ChatGPT Rated This Blog Post 8.5 Out of 10. Then Admitted It Couldn't Write It.

There's a question that comes up more than any other now.

A business owner looks at what Decodefy produces, seems impressed, and then mentions they're already using ChatGPT. Getting decent content. It seems to be working. So why pay for something that does the same job?

It's not an unreasonable position. And it deserves a straight answer rather than a sales pitch.

So I ran a test.

I took a blog post Decodefy built for a specialist craft business - a Yorkshire-based maker producing bespoke work for the premium end of their market. I pasted the article into ChatGPT and asked it to rate the copywriter. Then I asked what prompt I'd need to recreate the piece. Then I pushed further: compare this post against what a typical local service business gets from a standard AI prompt - and tell me what that difference actually costs them.

What came back stopped me. Not because it was flattering - though it was. Because ChatGPT described, in some detail, exactly what was in that post that its own default output doesn't contain. It named the gap. It put a number on it. It explained the commercial consequences well enough that I've been using the transcript in conversations ever since.

I know, it's nuts. An AI making the honest case against its own typical output.

But that's what happened. And I think it's worth sharing - not to score a point, but because most business owners genuinely can't tell the difference between content that builds trust and content that just exists on a website. The gap is real. It's just not always visible until someone shows it to you.

That's what this is.

The Verdict

The first thing I asked was simple. Rate the copywriter who wrote this post.

ChatGPT gave it 8.5 out of 10.

Not a polite score. A considered one. It broke down exactly why - buyer psychology, objection handling, tone, structure, proof placement, the way testimonials were used to support specific claims rather than dropped in as decoration. It called the writing "much better than generic SEO blog writing." It said the copywriter was "clearly above the level of a typical content writer."

Then it said something that mattered more than the score.

"This copywriter is good to very good. Strong at: persuasion, tone, structure, objection handling, premium service positioning."

That's not a description of AI output. That's a description of what AI output is usually trying - and failing - to be.

What It Would Take to Recreate It

The second question was the one I was really interested in.

What prompt would you need to produce this?

The answer ran to over 600 words. A detailed brief covering tone, buyer psychology, objections to address, specific business facts, testimonials with names and quotes, structural requirements, style notes, length guidance, and a list of phrases never to use.

Not a prompt. A full copywriting brief.

And even then - ChatGPT was honest about the ceiling.

"A usual local-business prompt like: 'Write a blog post about bespoke wooden gates for my website. Make it SEO friendly and include FAQs.' would probably produce something competent but noticeably flatter."

It scored that typical output. Six out of ten. Maybe seven.

The gap between a six and an 8.5 might not sound like much. It's enormous in practice. One produces content that exists on a website. The other produces content that changes what a reader thinks before they've picked up the phone.

What the Gap Actually Costs

The third question was the one most business owners never think to ask.

What are the commercial consequences of that difference?

"Your blog is much more likely to create enquiries from serious buyers because it reduces the exact doubts that stop people acting."

It went further than that.

"A usual AI-style blog can damage that impression slightly, especially for premium local services. Not because readers will always consciously say, 'This is AI.' More often, they feel it indirectly: it sounds like every other trades website, it uses vague benefits, it lacks lived experience, it has no real texture, it says obvious things, it does not answer the awkward questions, it feels like filler created for Google."

And then the line that's stayed with me since I read it.

Sit with that for a moment. You've spent years building expertise. Real expertise - the kind that clients pay more for, refer friends for, come back to. And the content sitting on your website, the posts going out under your name, are quietly undermining that impression every time someone reads them.

Not dramatically. Not obviously. Just enough to make you look like everyone else.

That's the cost nobody talks about. Not because it's hard to understand - because it's hard to see until someone points it out.

The Problem Isn't the Prompt

Most people who've tried ChatGPT and found it lacking draw the same conclusion.

They need a better prompt.

So they learn to prompt. They get more specific, more detailed, more structured. They add tone instructions and style notes and lists of things not to say. And the output gets better - genuinely better. Good enough that it's hard to argue with. Good enough that paying for something else starts to feel unnecessary.

Here's what that conclusion misses.

Better prompting is working harder inside a system that has a fundamental problem. You're improving the instructions to a tool that still doesn't know how to think - it only knows how to write. And those are not the same thing.

Every major AI model - ChatGPT, Claude, Gemini - does one thing at its core. It predicts what a competent piece of writing looks like based on patterns in everything it's ever been trained on.

That's genuinely impressive. It produces clean sentences, logical structure, reasonable arguments. It passes a surface-level read without embarrassing anyone.

But pattern-matching existing writing is not the same as understanding why certain writing works on a particular reader at a particular moment in their decision.

A senior copywriter doesn't sit down and think: what does a good blog post look like? They think: what is this specific buyer quietly worried about? What has already gone wrong for them? What do they need to believe before they'll act - and what's standing in the way of them believing it?

That thinking happens before a word gets written. The writing is just how the thinking comes out.

