You have a complaint to file, a contract to cancel, or a letter to your landlord to write — and you've been putting it off for two weeks. Not because you don't know what you want to say, but because you don't know how to say it without getting the tone or the wording wrong. I've been there. Since then, I use AI for these letters. The result: what used to take me an hour now takes ten minutes, and the final text is cleaner than anything I'd have written on my own.
Admin letters have a special talent: they paralyze us. Not because the situation is complicated. Because the format is.
A complaint to your phone carrier, a cancellation notice to your insurer, a letter to your landlord who keeps dragging their feet on fixing the boiler — in every one of these cases, you know exactly what you want to say. But the moment you open a blank document, you start wondering whether to write "Dear Sir" or "To whom it may concern," whether you need to cite a specific law, whether your tone sounds too aggressive or not firm enough.
According to a 2024 European survey by Visma, 55% of small businesses spend more than 40% of their time on administrative tasks. And a study by Shakked Noy and Whitney Zhang (2023) measured a 40% drop in time spent on professional writing tasks when people use AI — with output quality rated 18% higher.
That's not nothing. Those letters we keep pushing to tomorrow end up costing us: an unclaimed refund, a botched cancellation, a dispute that drags on.
AI doesn't solve the legal problem. But it solves the blank-page problem. And that's already huge.
You don't need a specialized tool. The major general-purpose models handle admin letters just fine.
ChatGPT is still the most accessible. Since May 2026, GPT-5.5 Instant is the default model, and the GPT-5.6 family (launched July 9, 2026) cut hallucinations by 52.5% on high-stakes prompts. For a complaint letter, that's reassuring: less risk of AI inventing a clause or a legal reference that doesn't exist.
Claude (Anthropic) is known for the quality of its prose. Claude Sonnet 5 dropped on June 30, 2026, and Claude Opus 5 on July 24, 2026. What I like about Claude: it produces text that feels less "robotic" than some competitors, which helps when the tone needs to stay human without getting emotional.
Google Gemini launched Gemini 3.6 Flash and 3.5 Flash-Lite on July 21, 2026, optimized for efficiency. If you're already in the Google Workspace ecosystem, it's a smooth option.
Mistral AI, a French company, offers Mistral Medium 3.5 (April 2026) and Mistral Large 3 (December 2025). The upside: a European option, with potentially clearer data usage terms for professional use.
To help you choose between these tools based on your needs, I wrote a dedicated article: how to choose your AI for writing.
For simple admin letters (complaints, cancellations), the free version of ChatGPT or Claude is more than enough. Upgrade to a paid plan if you're handling high volumes or want to keep an organized history of your letters.
This is where everything happens. A vague prompt gives you a vague letter. A precise prompt gives you something ready to send.
Here's the structure I use every time:
1. The type of letter. Tell the AI exactly what you need: "a complaint letter," "a cancellation notice," "a letter to my landlord."
2. The factual context. Who's writing to whom, why, and since when. No need for a novel — three sentences is plenty.
3. The tone you want. "Neutral and professional," "firm but polite," "formal." AI naturally tends to be too enthusiastic or too casual in admin letters. If you don't specify the tone, you risk getting something that sounds off.
4. What you expect as output. A complete letter? A rough draft to refine? Two different versions to compare?
A concrete example. Instead of writing "write a letter to my insurer," try:
"Write a formal complaint letter to my home insurer. Context: I filed a water damage claim on [date]. After three weeks with no response, I'm following up in writing. Tone: neutral, factual, firm. Format: complete letter with subject line, body, and standard closing."
You get something usable in thirty seconds. The AI structures it, phrases it, and picks the right closing for you.
For a deeper dive into building a good prompt, I have a full article on the topic: writing a good prompt when you're not a developer.
The complaint. This is where AI helps the most. When you write while you're annoyed, it shows — and it weakens your case. AI rephrases the facts in a neutral, structured way without losing the substance of your request. You supply the dates, amounts, and reference numbers. It builds the letter.
The cancellation. The wording is codified. AI knows it. You give it the contract name, the desired end date, and any clause you're invoking — it produces a compliant letter. Always double-check the notice periods in your actual contract before sending: AI doesn't know the specific terms of your agreement.
