Oops, did you really write this yourself? (Or how we know that your text is actually GPT-written)

The biggest mistake people make when asking an AI to write is failing to formulate what they actually want to say. They hand over the task and get back something that technically contains words but lacks personality, direction, and intent. The result is often so generic that it takes more time to fix than it would have taken to write it yourself.

I recently spoke with someone who claimed he was dictating the words (with an app) in his text. And I almost believed him. Certain patterns often reveal when a text is at least partially AI-generated or heavily machine-edited. Typical signs include unnecessary summaries, closing statements like “We hope this was helpful”, and English-influenced phrasing when the target language is Swedish. There are also structural giveaways: misplaced quotation marks, periods inside quotes, wrong types of dashes, and other formatting errors.

This particular case was so interesting so I got it analyzed with the o3 engine. It came to the conclusion that it was more likely written with help from AI, than with the personal touch: https://chatgpt.com/share/680cc747-1a24-8000-9ca7-5c6a1b8bc4b1.

So what is the real issue?

The real problem with you using AI is not just bad wording. It’s that you risk letting the machine do the thinking for you. And once you do that, your own edge starts to fade. The same goes for your shitty AI music, that we’ve spoken of before here.

AI should never be your brain. It should be your tool. You write. You think. Then, if needed, you let AI help shape the wording, clean up the flow, or simplify where it makes sense. Or just proofread it.

Especially when writing in Swedish, machine-generated texts often reveal their English roots. You’ll see misplaced punctuation, wrong dashes, and direct translations. Many of these things don’t belong in proper Swedish writing. Another common giveaway is the frequent use of phrases like “and yes”, “and no”, “so yes”, “so no”, or any sentence starting with “and” or “so” followed by a conclusion or final statement. These constructions are often overused in AI-generated texts to mimic conversational flow or artificially lead into a summarizing statement. This is a typical AI pattern, used to signal a closing remark or to simulate conversational tone in an unnatural way. Don’t use it! But I hope you find this well…

Or maybe not. Don’t find it well. Write it yourself for gods sake!

If you don’t control the AI, the AI will control you. Either way, it won’t get you smarter. It will make you more stupid and someone else smarter. If you don’t know exactly what you want to say, the machine will decide for you and then, your content will go sideways or even worse be pointed out as – surprise – machine generated!

How to became less stupid

Write first. Write the text ENTIRELY from the start to till the end!

Use your brain. Let the machine assist – but after you’ve done your part. One way to keep your writing human and distinctive is to clearly define your tone and set strict rules for your GPT. Make it reflect your personality, whether that’s cynicism, aggression, humor, or sharp directness. And last but not least: specify your punctuation style if you’re not american or english speaking.

Specify exactly which type of quotation marks it should use. Specify it all, from dashes to dots. The dashes and everything surrounding the language are just as important as your content. Ban clichés, generic phrasing, and predictable AI structures. And most importantly: never allow your GPT to end texts with hollow conclusions, empty summarizations, or polite fluff like “I hope this was helpful” or “To summarize this, it is always important to be politically correct and search for facts”. Is that really something you would say if you wrote the text? Maybe. Personally, I’d sometimes prefer to summarize it with “In short, you’re an idiot and people out there tend to not fact check things before typing it. Don’t be like them!” – if that fits the tone better. Most of the time it doesn’t, but neither do typical GPT summaries and clichés.

Even if you ask your GPT to proofread your text, you should always do it yourself too, because there are usually plenty of errors that don’t – and never will – reflect you, your personal style or anything else that makes the text unique. Make it unique. Make it yours. Make your licensing fee for ChatGPT matter.

Write rules that prevent the GPT from falling into the usual traps of generic phrasing. Since most users end up using the exact same style, your writing becomes indistinguishable from others and therefore extremely easy to identify as AI-generated.

Also, AI tends to overuse other elements like:

• Unnecessary listings like this one…
• Excessive use of symbols and formatting gimmicks…
• … and last but not least, the very standard and predictable way of wrapping up arguments.

Avoid these patterns. Stay sharp. Stay human. Stay smart, not an idiot.

Did you know how many hyphen-styles there are?

Another easy way to spot machine-generated text is through its inconsistent use of hyphenation and dash styles. Here’s a quick overview of the different types — and yes, there are quite a few:

• Hyphen (-): The standard dash used for compound words like “machine-generated” or “state-of-the-art”.
• En-dash (–): Slightly longer than the hyphen, traditionally used to indicate ranges (“2023–2025”) or relationships (“Sweden–Norway border”).
• Em-dash (—): Longer than both hyphen and en-dash. Used to insert a break in a sentence — like this. Very common in English, rarely correct in Swedish.
• Figure dash (‒): Same width as a digit, used in phone numbers like “070‒123‒4567”. Rarely used by humans.
• Horizontal bar (―): Rare and used for quoting or dialogue replacement. Very uncommon in modern writing.
• Minus sign (−): Looks like a hyphen but is technically a math symbol. Some AIs mix this up with hyphens.

When these are mixed, misused, or inconsistently applied, it becomes an easy giveaway that the text has been machine-generated or poorly proofread. Just like with quotation marks — where you have single quotes (‘ ‘), double quotes (” “), smart quotes (“ ”), straight quotes (” “), and even other localized variants — these details matter.

Humans who care about their writing usually know the difference. AIs typically don’t. Yet another reason why proofreading matters. This is also horrifying, since normally – to avoid this – you have to edit your text one last time after you got it printed by an AI. Even if you ask an AI to fix those errors, it sometimes won’t, so a search and replace will be necessary at some point. But hey, if you taking your time writing crap anyway, it is worth spending some more time to get it right and professional. Again, don’t be an idiot. Don’t lie to those of us, that already know about your errors.

Oh, right. One more thing. I used AI to generate the image…