If Your AI Content Sounds Generic, This Could Be Why

I've spent an inordinate amount of time arguing with ChatGPT. Not about politics or whether AI is going to take over the world. Not even about whether it's coming for my job.

I've spent much of the last few months arguing with it about paragraphs and flow.

As somebody who's made a living from writing for more than 25 years, I appreciate how ridiculous that sounds. Yet there I was, repeatedly telling one of the most sophisticated pieces of technology ever created to stop breaking perfectly good paragraphs into tiny fragments.

The reason is simple. I'm bored of reading AI-generated content and I know I'm not alone.

LinkedIn is drowning in it. Every day I come across posts written by smart, experienced professionals (even marketing professionals) who clearly know their subject matter inside out. Yet somehow, when AI gets hold of their ideas, everything starts sounding remarkably similar. The same dramatic openings. The same single line paragraphs structures. The same attempts to create suspense. The same slightly breathless rhythm that feels more like social media than thoughtful writing.

Instead of becoming the copywriter who spends the next decade complaining about it, I decided to channel my inner Ted Lasso and “get curious.” 

If businesses are going to use AI to create content, and let's face it, they are, I might as well put my skills and years of experience to good use – helping them to do it better. 

What started as a few experiments quickly turned into something of a mission. 

My challenge – create an AI toolkit that could consistently move AI output from generic to recognisably written by Sarah.

The early experiments
Like many people, I began with what seemed like a fairly logical assumption. If AI wasn't producing content that sounded like me, then presumably it simply didn't have enough information to work with. I figured the solution was therefore to give it more context.

I uploaded my blogs, newsletters, LinkedIn posts, website copy and client testimonials. I created a detailed 19 page toolkit to upload as a source file, explaining who I was, who my audience was and what I was trying to achieve. I gave it customer profiles, tone of voice guidance and background information about my business. I thought I’d left no stone unturned.

Nah. The output improved, but only very slightly. The content it created became more relevant. It picked up some of my vocabulary and preferred turns of phrase. But it still wasn’t hitting the mark. I know I set a pretty high bar, and I can’t expect it to ever sound 100% like me – but the output wasn’t nearly close enough.

At first, I couldn't work out what was wrong. The facts were correct. The grammar was fine. It wasn't misunderstanding the subject matter. Yet every time I read the content back, the essence of ‘Sarah’ was missing. And yes, you heard it here first - I confidently claim to have an essence!

The breakthrough moment
The breakthrough came when I stopped focusing on the words and started paying far more attention to the structure.

I noticed that the articles the LLM produced still had a tendency to fragment ideas into tiny pieces. It would introduce a thought and then abandoned it before it had chance to develop. Paragraphs were usually reduced to a single sentence. Every few lines there seemed to be another attempt to create emphasis, suspense or a dramatic reveal. That just smacks of AI.

One of the reasons so much content on LinkedIn now feels generic is because the LLMs are using use the same writing rhythm. AI naturally gravitates towards short standalone statements, repetitive emphasis and a style of writing that's optimised for skimming rather than reading. Chat GPT refers to it as “Linked In Influencer” style.

Professional writers generally work differently. I'm not suggesting that all copywriters, journalists, authors and marketers write in the same way. We absolutely don't. But most experienced writers share one characteristic. We trust our thoughts, our structure and we trust the reader.

We trust them to stay with an idea for more than a sentence. We trust them to follow a line of reasoning. We trust them to enjoy an example or story before we reveal the point we're making.

When I write a blog, my natural instinct is to introduce an idea, explore it, illustrate it with an example and then draw a conclusion. The reader arrives at the insight because I've guided them there. That's classic professional writing.

What I realised during my experiments was that LLMs weren’t struggling to understand my style. They were struggling to understand my “writing behaviour.”

There's a big difference.

Style is the language you use. It's the words, phrases and tone that make your writing recognisable. Writing behaviour is how you think on the page. It's how you build an argument, develop an idea and guide a reader from one point to the next.

The more I experimented with this distinction, the more important it became. Most businesses trying to improve AI-generated content are concentrating on tone of voice, customer profiles, messaging and the all important prompts. All of those things matter enormously. But they're only a small part of the picture.

If you want AI to produce content that genuinely sounds human, you also need to teach it how you communicate ideas. Do you naturally use stories and examples to explain concepts? Do you build arguments gradually or jump straight to conclusions? Do you write like a journalist, a consultant, an academic, a founder or a marketer? These are the things that shape how content feels when somebody reads it.

Once I started documenting and explaining my own writing behaviour, the quality of the output improved significantly. Not perfectly. And unfortunately not every time – because LLMs have a nasty little habit of defaulting to what’s become their conventional way of writing (like a LI influencer), but enough to convince me that most businesses are only scratching the surface when it comes to training AI.

The foundations need to go much deeper than most people realise and what surprised me most was just how deep they need to go.

What began as a slightly nerdy attempt to work out how to stop ChatGPT violating my paragraphs has turned into a much bigger exploration of how professional writers actually write and how I can teach AI to do a better job of replicating it.

Because AI is here to stay and the only way it's going to improve is if we collectively train it. I’m taking up the mantle. 

The more I tested, the more I realised that most businesses haven't actually documented the information AI needs in order to sound distinctive. They know who they are. They know their customers. They know what makes them different. They know how they like to communicate. But much of that knowledge lives in people's heads rather than in a format AI can understand and use consistently.

That's what led me to create a new service called Your Content DNA.

The idea is simple. Before you ask AI to create content, you first need to teach it who your business is, who your customers are and what they need to hear, how you sound, how you develop ideas and what your writing behaviour is. Only then can it begin to produce content that feels genuinely reflective of your business and sounds human, rather than a generic approximation.

I've spent the last few months trying to capture and document those foundations, both for my own business and for clients who want to use AI more effectively.

I'll share more of what I've learned in future blogs.

For now, if you're using AI to create content, ask yourself this.

Have you really taught it how your business communicates?

Or are you still asking it to fill in the gaps?