AI Has a Smell

And more and more people are smelling it.

AI-generated text is everywhere. AI-generated images are everywhere. LinkedIn posts, blog articles, emails, comments, proposals – all of it permeated by this unmistakable smell. And people? They’re becoming increasingly sensitive to it.

That’s what I mean by “smell.” Not in the technical sense, but in the intuitive one. A lot of people – especially those who read and write a lot themselves – smell it. After the first sentence. Sometimes just from the header image.

I’m Not an Anti-AI Evangelist

Before this gets taken the wrong way: I’m not the guy who says everything has to be handwritten, don’t you dare touch AI. I use AI myself. Many of my blog articles start with me recording a voice memo on a walk, and then at home I have AI turn it into a readable article.

And the results aren’t bad either. They carry my story, my narrative, my arguments. It’s not AI slop in the classic sense.

But it’s just not my writing style. Not the way I would have written it. The sentences are too smooth. The transitions too seamless. The word choices too… tidy.

And people smell it.

The AI Giveaways

How do you recognize AI-generated text? By now there’s a whole list of tells – the “AI giveaways” that sensitive readers catch immediately:

• The em dash (—): used to be the most important tell, but users have noticed that the models have evolved – and so have the patterns.
• The “quiet” obsession: quiet confidence, quiet truth, quietly growing – AI loves the word “quiet” in metaphors nobody needs.
• Forced hooks: “And honestly?”, “Here’s the kicker”, “And here’s the part most people miss” – sounds casual, but it’s formula.
• The rule of three: Adjectives always come in packs of three. “It’s fast, reliable, and efficient.” Every. Single. Time.
• The tapestry effect: Complex topics are described as a “tapestry”, “landscape”, or “testament.” Sounds sophisticated, says nothing.
• Therapist mode: “You’re not alone”, “You’re not broken” – unsolicited validation that’s just weird in a tech blog post.
• Structural self-narration: “Let me break this down”, “Here’s the breakdown”, “Let’s dive in” – the AI announces what it’s about to say before it says it.
• No typos. Ever. Zero. And that in itself is now suspicious.

Once you start noticing these things, you can’t stop. Like a smell you’ve picked up once.

Peter Steinberger Nails It

In a recent podcast with Lex Fridman, Peter Steinberger – the founder of OpenClaw – put it this way:

“I much rather read your broken English than your AI slop. You know, of course there’s a human behind it, and yet they prompt it. I’d much rather read your prompt than what came out. I think we’re reaching a point where I value typos again.”

He has a zero-tolerance policy: If someone replies on Twitter with AI-generated text, he blocks them. Immediately. No second chance. If a blog article starts with an AI-generated header image, he loses interest – because he thinks: If even the image is slop, why would the text be any better?

And then he says something that hit me:

“I experimented with creating a blog post with agents and ultimately it took me about the same time to steer the agent towards something I like. But it missed the nuances of how I would write it.”

That’s exactly it. You can steer AI toward your style. But it will never quite nail your style. Something’s always missing. The roughness. The edges. The personality.

Lex Fridman confirms the feeling too:

“I don’t know what that is, but I went through that too. I was really excited by the [AI generated] diagrams. And then I look at those now, and I feel like I feel when I look at Comic Sans as a font.”

AI infographics, AI diagrams, AI header images: A few months ago they were exciting, now they trigger an allergic reaction in a lot of people. “It’s a smell.”

My Own Struggle With This

I have to be honest: I feel the same way. When I see an article and after three sentences I think “AI slop” – I lose interest. And at the same time, I myself use a workflow where my articles go through AI.

That’s a real contradiction. And it becomes a problem when you think about why you write in the first place. I write so people read it. So it lands. So it has impact. But if the very readers I want to reach have a sensitive nose – then the AI polish is working against me.

It reminds me of something: When I give talks in English, I have a thick German accent. My grammar isn’t sophisticated. No British English, no polished TED Talk style. And yet – on several occasions people came up to me afterward and said: “That’s exactly what makes it likable. Do more of that.”

What I saw as “bad English” was, to others, authentic. Pareto English: 20% of the effort, 80% of the impact. And the remaining 20% of impact? That doesn’t come from more polish – it comes from the roughness itself.

Where Does It Smell the Most?

LinkedIn is the obvious battlefield. You scroll through the feed and smell it in every other post. People are developing outright aversions.

Emails to colleagues: If you send an AI-generated email and your colleague replies with an AI-generated answer – who actually had that conversation? You two, or your AIs?

Proposals and RFPs: Anyone writing B2B proposals is obviously going to use AI. But if the proposal smells like AI slop, the client might think: “They didn’t even properly engage with my problem. They just dumped it in.” Is that fair? Maybe not – there are hard evaluation criteria, and both sides use AI. But on a personal level, at first impression, when reading the intro? That’s where the feeling matters.

Recommendation

My call isn’t to stop using AI everywhere. My call is: Use it deliberately. And don’t unlearn how to write yourself.

Here’s my rule of thumb:

AI in:

• Writing documentation – better than nothing, nobody cares
• Letter to the insurance company – needs to be formally correct, no human reads that with love
• Technical summaries, internal meeting notes

Write it yourself:

• Emails to colleagues – just write them yourself, they’ll notice
• Personal blog – your voice matters here, not a polished version of it
• LinkedIn comments – especially where people have already developed an aversion
• Messages to friends – obviously, right?

Unsure::

• Proposals and RFPs
• Customer communication
• Sales emails

And if you do use AI? Then ask it at most to fix the worst typos and grammar mistakes. Not to “improve” the text. Not to “polish” it. Not to make it “clearer and more memorable.” Because that’s exactly where the roughness gets lost. The edges. The personal touch. The thing that makes you sound like you.

Or as Peter Steinberger puts it:

“There’s value in the rough parts of an actual human.”