We are becoming increasingly adapt at detecting the tell-tale signs of AI-generated text. We see uniform styles, just a little too polished, a little too symmetrical and now too familiar to us. We are familiar with the overused vocabulary: leverage, plethora and so on (Side note: I, for one, want to reclaim ‘plethora’ for human use. I like the word). However, there are more subtle devices that are equally overused. This is where LLMs, such as ChatGPT, for all its fluency, often fall back on a small arsenal of rhetorical and stylistic devices that, while effective in isolation and when sparingly used, can become repetitive and formulaic when overused. This article unpacks some of the most overused stylistic ‘crutches’ that tools like ChatGPT lean on.

1.       The False Reframe: “It is not, X, not Y, but Z”

Example of use: This is not just a tool. It is not just a co-creator. It is a revolution.

This is a much-used device in TED-style communication. It can also be used to great effect in advertising. The UK department store Marks & Spencer’s series of commercials employed this beautifully with now-famous (and delightfully parodied) catchphrases such as ‘This is not just salmon; this is M&S pan-seared salmon’.

The device sets up a minor claim, only to dismiss or transcend it in favour of something more dramatic or profound. So, the reader is given an ascending pattern: mild to moderate to grand. It is intended to be persuasive and insightful. It is a shortcut that creates an illusion of depth without giving any actual complexity.

It can be useful in a hook, but it is very predictable now.

“It is not just using a tool.

It is not merely streamlining your workflow.

It is boldly repackaging autocomplete as personal genius.”

2.       Triplet Paralysis: The Rule of Three

Example of use: Fast, scalable and secure.

This is a real shame because the triadic structure (or triple enumeration, the rule of three) works very well in public messaging. We remember so many of them from the Covid-19 pandemic: ‘Hands. Face. Space’ in the UK or ‘Know. Prevent. Protect’ from Canada.

The rule of three is inherently effective because it is satisfying and easy to remember. It is also a device that is used in many languages. Tools like ChatGPT leans heavily into this pattern, especially when summarising benefits, characteristics or steps. Then it becomes overused and the consequence of this is that it offers rhythm, but not depth. Filler text, essentially.

Engage

Inform

Forget instantly.

3.       Balancing Hedging

Example of use: “While X has benefits, it also carries risks”.

This is a perfectly normal and common device to bring about symmetry to a sentence. It serves to frame both sides of an issue and often ends in a noncommittal conclusion. It appears constantly in AI-generated text, and it is a hallmark of LLMs’ risk-averse design. It avoids controversy. It substitutes real human judgment with vague neutrality.

“While AI-generated content can enhance productivity, it may also result in the total erosion of originality—but that depends on how you use it.”

4.       The Epiphany Device

Example of use: “Because thinking isn’t just about saying something well. It’s about becoming different by the end of the sentence than you were at the beginning.”

This is similar to the False Reframe described earlier. This is a rhetorical flourish that uses antithesis, i.e. saying versus becoming, parallel structure (“it’s about…”) and climactic progression.

It is effective because it conveys emotional and intellectual resonance. The statement becomes a relevance. Caterpillar to chrysalis to butterfly.

But AI-generated text ends too many paragraphs in this way. If everything is an unveiling - the butterfly revealed -  then none are.

“It’s not just about reaching a conclusion. It’s about sounding like you’ve had one.”

5.       Templates for Artificial Warming and Empathy

Example: “Navigating career transitions can be challenging. It’s completely normal to feel uncertain or overwhelmed at times. That’s why it’s so important to take small, consistent steps and be kind to yourself along the way.”

This sounds gentle and reassuring, but it’s also risk-free, and emotionally prefabricated. You’ll find this tone in self-help articles, HR copy, and countless AI-written blog posts. It is overused because programmed empathy is better than dealing with risk.

“It’s okay to feel lost. Many others have, too. Fortunately, copy like this exists to cradle your anxieties in algorithmic warmth.”

6.       Excessive Signposting

“Let’s explore three reasons why…”, “In conclusion…”

These are crutches that we humans always lean into in natural writing. They help organise ideas. The reader can also navigate their way through.

However, when overused, it does nothing other than flatten the prose into textbook scaffolding. That said, this is probably the least offensive of the overused AI devices (this is only because I always do them).

Symmetry without Surprise

These stylistic patterns employed by LLMs are designed to bring clarity to the reader. They are also designed to optimise readability and to tread a path that is safe, formulaic and uncontroversial. No alarms and no surprises, please.

To escape this, maybe to err is to be human – I have read this line a lot in relation to AI. It is true. Maybe we should write normally and freely, with asymmetry, statements that introduce risk, conflict and contradiction – you know, just like any human-written article in the legacy press that so many like to look down their noses on. Even better than that would be to introduce a structure of very simple dialectical thinking. This will be the subject of the next article.

Epilogue

I asked ChatGPT to finish this article with a piece of prose incorporating all these examples. Who says that it cannot be funny?

It’s not just a tool.
It’s not even just a co-pilot.
It’s your literary life raft when you haven’t written a decent sentence since secondary school.

Fast. Elegant. Empty.

While AI can improve efficiency, it might also let you outsource all critical thought—but that depends on how shameless you're feeling.

Because writing isn’t just about finishing a paragraph.
It’s about sounding like you became wiser somewhere between clause two and three.

Feeling overwhelmed is normal. Many others have felt the same.
That’s why this paragraph was pre-soaked in algorithmic empathy just for you.

Let’s explore three reasons this seems like good writing:
It’s structured. It’s smooth. It vaguely resembles something meaningful.

In conclusion,
it’s not quite writing.
It’s not quite thinking.
It’s just vibes with punctuation.”

Even ChatGPT clearly disapproves.

Thank you for reading.

Mark Gibson

Leeds, United Kingdom, May 2025