I was just writing an email to a client, and realised I was sounding like… AI? I have always been able to switch on a formal tone when writing business emails and things alike, so this style of writing isn’t new for me… so why does it only feel so wrong now?

I pondered the thought (see, formality is in my nature!) and noticed a lot of similarities in my writing to something that Gemini or ChatGPT would produce. So I wondered, does writing formally automatically make you sound robotic? Let’s dive deeper.

How Does AI Formulate Its Tone

By now, we know that AI uses Large Language Models (LLMs) to learn how humans speak. Artificial intelligence has learned from millions of emails and figured out the common vocabulary, sentence structure, and tone with which they’re written.

Naturally, AI follows a very strict structure with a set of rules, which can lead to results that sound cold, neutral, and overly corporate. AI is known for its excessive use of hyperbole and adjectives that a sentence could definitely do without.

When AI has spent so long scanning text across the entire internet, it’s obvious that it was eventually going to pick up on these patterns. Yet in doing so, it believes that it is doing what it is supposed to in order to sound ‘formal’ or ‘correct’, but instead it’s making its results sound lame and predictable.

Does Prompting Solve the Issue?

With the growth of AI, I have seen people put a growing emphasis on the importance of prompting. I remember trying it for myself at the beginning of ChatGPT, and I did indeed notice a difference in responses.

What always irritated me the most about language models was the introductory paragraph to their responses:

• ME: I am struggling with X, can you please do Z
• AI: It’s the classic X problem… [speaks about X for a line or two]. Here is Z that we can do to help you…

Or something like that… every TIME. What I’m getting at was that this slowed once I prompted AI to stop doing it – so can’t we do the same for drafting emails?

To a certain degree, yes. There are certain patterns and traits that an AI passage of text has, that when commanding it to exclude, can make it slightly harder to detect. Some of these traits include:

• Em dashes (—)
• Rule of Three: “It’s not X, it’s not Y, but it’s Z”
• Repeated clarifications and over-explanations
• Overuse of metaphors and similes that don’t work

These are four, to name a few. But don’t get ahead of yourself, because even with their absence, it is likely that people will still be able to detect AI in your writing. According to a 2024 study by Bynder, 55% of US consumers surveyed could accurately identify AI-generated content. So… you’re probably going to get caught out.

Will AI Ever Become Undetectable?

There are constantly more advanced AI models being released that claim to be less detectable. At this point in time, there are no models that are 100% undetectable to everybody.

Whilst researching for this section, I came across Undetectable AI. I wanted to test if this actually works to make an AI passage undetectable, so I asked ChatGPT to write a paragraph on whether AI will ever become fully undetectable – what a mind f*ck, huh? Here’s what it produced:

AI becoming completely undetectable is unlikely in any absolute sense, but it will continue to get close enough that detection becomes unreliable in practice. Detection systems rely on statistical patterns, training data fingerprints, and behavioral cues, all of which can be iteratively minimized or adversarially optimized away by more advanced models. However, there’s a structural limitation: any system producing outputs at scale leaves traces—whether in distributional quirks, metadata, or interaction patterns—so detection shifts from content analysis to context and provenance. The real outcome isn’t “perfectly undetectable AI,” but a collapse of trust in detection tools themselves, forcing platforms to rely more on cryptographic verification, watermarking, and source authentication rather than trying to infer authorship after the fact.

Besides the em-dashes, I wouldn’t say this was horribly ‘AI-y’. I would say vocabulary such as ‘iteratively’ and ‘adversarially’ are a bit excessive, but I don’t doubt that an uber-clever person writing a thesis on this topic wouldn’t potentially utilise those words as well. Whilst not necessarily the Rule of 3, you can also see a repeated use of comparison in the writing, as well.

Oop – time’s up! Undetectable AI just produced it’s passage and-

It is unlikely that the complete lack of detectability of the AI would become an absolute reality, although the level of proximity to it will increase. This will happen because of the fact that detection tools work based on statistics, data training fingerprints, and behavior of AI itself, which can be gradually reduced by even more sophisticated AI. Nevertheless, one should note another limitation, namely, the presence of a trace left by the program, no matter how subtle it may seem, because detection changes from the analysis of the output itself to the context and provenance of the output. Thus, the result does not lie in the creation of a “perfectly undetectable” AI, but in the loss of reliability of detection systems.

It’s… worse?

I think it’s safe to say that AI is not going to be wholly undetectable anytime soon. But if that’s the case, why does my human-written content sound AI-written?

Why is Authentic Writing Starting to Sound AI?

In an effort to be formal, it is easy to adopt the patterns that AI has learnt. We, too think we are sounding ‘correct’ or professional, but as I previously mentioned, we end up sounding predictable.

When AI was trained on ‘good’ human writing, it makes sense that it shares similarity with authentic content. It hasn’t created this tone out of nowhere, but rather mimicked the performative formality that we see in day-to-day life. At the end of the day, formal writing is a performance of sorts, as true informal writing is messy and sporadic (to a certain degree). AI is simply bridging this gap.

Also, it’s a fair point to raise that our increased exposure to AI written content could be causing us to subconsciously mirror it, and adapt certain patterns that it utilises. In wanting to sound polished, we are perhaps inadvertently sounding robotic.

I have previously spoken in other blog posts about people craving human authenticity, and I think that’s becoming true even in formal mediums such as an email or a LinkedIn post. The easiest way to mitigate the risk of sounding like AI is to add your own personal flair into your writing, such as using particular words or phrases that you personally like to use. This individuality is what will make you sound human – we are not meant to create perfect emails! It’s definitely a lesson that I myself need to learn, too.

Final Thoughts

I remember I once posted a passage online for review, and somebody in the comments claimed it was clearly written with AI. They were absolutely adamant. I was gobsmacked and a little offended because it felt like my hard work was being taken away from me. However, perhaps this was a compliment – my writing was being compared to a model that has literally learned what is ‘good’ writing. Hats off to me, I guess.

I don’t need to give another cliché, powerful sentence about how AI is taking over, and we need to learn to adapt – you know this (and if you don’t, read my other blog posts). But what this lesson has taught me is that maybe there is logic in making your emails a little more you, and a little less everybody else.