7 giveaways that AI wrote it for you.
It's not one word like "delve." It's patterns, frequency and corporate polish. Here's how to spot them — and a prompt to fix them.
Liked this? New guides land here first. No spam, ever.
AI writing is detectable not by any single word but by the combination of patterns it keeps reaching for: the em dash as a thinking pause, the trio of adjectives, the fake-profound formula, the relentless positivity. This guide walks through the seven most common giveaways, and ends with a copy-paste prompt you can use right now to strip them out and make your AI writing sound like an actual human wrote it.
Why the giveaway is never just one word.
The "is this AI?" conversation keeps landing on individual words. "Delve" became the joke. "Tapestry" did a stint. "Groundbreaking" gets flagged in every detector. But if you swap out those words and leave everything else the same, the writing still sounds like AI. Because the tell isn't vocabulary — it's structure, pattern, and rhythm.
AI has defaults. It defaults to three-part lists. It defaults to wrapping up with a neat conclusion. It defaults to enthusiasm. It defaults to balance. Real humans don't do any of that consistently, because real humans have opinions, get distracted, and go off on tangents. Once you know what to look for, you can't unsee it. Here are the seven patterns worth knowing.
The seven giveaways.
1. The em dash obsession
AI loves the em dash because it signals a thoughtful pause — a moment of consideration — before the real point lands. Except real people don't write like that. We use commas. We use full stops. We start a new sentence. The em dash is doing a specific job in AI writing: it's performing depth. One is fine. Three in a paragraph is a pattern.
2. The fake profundity formula
"It's not about X, it's about Y." You've seen this everywhere. "It's not about working harder, it's about working smarter." "It's not about the destination, it's about the journey." Sounds deep. Says nothing. This is AI's go-to move for sounding philosophical without actually committing to an idea. Real opinions are more specific and more uncomfortable than this.
3. Everything comes in threes
"Clear, concise, and compelling." "Bold, brave, and brilliant." "Fast, flexible, and future-proof." AI has absorbed enough marketing copy to know that things come in threes. It's not wrong that three-part lists sound good. But when every single point in a piece of writing arrives in a trio, that's a pattern, not a choice.
4. The Wikipedia voice
Balanced. Thorough. No strong opinion. AI defaults to presenting all sides fairly, which produces writing that sounds like an encyclopedia entry: technically accurate, completely inert. Real humans pick a side. They go off on tangents. They contradict themselves occasionally. They don't wrap everything in a neat bow at the end. If a piece of writing could have been written about any person, product or service, that's the Wikipedia voice at work.
5. Cliché on cliché
"Rich tapestry." "Rapidly evolving landscape." "In today's fast-paced world." "Game-changer." "Unlock your potential." When did you last actually say any of those out loud? These phrases exist in AI writing because they exist in the training data — in press releases, in corporate comms, in a thousand SEO articles from 2015. They're not wrong, exactly. They just belong to no one.
6. Enthusiasm for everything
Even the mundane is "exciting." Even the obvious is "powerful." Even the smallest update is "groundbreaking." AI has one volume setting and it's hype, because hype is what gets positive feedback in the content it's learned from. Real humans don't find everything exciting. Sometimes something is fine. Sometimes it's fine is the honest take, and the honest take is what sounds human.
7. The setup that goes nowhere
Big bold opening claim. "The key is..." Generic conclusion that could apply to literally anything. This is the structure AI reaches for when it hasn't been given enough specific material to work with. The hook is striking, the middle is competent, and then the ending just... resolves. Neatly. Without actually saying anything you couldn't have predicted from the title.
What to do about it.
The fix is not to write everything yourself or avoid AI altogether. The fix is to give AI clearer constraints before it starts — and a specific prompt to run when the output needs pulling back toward human.
The most useful thing to add to any AI writing prompt is a "never use" list. Not just words, but structural patterns: no em dashes, no three-part lists unless they're genuinely the best format, no conclusions that restate the opening, no sentences that could apply to any topic. Give AI evidence of what your writing actually sounds like. And when it gives you something that's 80% right and 20% AI-polish, give it one specific note rather than starting over.
Here's a prompt you can paste directly into ChatGPT or Claude. It's designed to take AI-generated text and pull it back toward real, human writing. Use it on anything that sounds a bit too polished.
Rewrite this to sound like a real human wrote it — not a content generator, not a marketing brochure.
Rules:
— Use clear, simple language and short sentences. Active voice.
— Talk directly to the reader using "you" and "your."
— Remove all em dashes. Use commas or a new sentence instead.
— Remove any three-part adjective lists ("bold, brave and brilliant" etc).
— Remove fake-profound formulas like "it's not about X, it's about Y."
— Remove all clichés: "game-changer," "rapidly evolving landscape," "in today's world," "unlock," "tapestry," "groundbreaking," "powerful," "exciting" (unless genuinely warranted).
— Don't wrap up too neatly. Real writing sometimes ends mid-thought or with an open question.
— If the original has an opinion, sharpen it. Don't hedge.
— Keep bullet points if they're genuinely the clearest format, but don't force structure where flowing prose works better.
Here's the text to rewrite:
[paste your text here]
In Claude, create a new Project and paste this prompt into the Project Instructions. Every conversation you start inside that project will automatically run with these rules — no copy-pasting required. Just open the project and start writing.
The bit that's still yours to add.
The humaniser prompt strips out the obvious tells. But there's one thing no prompt can add for you: the specific, real moment that makes writing feel true. A particular client conversation. The exact number. The thing that happened on Tuesday. AI can structure around your material. It can't invent the material itself.
The pattern of AI writing — the polish, the balance, the relentless optimism — is a symptom of having nothing specific to say. Give it something specific and most of the giveaways disappear on their own.