The short version: the ChatGPT prompts that work best for copywriting aren’t clever one-liners, they’re the ones that feed the model your actual customer language, your actual offer, and a clear format to copy, and the prompt matters far less than what you put into it before you hit enter. I’ll give you the exact prompts I use every week, but the uncomfortable bit is that most people fail at this before they’ve even opened the chat window.
Worth reading next: How Do You Find Housewife Friendly Work From Home Jobs in 2026.
Worth reading next: how many prompts can you send chatgpt in a day.
Why “write me a sales email” never works
I get asked for the magic prompt roughly once a week, usually by someone who’s tried ChatGPT for copy, got something bland back, and decided the tool is overhyped. Nine times out of ten, they typed something like “write a sales email for my coaching business” and got exactly what that deserves: generic, adjective-heavy, sounds-like-everyone-else copy.
That’s not a ChatGPT problem. That’s a garbage-in problem. The model has no idea who your customer is, what they’re afraid of, what they’ve already tried, or how you talk. You gave it nothing to work with, so it filled the gaps with the most generic marketing language it’s seen a million times. This is exactly why so many small business websites sound identical now, and it’s why so much AI-written copy has that same flat, corporate hum to it.
The fix isn’t a better prompt. It’s a better brief. Every prompt below assumes you’ve done a bit of homework first, because that’s where the actual copy quality comes from.
The five-minute prep that makes every prompt better
Before I open a chat for copywriting, I do this:
- Paste in 3 to 5 real quotes from customers, reviews, or sales calls, in their own words
- Write one sentence on the specific outcome this piece needs to drive (a click, a reply, a booked call)
- Name the exact competitor or alternative the reader is comparing you against
- Paste one example of writing that sounds like the brand’s actual voice, even a Slack message or an old email
That’s it. Four things, five minutes, and it changes the output more than any prompt phrasing ever will. I learned this the hard way rebuilding my own newsletter list after a quiet few years. My early AI-drafted emails read fine and got a 14 percent open rate, which is average and forgettable. Once I started feeding ChatGPT real subscriber replies and my own past emails as voice samples before asking for a draft, open rates on the same list climbed to 31 percent within six weeks. Same tool, same me, completely different input.
The core prompt structure I use for almost everything
Once the prep is done, I use a version of this every time, adjusted for the format:
“Act as a direct-response copywriter. Here is my offer: [describe it]. Here is my ideal reader: [job, problem, what they’ve tried before]. Here are three real quotes from actual customers: [paste them]. Write a [format] that leads with the reader’s specific problem, not the product. Use short sentences. No hype words like unleash, revolutionary, or game-changing. Match this tone: [paste sample]. Give me three versions with different opening lines.”
Three versions matters more than people realise. The first output from ChatGPT is almost always the safest, most obvious angle. The second and third are where the interesting stuff shows up, because you’ve already used up the predictable option. I never ship the first draft ChatGPT gives me. Not once. Anyone telling you they copy-paste straight from the chat window into a live campaign is either not checking their results or not telling you the truth.
Prompts by format, with real examples
Headlines and subject lines
Prompt: “Write 15 subject lines for this email about [topic]. Mix these angles: curiosity, a specific number, a direct question, and a slightly uncomfortable truth. Keep each under 50 characters. No exclamation marks. No emojis.”
Real example: for a client’s bookkeeping firm sending a year-end tax reminder, ChatGPT’s first pass gave me things like “Don’t Miss Your Tax Deadline!” (deleted immediately). Pushed for the “uncomfortable truth” angle, it produced “Most of your invoices are already wrong” which we used, and it pulled a 42 percent open rate against the firm’s usual 22 percent average. The angle, not the tool, did the work, but the prompt is what surfaced the angle.
Sales pages and landing pages
Prompt: “Write a landing page outline for [offer] using this structure: headline, one-sentence problem statement, three specific consequences of not solving it, the mechanism of my solution, three proof points from these testimonials [paste], objection handling for the two most common hesitations [name them], and a single clear call to action. Write in plain, spoken English, as if explaining it to a smart friend who is busy.”
I ask for the outline first, not the finished page. Full pages written in one pass tend to pad every section evenly, which is how you end up with 400 words of throat-clearing before the actual offer. An outline lets me see the shape and cut the weak sections before any prose gets written.
Cold outreach and LinkedIn messages
Prompt: “Write 5 versions of a first outreach message to [role] at [type of company]. Reference this specific detail about their business: [something real, not generic]. Keep it under 60 words. No ‘I hope this finds you well.’ End with a low-pressure question, not a pitch.”
This is where most people leave money on the table, especially anyone doing freelance client outreach at volume. Generic outreach gets ignored regardless of how well it’s written, because the reader can smell a template from the subject line. The specific detail is doing 80 percent of the work, not the copy structure.
