Is there nothing more than a dash between artificial and human intelligence?

Human and artifical intelligence face each other eye to eye. In between is the dash.

There is currently a rumour circulating on LinkedIn about a supposed “hack” for detecting AI-generated text: artificial intelligence's disproportionate preference for – wait for it – dashes. But is this really true?

I took a look at my ChatGPT history.

I don't use many dashes in longer, more detailed texts. But when it comes to shorter texts for ads and headlines, it's true – almost every text contains at least one.

Did people simply not use dashes before?

I think anyone who has ever written marketing texts themselves knows that this is not true. The shorter the texts need to be, the more precise, concise, and lively they need to be. A dash is the perfect tool for this. It works like a cliffhanger, building suspense in the sentence that can then be resolved in the second half. Multiple pieces of content can be combined in one sentence.

And the problem?

If this opinion becomes more widespread, you will eventually no longer be able to use dashes in your texts because they will be viewed with suspicion from a stylistic point of view. That would be a shame—and would further strain the relationship of trust between service providers and customers.

Does it ultimately make any difference whether the text was written with or without AI? Or should we ultimately focus more on the content?

Types of dashes

Since this topic has generated so much interest, I would like to shed some more light on it here. First, here is a list of all the different types of dashes that exist:

1. Hyphen (separator, quarter-width hyphen, divis)

Character: - (U+002D)

Usage:

• Compounds: E-Mail

• Hyphenation at the end of a line

• Double names: Müller-Lüdenscheidt

This is the “normal” hyphen on the keyboard.

2. Dash (en dash)

Character: – (U+2013)

Usage:

• Inserts: This is – As you know – not an easy decision.

• Distance information: Berlin–Hamburg

• Time periods: 2020–2025

Longer than the hyphen; in Word, it can often be inserted using Ctrl + - on the numeric keypad.

3. Em dash (—)

Character: — (U+2014)

Usage (especially in English):

• Strong pauses in thought: He said it—no hesitation—then walked away.

• No surrounding spaces!

Rarely used in German, but common in English literature. AI systems like to use this dash. That is why the dash is considered such a strong indicator.

5. Protected hyphen

Character: (invisible, but looks the same: ‑) (U+2011)

Usage:

• Prevents line breaks: "E-Mail" stays together and is not split up.

Often used in professional typesetting (e.g., in PDFs or printed works).

** Please note: I am not a native English speaker, so if you spot any mistakes, feel free to write me an email with the correction: wolny.frederic@gmail.com **

So the question is: -, –, or — ? What looks to me like mumbling about Morse code is currently the hottest topic in AI. And since ChatGPT and Co. like to use the —, even though we tend to use - and – in German, the dash is often used as an indication. The reason is probably that most of the training data for language models comes from English-speaking countries.

So could there be something to it after all?

The dash length suggests so, but in my opinion, AI should then also not put spaces before its dashes. However, it does.

Further (alleged) clues for recognizing AI-generated texts

In addition to the dash, several other clues have gained a certain degree of recognition. The list is as follows:

Use of dashes

Just for the sake of completeness.

Use of tricolons

Another indication is supposedly the preference of AIs for groups of 3. Three bullet points, three arguments, etc.

Use of certain emojis

For example 🚀 or ✅

Smooth language

AI texts are usually considered unusually smooth, clean, and bland. Some people also describe them as stiff or sterile.

Clean structure

Where language models are far ahead of us: each of their answers is structured. This is actually a good thing, but in most cases the structure is more consistent than that used by humans.

Spacing after numbers

For example, 20 % instead of 20%.

My list is neither exhaustive nor accurate.

I just want to provide an overview.

The last question that remains for me is:

Why is it so important for people to be able to recognize AI texts?

Not an easy question. I have two theories:

Assumption 1:

Everyone wants to save work with AI, but as soon as we suspect that someone else has used AI for a text or a task, we feel that someone is saving work. And when someone else saves work, we feel that they are lazy.

Instead of making the effort to write themselves, someone has taken the cheap shortcut. And strangely enough, that hurts a little. Not much. But a little bit.

Assumption 2:

It's simple: people want to feel better than others.

This is this big, new thing called AI, and no one knows exactly how to deal with it. The boundaries between truth and fiction are blurring.

And it's a natural human need to want to feel better.

No one wants to be fooled, so we think:

If I recognize that something is written by AI, then I still have the upper hand. The control. You can't fool me.

I belong to the small elite group of super brains who can recognize AI texts with razor-sharp precision – based on small details.