The irony is not lost on me

I work with AI writing tools every day. I test them, review them, and occasionally use them to help draft content. So I'm acutely aware that writing an article about how to spot AI content is a bit like a magician explaining how the card trick works. But you know what? Someone should, because the tells are everywhere and most people can't see them.

Once you learn to spot AI writing, you can never unlearn it. It's like seeing the arrow in the FedEx logo. Except instead of a clever design detail, it's the slow realisation that about 30% of the internet is now being written by algorithms pretending to be people.

The em dash epidemic

This is the easiest tell. Raw, unedited AI text is absolutely riddled with em dashes. You know the long dash? The one that looks like this: "The tool was impressive, powerful, and intuitive." Read any ChatGPT output and count the em dashes. It's wild.

Human writers use em dashes occasionally. AI uses them in almost every paragraph, sometimes multiple times per sentence. If an article has more than two em dashes per page, your spider senses should be tingling.

Real humans tend to use commas, full stops, or just start a new sentence. We're lazy that way. AI has been trained on a lot of formal writing and it loves its em dashes the way a barista loves oat milk.

The vocabulary of the damned

AI has favourite words. Specific words that appear in AI output with a frequency that borders on obsessive: - "Landscape" as a metaphor. "The AI landscape." "The competitive landscape." No human has unironically said "landscape" in a non-geographical context since 2019. - "Delve." Nobody delves. Nobody has ever delved. If you see the word "delve" in any piece of writing, it was either written by AI or by a person who has been reading too much AI output. - "Robust." AI thinks everything is robust. Robust features. Robust solutions. Robust frameworks. Real humans say "solid" or "strong" or just... describe the thing. - "Tapestry." As in "the rich tapestry of." Stop it. - "It's worth noting." AI loves this phrase. It's the AI equivalent of clearing your throat before saying something. - "Stands out." Everything stands out. Nothing blends in. Every tool "stands out in a crowded field." - "Comprehensive." AI describes everything as comprehensive, including things that are demonstrably not comprehensive.

If a piece of writing uses three or more of these words, it was almost certainly AI-generated, or written by someone who's absorbed so much AI output that they've started writing like one.

The structure tells

Opening with "In today's..."

"In today's fast-paced world." "In today's digital age." "In today's competitive market." If an article opens with "In today's..." followed by any adjective-noun combination, close the tab. You're about to read 1,200 words of AI-generated content that will tell you nothing you didn't already know.

The list of three with escalating grandeur

AI loves to list three things, and it loves to make each one sound slightly more impressive than the last. "This tool is efficient, powerful, and transformative." "The platform is intuitive, versatile, and truly game-changing."

Real humans don't escalate like this. Real humans say things like "it works well and the interface is decent."

The balanced paragraph pattern

AI paragraphs follow a predictable structure: claim, explanation, and a neatly wrapped conclusion. Every single one. It's almost hypnotic in its consistency. Real human writing is messier. Paragraphs vary in length. Some thoughts don't get conclusions. Sometimes you just move on to the next point without wrapping up the previous one.

The diplomatic both-sides treatment

Ask AI to compare two things and it'll find roughly equal praise for both. "Tool A excels at X while Tool B shines at Y." Real humans have preferences. Real humans say "Tool A is better" without immediately rushing to reassure Tool B's feelings.

The tone tells

Everything is exciting

AI writes like everything it's describing is the most fascinating thing it's ever encountered. There's an unbearable positivity to raw AI output. Everything is "exciting," "impressive," or "remarkable." Human writers are more cynical. We say "it's fine" or "it works" or "I mean, it's alright I suppose."

Nobody is ever wrong

AI doesn't criticise. Or rather, it criticises in the gentlest possible way. "Some users may find the interface slightly challenging." What a human would say: "The interface is confusing and I wasted twenty minutes trying to find the settings menu."

Hedging everywhere

"This could potentially help users to..." Just say "this helps." AI hedges everything because it's trained to avoid saying anything definitive that might turn out to be wrong. The result is prose so non-committal that it essentially says nothing.

How to check with tools

There are AI detection tools like GPTZero, Originality.ai, and even Grammarly has some detection features. Here's the truth about them: they're unreliable.

False positive rates are high enough that using them as definitive proof is irresponsible. I've seen these tools flag clearly human-written academic papers as AI-generated, and I've seen them give AI-generated marketing copy a clean bill of health.

Tools like Grammarly and ProWritingAid are more useful for cleaning up your own writing to not accidentally sound like AI. They won't tell you if something was AI-generated, but they will flag the telltale phrases and patterns that make text sound artificial, which is arguably more useful.

The uncomfortable truth

Here's the thing nobody in the AI detection industry wants to admit: detection is a losing battle. AI models are getting better at producing human-like text faster than detection models are getting better at catching them.

Within a year or two, the textual tells I've described above will probably be gone. AI will learn to vary its sentence structure, avoid its favourite words, and produce text that's indistinguishable from human writing. Some argue it's already there.

The real question isn't "was this written by AI?" It's "does it matter?" If an article gives you accurate information, helpful advice, and is actually pleasant to read, does it matter whether a human or an AI wrote it?

I think it does matter, because AI writing lacks genuine experience. An AI can tell you that a tool is good, but it can't tell you that it stayed up until 2am swearing at it before the configuration finally worked. That lived experience is what makes writing worth reading.

How to make your own writing less AI-like

If you use AI to help with drafts (no shame in that), here's how to make the output sound human: Delete every em dash. Replace with commas or full stops. Search for the forbidden words listed above. Replace them with something a normal person would say. Add imperfections. Real writing has tangents, opinions, and the occasional sentence fragment. Leave them in. Disagree with something. AI never disagrees. Having an actual opinion is the most human thing your writing can do. Cut the conclusions. AI wraps everything up in a tidy bow. Real articles sometimes just... stop.

Now, was this article written by AI? I'll leave that as an exercise for the reader.

(It wasn't. But the fact that you have to wonder is kind of the whole point.)