Ad copy
Prompt: “Write 10 Facebook ad variations for [offer]. Use these five hooks: a stat, a question, a myth-bust, a short story opener, and a direct callout of the reader’s situation. Each under 125 words. Match this tone: [sample]. Flag which two you’d test first and explain why.”
Asking it to flag its own top picks and explain the reasoning is underrated. It forces the model to apply actual criteria rather than just generating variety, and the reasoning it gives you is often more useful than the copy itself, because it tells you what pattern it thinks is working.
The contrarian bit nobody wants to say out loud
Here’s what most posts on this exact topic won’t tell you: if your offer is weak or your pricing is confused, no prompt fixes that. I’ve had clients hand me beautifully specific customer quotes, perfect voice samples, all the prep done right, and the copy still didn’t convert, because the underlying offer was trying to be everything to everyone. ChatGPT will happily write persuasive copy for a confused offer. It won’t tell you the offer is confused. That’s still your job, and it’s a big part of why I built how I approach pricing small business services with AI around getting the offer clear before the words go anywhere near a prompt.
The second uncomfortable truth: prompt libraries you buy for £27 on a landing page are mostly the same fifteen prompts recycled with different formatting, and half of them use the word “use” in the prompt itself, which should tell you something about how generic the training behind them is. You don’t need 200 prompts. You need about six that you understand well enough to adjust on the fly, which is the entire point of this post.
Prompts for editing, not just drafting
The editing prompts get less attention than the drafting ones, and they’re where I spend most of my actual time now:
- “Read this draft and cut it by 30 percent without losing any specific detail or example. Cut adjectives and hedging first.”
- “Which three sentences in this draft could a competitor’s copywriter have written word for word? Rewrite those to be specific to my business.”
- “Read this as my most skeptical potential customer. What would make you stop reading, and where would you not believe me?”
- “List every sentence in this draft that starts with ‘we’ or ‘our’ and rewrite it to start with ‘you’ or a specific reader problem instead.”
That third one is the most useful prompt on this whole list. Ask ChatGPT to argue against your own copy and it finds the weak claims and vague promises faster than most human editors, because it’s not attached to having written them.
A step-by-step process for one piece of copy, start to finish
- Gather 3 to 5 real customer quotes and one sample of your actual writing voice
- Write a one-sentence brief: audience, problem, desired action
- Ask ChatGPT for an outline first, not a finished draft
- Ask for three full drafts with different opening angles
- Run the “skeptical reader” prompt against your favourite of the three
- Cut 30 percent, replace every generic sentence with a specific one
- Read it out loud before it goes anywhere near a client or a send button
That last step sounds basic and it’s the one people skip most. Copy that reads fine on screen often trips over itself out loud, and your ear catches rhythm problems your eye slides past.
Where this fits into a bigger AI marketing setup
Prompts for copywriting are one small piece of a much bigger question about how a small business uses AI across marketing, not just for one email at a time. If you’re building this out across a team or an agency, the prompt-by-prompt approach stops scaling fast, which is exactly the gap I cover in what works with AI tools for marketing agencies. And if every piece of copy your business puts out still sounds like it was written by the same anonymous AI voice as your competitor down the road, that’s a positioning problem before it’s a prompting one, which is the whole argument in why small business AI marketing sounds exactly the same.
What I’d stop doing immediately
Three habits I see constantly that quietly wreck copy quality:
- Asking for a “professional tone” as if that’s a specific instruction. It isn’t. It just means “sound like everyone else’s LinkedIn bio.”
- Accepting the first draft because it’s fast, not because it’s good. Fast and good are different problems.
- Never telling ChatGPT what not to say. A five-second list of banned words and phrases (unleash, game-changing, cutting-edge) does more to fix generic copy than almost any other single instruction.
For a quick check, run it through my tool that writes the prompt for you.
Frequently asked questions
What is the single best ChatGPT prompt for copywriting?
There isn’t one universal best prompt, but the highest-impact habit is feeding ChatGPT real customer quotes and a sample of your actual writing voice before you ask for a draft, then always requesting three versions with different opening angles instead of accepting the first output.
Can ChatGPT write copy that converts?
Yes, but only when the input includes specific customer language, a clear single call to action, and an editing pass that cuts generic sentences, because a strong offer with weak copy will often outperform weak copy dressed up as a strong offer.
Why does ChatGPT copy sound generic even with a good prompt?
Usually because the prompt asked for a tone (“professional,” “friendly”) rather than providing an actual voice sample or real customer quotes, which leaves the model filling the gap with the most common marketing phrasing it has seen rather than your specific voice.
How many ChatGPT copywriting prompts do I need to know?
Around six: one for outlines, one for full drafts with multiple angles, one for headlines or subject lines, one for cutting length, one for a skeptical-reader review, and one for banning generic phrases, adjusted slightly for whatever format you’re writing.