Generic AI skips that layer entirely. It goes straight to the writing. Which is why it produces content that looks right but doesn't quite land. Technically correct. Structurally sound. And somehow beside the point.

The Other Thing Nobody Mentions

There's a reason British buyers sometimes feel subtly uncomfortable with AI-generated content - even when they can't put their finger on why.

Every major AI model was built by consuming the internet. Billions of words of text, scraped, ingested, pattern-matched. And the vast majority of that content - the marketing copy, the blog posts, the sales pages, the business writing - was written in an American register.

Not just American spelling. American communication culture.

That culture has a particular character. It's urgent. It's superlative. It announces expertise rather than demonstrating it. It tells you something is incredible before it's shown you anything. It uses phrases like "unlock your potential" and "dominate your market" and "the ultimate guide to" with a straight face.

British buyers feel that mismatch even when they can't name it. There's a pressure to it, a slight overreach, a confidence that hasn't quite been earned yet. It's the difference between a salesman who walks in telling you he's the best in the business and a craftsman who just shows you the work.

Decodefy is built for British English from the ground up. Not just spelling - though behaviour, colour, and organise matter - but the entire confidence register. Understated. Earned. Direct without being pushy. The kind of writing that a Yorkshire tradesman or a London solicitor or a Fenland accountant would actually put their name to without wincing.

Which, if you used a standard AI tool to write it, it did.

Same Brief. Completely Different Output.

Here's the part that tends to stop people mid-conversation.

To produce that 8.5/10 post - the one ChatGPT spent 600 words explaining how to recreate - the actual Decodefy prompt was this:

"Write a blog post addressing the five most common myths and objections that stop homeowners from ordering bespoke wooden gates."

That's it. Roughly twenty words.

The 600-word ChatGPT brief covered tone, buyer psychology, style rules, objections to address, business facts, testimonials with names and quotes, structural requirements, length guidance, and a list of phrases never to use. It had to - because without all of that, ChatGPT has no foundation to work from. Every piece starts from nothing. Every prompt has to rebuild the entire context from scratch.

Decodefy already had everything else.

The Master Brand Compass held the voice - the rhythm, the vocabulary, the confidence style, the way this particular business owner explains things to people who are genuinely interested. The RAG database held the business knowledge - the technical specifications, the methods, the testimonials, the real objections that come up in actual sales conversations. The AEC Fractal Framework held the thinking layer - the psychology of how to sequence an argument, where to place the proof, which doubt to address first, when to stop.

The prompt didn't need to carry any of that. It was already there.

So when ChatGPT said the biggest limiter wasn't writing ability but input quality - "give me generic inputs and I will make a polished generic article" - it was describing exactly what happens when there's no permanent foundation underneath the prompt.

You get out what you put in. And most people, most of the time, are putting in twenty words and hoping for a senior copywriter.

The Question Worth Asking

Nobody sets out to publish content that makes their business look less specialist than it is.

It happens gradually. A post goes out that's decent enough. Another one. The website gets updated. The LinkedIn feed stays active. From the inside, it looks like marketing is happening. From the outside - from the perspective of a potential client reading it for the first time - something slightly different is being communicated.

Not that the business is bad. Just that it's like every other business in the sector.

And for an expert who's spent ten or twenty years building genuine depth - that's a quietly expensive place to be. Not dramatically. You won't see it in the numbers this month. But the client who chose the cheaper option because nothing in your content gave them a reason not to - that's a real cost. It just doesn't arrive with a label on it.

This isn't really about AI tools. It's about what you're asking your content to do.

If content is a box you're ticking - something that needs to exist, be vaguely professional, and not embarrass anyone - then a standard prompt and a decent AI tool will probably do the job. There are plenty of businesses where that's fine.

But if you're an expert. If the quality of your work is genuinely better than most of what's out there. If your clients, once they've worked with you, tend to stay and tend to refer - then your content has a harder job to do. It has to communicate that expertise before someone has experienced it. It has to close the gap between what you know and what a stranger can see.

Generic AI content doesn't do that. It can't. It doesn't know what you know, how you think, or what your clients were quietly worried about before they called you. It produces the version of your business that could have been written by anyone who'd spent ten minutes on your website.

Which is usually not the version worth paying for.

The business owner who commissioned that gates post came to Decodefy the same way most people do - with something they needed to say and no reliable way to say it properly.

They left with a piece of content that, by ChatGPT's own assessment, a skilled direct-response copywriter produced on a very good day.

Not because the prompt was clever. Because the foundation was right.

That's what compounds over time. Every piece gets built on the same permanent base - the same voice, the same business knowledge, the same understanding of what the right reader needs to hear before they act. Month one is good. Month six is significantly better. Not because the AI got smarter, but because it knows more about you.

If you've been wondering whether what you're currently producing is working as hard as it could be - that's probably worth a conversation.

Not a long one. Just an honest one.

Find out how Decodefy works →