How to spot AI in posts

A lot of what you scroll past now is written by AI, and it tends to sound the same. Smooth, confident, and saying very little. Once you know the tells you can't unsee them, and the feed starts to look a bit different.

One thing to keep in mind before we start. These tells are clearest when someone has used a standard AI tool on its default settings, with no instructions and no context. Give the same tool your tone and a bit of background and most of them disappear. So take this as a guide, not a lie detector.

Common tells across most AI

• An emoji on every single line. People don't do this. We might chuck one or two in. AI lines them up like a little parade, and the green tick and red cross are its favourites.

• Far too many words. Ask AI something simple and it writes an essay. You get to the end and realise the whole thing could have been one sentence. It's trained to be thorough, and it cannot help itself.

• Nobody home in the comments. Plenty of likes, a busy thread, and the poster replies to no one. Or the replies come back three paragraphs long for what should have been a "cheers mate". Dead giveaway.

• The same phrases on repeat. "At its core." "It's crucial to remember." Once you notice them you see them everywhere, and you're a little annoyed every time.

• Stories that are a bit too perfect. A tidy lesson, a neat ending, a stranger who says exactly the right thing at exactly the right moment. Real life is messier. AI loves a fairytale.

• Openers that say nothing. "In today's fast-paced world." "Let's be honest." "Here's the thing." All throat-clearing, no point. People who actually know their stuff just start.

A few more that give it away

• Em dashes everywhere. The classic. AI scatters them through a post like confetti. Most of us don't reach for an em dash mid-post, so a whole run of them in suspiciously tidy sentences is a tell.

• The "not this, but that" thing. "It's not about the software. It's about the people." Sounds deep, says nothing. One is fine. Three in a post and you're reading a machine.

• Everything in threes. Faster, cleaner, smarter. Build, measure, learn. Real writing isn't that neat. AI defaults to it because it sounds good, and it sounds good because we've all heard it a thousand times.

• The needy question at the end. "What's your take? " It never quite connects to the post. It's there because AI knows posts are meant to end on a hook, like a kid who's learned the shape of a joke but not the punchline.

• Too clean. No typos, no half-thought, no personality. Real people writing between meetings leave a few rough edges. Perfectly smooth top to bottom, with nothing odd or specific in it, usually had help.

None of these nails it on its own. It's when a post hits four or five at once that you know. And really, the question isn't "is this AI" so much as "did a human bother to show up".

Tells specific to Claude

For the Claude users trying not to get sprung by the boss:

• Suspiciously tidy lists and headings. Everything perfectly balanced and aligned. Rough up the formatting so it looks like a person did it in a hurry.

• "In conclusion". Claude loves a formal sign-off. Nobody ends a Slack message like they're wrapping up a uni essay.

• Relentlessly nice. Claude writes like the loveliest coworker you've ever met. Thoughtful, warm, considered, every single time. After a while it gets suspicious.

The real giveaway is the setup

Default AI is easy to spot because it's never met you. So, it falls back on all of the above, and the less you give it, the more obvious it gets.

Set it up properly though, with your voice, your context, and a human reading over it before it goes out, and the tells fade. That's the whole game.

Using AI without losing yourself

The point isn't to trick anyone. It's to get a hand without sounding like a press release.

• Give it your context. Tell it what you do, who you're talking to, how you write, and show it something you've actually written.

• Read every word before it goes out. Cut the stock phrases, the em dashes, the needy question. Add something only you'd know: a real number, a real project, an actual opinion.

• Use it for the hard bit, not the whole thing. Beating a blank page, fixing clumsy grammar, trimming a rant. That's just a colleague reading your draft. The post is still yours.

Do it well and nobody goes hunting for tells, because there's nothing to find. The rhythm is yours, the opinions are yours, the odd rough edge is yours. You didn't hand a prompt to a machine and paste back the result. You worked out what you wanted to say, then used a tool to help you say it better.

Because this was never really about catching AI. It's about whether there's a person behind the words. The posts worth reading still come from someone who bothered to think, to have a view, and to put their name to it. AI can tidy the grammar and speed up the boring parts, but it can't decide what you believe. Get that right and it won't read like AI, or like someone hiding it. It'll just read like you.