There’s a new credibility tax on written content, and most people don’t realise they’re paying it.

You might be doing the actual thinking (as you should!). You might have a genuinely interesting idea. But if your writing smells like AI, readers will skim it, dismiss it, or outsource their reading of it back to AI. Which is a deeply ironic loop.

The problem isn’t using AI to help you write - you should use it as your editor and critique. The problem is when you let AI sand off everything that made your writing sound like you in the first place.

Neo Aplin and I spent an entire episode of How I AI going through exactly this. Here’s the Cliff’s Notes.

The phrases that “out” AI immediately

Some phrases have become AI’s fingerprints. If your writing contains more than a handful of these, people will notice - even if they can’t explain why.

Empty openers are the most obvious offenders: In today’s fast-paced world. In an ever-changing landscape. When it comes to... These phrases say nothing. They exist purely to delay the point, and AI loves them.

Then there are the fake-profound hooks. Here’s the truth. Here’s the thing. Here’s the uncomfortable truth. I’ll confess: I used “here’s the thing” constantly before AI turned it into a cliché. Annoying.

And the “it’s not X, it’s Y” construction - corrective antithesis, technically. AI deploys this move constantly because it sounds clever. It does, the first time. By the hundredth time, it just sounds like a language model that watched too many TED talks.

The rule of three (and why AI can’t stop)

AI defaults to three examples, three reasons, three steps. Always exactly three. There’s something almost compulsive about it.

Real humans use one strong point. Or two. Or seven, if the situation calls for it. The relentless triplet is one of those things you might not consciously clock, but your brain registers it as slightly off.

Tell AI explicitly: don’t default to three. Use as many examples as the point actually needs.

The sentence length problem

AI writes in eerily consistent sentence lengths. You might not notice it consciously, but you feel it - that sensation where something reads fluently but feels somehow inhuman.

The fix is deliberate variation. Short sentences land harder. Longer ones give you room to build an idea, add nuance, or let something breathe before you pull the reader to the next point. Mix them up and suddenly the writing has a pulse again.

Words worth banning

Delve is the poster child. Research from 2023 found that the most prolific academic authors that year over-indexed heavily on the word. The AI fingerprints were already showing up in peer-reviewed literature. Peer-reviewed literature! Nowhere is safe.

Other words to consider pulling: quietly (it’s everywhere now), ultimately, synergistic (if this word has never left your mouth in conversation, why is it in your writing?).

My advice: build your own banned list. The words that grate on you are probably the ones that don’t sound like you. Start there, and keep adding as you go.

The worst combination

Using AI to write everything for you is one problem. But the thing that actually damages your credibility is subtler: you do the real thinking, you write something genuine, and then you run it through AI for a quick “cleanup” - and it AI-ifies the whole thing.

Your ideas, nobody’s voice. Or rather, everyone’s voice averaged together into a beige paste.

If you would like an anti-AI slop prompt that we created at Inventium.ai, grab it here. It covers everything above and more, and it’s built to be customised. Read it, add your own pet peeves, and run your writing through it before you hit send.

And if you know someone who’s doing the thinking but letting AI undo it, send this their way. Consider it a public service.