Why you shouldn’t trust ChatGPT to give feedback on copy.

AI has notes? Of course it does.
This is becoming a familiar tale. A professional copywriting team has produced a piece of writing for a client, maybe a blog or a web page. It’s been written in line with their branding, their tone of voice, and with the right audience in mind. It’s been checked against the brand strategy, and sent to the client for approval.
No-one can be 100% sure what happens between then and the next step, but we can take a good guess. Somewhere along the line, the client asks themselves, “I wonder what ChatGPT thinks of this?”
Quick as you like, the copy has been pasted into an AI tool with the simple instruction: “review this”. Now we’re off to the races. Suddenly, the opening is “too slow”, the mid-section “needs restructuring”, and the conclusion could be “more impactful”. Apparently, lots of great writing needs replacing with generic gumpf that sounds like it was written by a management consultant who’s just discovered the word ‘elevate’.
The client then wants their outsourced copy team – who they’ve already paid to write the piece – to implement these changes.
It's a slightly strange way of working. On the face of it, there’s nothing inherently wrong with using AI to interrogate a piece of writing. Using prompts that are likely to give you useful information back is great. That gives the human writer something to work with, and real reasons why something might be worth changing.
If you’re going to use AI for feedback, you should really know what you’re asking, and what you’re likely to get in return. We can help you to get your prompts sorted – read on to find out how.
If all you have is a hammer, every problem looks like a nail.
If you ask an AI tool to review something, it will review it. It will find things to question and improvements to make, because that’s what you asked it to do. The fact that it’s produced a list of changes does not, in itself, demonstrate that the original work was ‘bad’. All it proves is that the list-making machine is very good at producing lists.
ChatGPT is, in effect, a blunt instrument. And it’s certainly not more qualified to write or comment on your blog than the professional copywriter that you’re already paying to do so.
Why not try a little experiment?
There is an easy way to see this for yourself. Take a piece of writing you already love. It could be an advert, a speech, a landing page, a short story, or anything else. Put it into AI and ask it to critique the work. Then open a new chat, and ask why the same piece is so effective. Read the two responses side by side, and watch the machine make a convincing case for and against exactly the same words.
Is ChatGPT broken? Nope, it’s working exactly as you've asked it to. What it’s doing is identifying possibilities based on patterns in language, and on the instructions it’s been given. It doesn't have a proper understanding of your business, your customers, the conversations that happened before the brief was written, or the reason the writer chose any given phrase. It has no idea why the piece was written in the first place, or what you want to achieve.
And honestly, you can do this with anything. We put Hamlet's “To be or not to be” soliloquy through Chat, and it suggested making the capitalisation “more consistent,” and “modernising” the grammar and punctuation. We fed in Martin Luther King Jr’s iconic “I have a dream” speech, and Chat suggested “varying the repetition,” so that Dr. King no longer “has a dream,” but rather “can imagine a day when…” How profound.
Different does not mean better.

AI can be extremely useful when something genuinely needs attention. It can identify an awkward sentence or highlight language that someone outside your industry might struggle with. The problems start when the real people using AI forget that it’s just a tool, and that its suggestions aren’t necessarily improvements.
That’s how distinctive brand language – the things that make your brand identity, voice unique, and help you stand out against the competition – get ironed out. Deliberate quirks disappear, and character is replaced with a generic version of ‘professional’. After enough rounds, you end up with writing that’s perfectly clear, perfectly polished, and perfectly identical to every other thing produced with the same tools.
It’s Groundhog Day for copy amends.
There’s another problem, too. Ask AI to improve its own suggestions, and it will happily carry on doing so. You can keep refining a piece of copy indefinitely, because there’ll always be another possible way of saying the same thing. At some point, somebody needs to exercise judgement, and decide that the version they’ve got is the one they’re going to go with. That someone should probably be a human being.
So how can we avoid these pitfalls? It’s relatively simple, and will actually make you feel a lot better about your content.
First, before putting any piece of work through AI, have someone who actually knows the context read it. Get them to check if it’s factually accurate. Does it reflect your business, your audience, and your current priorities? Does the tone of voice that it’s written in sound like your brand? And most importantly, is it doing what it’s meant to do?
These questions are much more important than whether an AI tool can produce a slightly different way of saying something.
Secondly, remember that you do actually trust the people you’re working with, otherwise you wouldn’t be working with them in the first place. The real people involved in your project know things that won’t have made it into an AI prompt. Things like internal sensitivities, or the reasons why previous decisions were made. They’ll have a handle on your commercial priorities and stakeholder concerns – something that AI simply cannot fathom.
All this feeds into a wider knowledge, and a three-dimensional view of why a particular point has been made in a particular way.
You should still absobloodylutely give feedback!

If something doesn’t feel right, just say so. “I don't think this is quite right for our customers” is useful feedback. So is “we need to be more careful about this claim” or, “this doesn't reflect the position we agreed last week.”
Those comments give a copywriter something to work with, because they’re based on real, human interactions and actual reasoning. “AI thinks we should make this more engaging” does not.
Feedback should come from you. Never let your own opinions be outsourced to a machine.
If you use AI, give it a proper job.
If this whole piece has sounded like a bit of a rant, we’re sorry. It’s not meant to come across as one long complaint, and we’re not trying to point any fingers. And actually, there are plenty of sensible ways to use AI to review copy. The trick is not to ask it for a sweeping verdict on the whole thing.
If you’re going to feed a piece of writing into an LLM, ask it to flag wording that a non-specialist might not understand. Ask it to identify repeated points without rewriting them. Ask it to find factual claims that might need independent verification. Or ask it to compare the draft with your tone of voice guidance (which you’ve previously fed into it), and check whether the call to action is clear.
These are all useful jobs, because they produce information that you can then assess, without implying that the entire thing should be ripped up.
So, if you are going to use AI, a good prompt might be:
“Review this draft for [clarity/repetition/tone of voice/call to action]. Its purpose is [purpose], its audience is [audience], and we want the reader to [desired action].
“Use the supplied brand guidance as your standard [upload PDF].
“Quote and explain up to five issues. Do not rewrite the full piece or recommend changes purely as a matter of preference. Flag anything requiring human judgement or independent verification.”
Would you like me to produce a punchier version…?

The idea isn’t to ban AI from having anything to do with your published copy, but to ensure that all key decisions on what gets put out are made by people. People who understand your brand, and have genuine hopes and aspirations for it to succeed.
Remember that AI will always find something to change, but that doesn't mean you should. Feedback for feedback's sake isn’t useful at all, and if you’re paying for copywriting, it might be best to trust the human professional, rather than an algorithm that has no idea who you are. And remember, a good copywriter will always take real feedback on board – just as long as it comes from you!
Want to learn more about how the (human) Ketchup Crew crafts all kinds of copy, for lots of different industries and audiences? Or perhaps you’d like to tell a new brand story to your customers, or rethink your marketing strategy to position yourself completely differently in your market? Whatever your copy needs, check out our copy services page for more information.