The letter to your insurer. This type of letter often requires citing specific references (policy number, date of incident, contractual clause). AI structures this kind of letter very well. Give it the exact references — it places them in the right spots.
The letter to your landlord. The tone here is tricky: you want to stay cordial while being clear about your rights. That's exactly the kind of nuance you can steer in your prompt by writing "firm on substance, polite in tone." AI handles that tension better than most of us would under pressure.
Don't ask AI to cite specific legal articles without verifying them yourself. Models can hallucinate legal references — meaning they can invent articles that don't exist or attribute them to the wrong code. For more on this specific risk, check out the glossary entry on hallucinations. If you need a legal reference, verify it from an official source before including it.
On tone. Admin letters need to be neutral and objective. By default, AI tends to produce phrasing that's too warm or too enthusiastic. Always specify in your prompt that you want a "strictly formal" or "standard administrative" tone. Then read the output with that lens.
On privacy. This is the point I care about most. Never paste your sensitive information as-is into a public tool: social security number, bank account details, full contract number, full address. Use placeholders in your prompt ("my contract number: XXXX," "my bank details: to be filled in") and add the real information in the final document, outside the tool. Data you send to a public AI tool can potentially be used as training data. I have a full article on this: personal data and AI — what you need to know.
On proofreading. AI produces a quality draft, not a certified document. Always read through the generated letter with two questions in mind: are all the facts accurate? Does the tone match what I actually want to convey? Two minutes of proofreading prevents a lot of headaches.
The techniques here apply to other forms of written communication too. If you want to use them for your everyday emails, I covered that case in the article on writing your emails with AI.
There are situations where AI writes well, but the letter alone isn't enough.
If you're in a dispute with a landlord who refuses to make repairs, if your insurer is contesting a claim on the merits, if you've received a formal notice — that's when you need human advice. A lawyer, a consumer association, a mediator. AI can help you articulate your position clearly, but it can't assess your actual legal situation.
Similarly, if you need to sign a document or take on legal responsibility, have someone who knows the subject review it. AI is a writing tool, not a legal advisor.
To understand what these tools actually do under the hood — and therefore better grasp their limits — you can check out the glossary entry on LLMs.
It writes quickly and well on the formal side. But it can get things wrong when it comes to facts you didn't provide, or it might invent a legal reference. Always read through the result before sending, and verify the factual details (dates, amounts, references) yourself.
ChatGPT (free version with GPT-5.5 Instant as the default model since May 2026) and Claude (Anthropic) are the most accessible. Both let you write quality letters without a paid subscription, within daily usage limits.
Never enter your sensitive data directly into the tool. Use placeholders in your prompt (XXXX for a contract number, [to be filled in] for an address) and add the real information in the final document. For a deeper dive on this topic, the article on privacy with AI is exactly what you need.
Yes, that's one of its strengths. You specify the tone you want in your prompt — "formal and neutral," "friendly but firm," "strictly administrative" — and it adjusts. If the first result doesn't feel right, rephrase the tone instruction and ask for a new version.
Yes. There's no legal restriction on using AI to write letters, including administrative documents. You remain the author and signatory of the document. The responsibility for the content is yours — AI is a writing tool, not a party to the contract.
Ask the AI itself to review it: "Are there any technical terms or jargon in this letter? If so, suggest simpler phrasing." You can also have someone unfamiliar with the topic read it — if they understand it on the first read, the letter is good.
A few years ago, I never would have guessed that admin letters would be one of the AI uses I'd reach for most often. And yet.
It's not flashy. It's not what you see in demos. But it's useful, concrete, and it removes a real friction from everyday life. A complaint letter I would have put off for ten days, I now write in ten minutes — and it's better than anything I'd have written alone under pressure.
The key is staying in the driver's seat. AI proposes, you decide. You supply the facts, it shapes them. You proofread, you tweak if needed, and you sign. That's it.
I test AI for real and share what actually works, no jargon, no hype. If this article was useful, the easiest way to stay in the loop is my Friday letter. And if you have a question or a doubt: write back — I read everything.