I use AI in my work. A lot. I have a paid subscription and multiple custom instances for myself and even one for my client.
I started with ChatGPT but later switched to Claude because its responses were more accurate and to the point, and its feedback more critical. I’m also a big fan of its Projects interface that’s more transparent than a ChatGPT custom instance for my use cases.
But!
I don’t use AI in a way you think a copywriter would use it, giving it a green light to come up with words on my behalf. That would be very harmful to both my clients and my reputation.
What I do NOT use AI for (and why)
- Coming up with new ideas: They won’t be new but someone else’s, since AI has no ideas of its own.
- Writing something from scratch: I can do a much better job myself.
- Asking it for an opinion on my writing: It doesn’t have its own opinion in the sense that it can’t give you an honest one. It’s programmed to give answers that keep its Master happy, so I wouldn’t trust its answers to be objective.
- Reviewing client websites: I tested it to see if I can improve my website audit processes, and it turned out to be a totally unhelpful and unreliable disaster.
- Marketing/business strategy: This topic is too complex and depends a lot on your business context. I saw it produce complete rubbish multiple times, so I don’t trust it here.
- Any topic outside my expertise: My experience shows that I can’t trust its responses 100% (sometimes, not at all). So, I only give it tasks that I’m knowledgeable at enough to be able to assess the quality of its responses.
What I do use AI for (and only under these conditions)
- Proofreading copy (the whole piece at once)
- Making copy more succinct (only a couple of sentences at once maximum, otherwise, it will cut way too much and you lose important details, your voice, or both)
- Repurposing my own content: creating drafts for shorter posts for LinkedIn based on my in-depth articles (but I always finalize the drafts myself)
- Brainstorming titles/email subjects
- Data sorting and analysis (only if I can double-check whether the outcome is legit quickly)
- Any other task I’m equally good at, but that AI can do faster (keeping track of certain search rankings, creating a task list from a long client email, etc.)
Basically, I treat it not as an expert consultant, but as a junior assistant that saves me time but whose work I double-check every single time and only “approve” if it meets my standards. And even that only works if I give it super-specific instructions.
Now, let’s talk about you.
I’m pretty certain you use AI, too. Which as a business owner on a budget who has to take care of your website and marketing yourself, you absolutely should!
Yet, you may have noticed that it’s not that easy to make it respond in the way you want it to.
Below I’ll share with you a way to get it to proofread your writing without rewriting the whole thing. I tested it in ChatGPT, but it works with any other generative AI model.
The one prompt that helps me proofread without losing my voice (and sanity)
I’m sure this happened to you way too many times:
You ask it, “Proofread this text,” and it goes overboard and changes way more than grammar mistakes, among other things, sprinkling random “em” dashes where they do not belong 🙄
Why? Because it’s trying to be “helpful”, which in ChatGPT world, that often means doing something, even if nothing needs to be done. Like a guest who insists on rearranging your furniture just to feel useful.
Use the prompt below to make sure:
- It only corrects parts with actual mistakes and doesn’t change the rest
- You can easily double-check its suggestions
Main prompt:
Consider this text: [insert your text] **Instructions:** Correct only objective errors: grammar, punctuation, spelling, and clearly incorrect word usage. Do not rephrase for style, clarity, tone, or personal preference. Respond in before/after format, and briefly explain each correction. Only include the sentence(s) that contain changes. Do not reprint the entire text. If there are no errors, respond with “No changes needed.” In case you’re curious about the prompt’s structure:
- I first give it the text and then the instructions, since it regards the last parts of the prompt the most important.
- I separate the instructions part from the text part with a marker **Instructions:** to make it crystal clear where instructions start.
- I phrase each instruction in one sentence to make it as simple as possible for it to follow.
The general idea is this: keep your instructions as clear and unambiguous as possible, breaking them into clear logical steps—an approach recommended by AI engineers to prevent AI from making up stuff or going off-prompt.
“Another approach involves introducing structure into the model’s reasoning. By prompting it to check its own outputs, compare different perspectives, or follow logical steps, scaffolded reasoning frameworks reduce the risk of unconstrained speculation and improve consistency.”
Here’s how the prompt works.
A real-life example of a proofreading prompt in action
Greetings from the future! I’ve just finished writing this article and want to proofread it. Here’s what happens if I use a simple “proofread this text” prompt.
Prompt: “Proofread this text: [Full article]”
ChatGPT response: Reprints the full text. Corrects more than just grammar. You can’t identify the exact changes right away.
But don’t worry. Here comes our custom prompt to the rescue.
Prompt: Custom proofreading prompt from above
ChatGPT response: Corrects only grammar mistakes. Before/after parts clearly identifiable.
How to save your prompt as a shortcut (optional)
If you’d like to use the same prompt in the same chat faster, let chatGPT save it under a certain shortcode with this prompt:
Bonus prompt: Save it as a shortcode to reuse later
From now on, when I type ProofreadThis: [Text], it means: Correct only objective errors: grammar, punctuation, spelling, and clearly incorrect word usage. Do not rephrase for style, clarity, tone, or personal preference. Respond in before/after format, and briefly explain each correction. Only include the sentence(s) that contain changes. Do not reprint the entire text. If there are no errors, respond with “No changes needed.”
Now you can just type:
ProofredThis: [insert your text]
…and it will use the same proofreading instructions.
Or if you keep using the same prompt several times in a row (either this one or another one), just type:
Same: [insert your text]
This asks chatGPT to apply the previous prompt (whatever it was) to the new text you provided.
Lazyness productivity level: epic 😁

You can save any nother prompt under a shortcode of your choice by using this prompt template:
From now on, when I type [Shortcode]: [My text], it means [describe exactly what you want chatGPT to do with the text].
Important: The shortcode you defined this way will only work within one chat. If you’d like it to work in another chat, you have two options:
- Define the shortcode in a new chat again, or
- Create a custom chatGPT instance and add its definition under its instructions (in this case, the shortcode will work across all new chats you’ll have with this particular instance)
Final words of wisdom
AI writing tools are great, but only as long as you don’t let them take the steering wheel. You’re still the one putting your name (and/or your money) on the final version, which means you need to quality-control every word.
And you can’t do that unless you have a solid understanding of:
- your ideal clients,
- your value proposition, and
- what good copy actually looks like.
If you want help with that, check out my online course on zeroing in on your audience and uncovering your UVP.
Or 📕 grab my book “Making Your Website Work”, which will teach you what good copy is and how to improve bad copy in no time.

There are so very many ways the idiot ‘AI’ can and does mess up, that I think it’s best for me to stay with the HUMAN version (mine).
Everyone is enamored of the speed of generating ‘content’ – and most people will NOT be diligent about checking what they get back.