“I hope this email finds you well,” or how not to sound like ChatGPT

We’ve all heard the complaints from colleagues or seen it for ourselves: we are deep into the AI era of identical job applications and generic bids for work. You’ve felt it: the uncanny sense that something you read wasn’t wrong per se but also wasn’t quite… human. Sometimes it’s because you read the same thing in two different emails from two different people but often, honestly, it’s because it’s suspiciously free of errors and/or a bit… dull.

There’s an easier blog I could write. I’d call it: “Kids today, eh?” It would be about how awful it all is and how we all had to work harder back in the day* - but John C Maxwell** said, "Change is inevitable. Growth is optional." He was right. Let’s move on.

AI can help you grow, so instead of wasting your time whining about how other people are using it, I’m going to tell you how to make natural language tools sound more like you – but if you had perfect spelling and grammar and could write at lightning speed. Growth as a professional today must include using AI well, otherwise you’re just a hobbyist.

What kind of writer is ChatGPT?

ChatGPT is the most complicated and competent bad writer of all time. Awesome in its accuracy, pace, professionalism and adaptability, but - if you’re like me and use the technology every day – it has a set of tells that I have to fix Every Damn Time.

Problem one: It plays bullshit bingo

You might already know this but, in case not, here’s our handy guide to the things you should just delete because they’re shockingly over-used in all ChatGPT prose. You should also stop using them if you are writing them without the help of AI, because people will start to suspect you’re a robot. I have helpfully put them in bold and merged them all into one glorious slice of nonsense that I made my private GPT write for me – which is kind of a pass agg move but, as ever, the AI was happy to oblige:

I hope this email finds you well.

In today's digital age, the world of information is a treasure trove waiting to be explored. Let's dive deeper into this intricate and multifaceted landscape. A tapestry of data, ideas, and innovations forms the foundation of our modern existence.

It is worth noting that technology has been a game-changer, transforming how we access and process information. Furthermore, the nuanced and complex nature of digital content requires a critical eye to discern its value.

Delve into this realm, and you'll find that understanding the intricate details is crucial. In conclusion, mastering the essential skills to navigate this environment is more important than ever.

In all honesty, I kind of love it. On the barometer of bad writing, I’d put this at ‘hybrid car advert’: the words are perfect and well-chosen but also meaningless. It takes a really skilled set of technicians to have AI generate this after reading the whole of the internet. It’s amazing to me that large language models don’t just turn into the bottom half of the internet, join the culture wars, abandon the basic rules of spelling and grammar, and spout unfounded accusations and profanities at the other GPTs.  

Problem two: It half-arses things

This isn’t something your AI is doing wilfully. ChatGPT is a giant probability machine and has to make a call about what you want to be told, at the first pass. You could ask it to tell you everything it knows about cats, but if it actually did that you’d be trying to pour half the internet into one chat. Everything would break and it would be absolutely no use to you even if it didn’t. Your poor AI has to make a call about how much you actually want to know, and it has to assume you’ll come back for more or refine your question if you need to.

So it says less, misses important things, uses a LOT of bullet point lists and even more Unnecessary Capitalisation, and (although this is happening more rarely over time) skips a few fact-checking steps. User beware. Ask it more questions.

Problem three: It doesn’t know what it’s saying

This is the heart of the problem. Bad AI writing is the stuff that’s generated when the AI doesn’t know what point it’s trying to make: there’s no personal insight, no call to action and no anecdote – unless you ask it to invent them, which won’t feel authentic. It needs to make a point. It needs you.

How to fix it

Verify the facts to remove errors, delete the over-used stock phrases and sort the bullet point lists (“Hey, ChatGPT, add more exposition and use fewer lists”).

After that, the only trick is to give it a bit more thought:

1. Tell a story

According to Southpark creators Trey Parker and Matt Stone, effective stories connect events using "but" or "therefore" instead of "and then". It’s become known as the “But & Therefore Rule” – which is perfect because there’s not a character in Southpark who wouldn’t lean in to a butt gag.

It’s a simple bit of genius: "But" introduces a complication, “Therefore" establishes causation or consequence. A boring story might say: "This happened, and then this happened, and then this happened." It’s more interesting if you follow the rule: "This happened, but then this happened. Therefore, this happened."

It’s how you create causality, conflict, rising tension and a logical progression of events rather than just a sequence of stuff. And you can either get it right by telling the AI the story in your prompt, or fix it yourself when you edit the generated content. Either works.

If you think that only applies to longform creative content like blogs, think on. Good marketing, your best email, your sales pitch or your closing arguments should all follow the same rule if you want people to stick with you.

2. Mix it up

AI needs to follow rules, so the phrasing, sentences, paragraphs and sub-sections all tend to have a regular length and rhythm. That’s nice when you create something short, but boring as all hell in a longer bit of content. Ask your AI to mix it up a bit.

3. Use this prompt

At Engine Towers we worry sometimes about giving away the recipe for the secret sauce, but you can have this: a prompt you can drop in to fix most of the annoying ChatGPT issues in one go – and I’m going to assume you are up to the task of editing it to suit your cultural context and tastes:

“Write in UK English including using ‘s’ not ‘z’ for words like organisation. Only capitalise the first word of headings and subheadings, unless using a proper name. Use the active voice rather than passive voice. For longer text, vary sentence length to avoid monotony. If drafting an email, do not start with ‘I hope this email finds you well’. Use a range of alternatives. Tone should be professional but friendly rather than too formal.”

Better yet, get a private GPT and bake all this plus your house (or personal) style into the background training. Enquire within – info@engine-ai.co.uk – to find out more.

4. Be yourself

Taylor Swift said, “Just be yourself, there is no one better." Everyone else is taken anyway. Trying to sound like someone else – someone you want people to think is smarter, more eloquent or more professional – is exhausting. Be you.

Your AI doesn’t know as much about you as it will claim: you’ll need to tell it. Having a swipe at stale newspaper columnists, adding a butt joke, quoting a pop music legend, waging war on dull writing… none of that was ChatGPT’s idea. But it’s why you might remember this advice.

Get the AI to check your spelling and grammar, though. It’s better at both. Trust me.

*I asked Perplexity and these op-eds have, of course, already been written, but they are most often to be found behind the Telegraph paywall. Mercifully.

**Look him up. The very short biog will tell you he’s big on the leadership speaker circuit, but the longer read includes the less common fact that he received the Mother Teresa Prize for Global Peace and Leadership.

‍