic?
Feed it specific facts, names, dates, and one concrete detail before you ask for the draft, then add or change at least one sentence yourself after. The generic feel comes from AI reaching for the safest average phrasing, which only you can break by putting something specific back in.
Is it obvious when a letter was written by AI?
Often, yes, particularly to people who read a lot of them, recruiters, HR managers, editors. Repeated phrases like “I hope this finds you well” or “I trust you understand” are giveaways. A heavy edit pass that removes the stock phrasing and adds one specific personal detail usually fixes this.
What’s the fastest way to get a usable draft from AI?
List your facts as short bullet points before you ask for anything, then request “a first draft” with a word limit rather than an open-ended request. This one change, giving context before asking for output, is the single biggest factor in getting a usable draft on the first try.
