You can tell when something was written by AI almost immediately. Because it sounds familiar in the worst possible way. The sentences feel pre-approved. The pacing is clean but lifeless. The tone is confident without taking any risks. Your brain can pick it up before you even process it.

One of the clearest signs is a sentence structure that has silently infected the internet: “It’s not this. It’s that.” It pretends to be clarifying, in reality it’s contrast for contrast’s sake. A highlighter over an idea that doesn’t need it.

Once you see it, you see it everywhere, because AI loves it and humans have stopped questioning.

That phrase almost never appears alone. It comes with a whole family of expressions that serve the exact same purpose: to slow the reader down, inflate the text, and make everything sound reassuring. “Let’s break it down.” “That said.” “The key point is…” “This raises the question.” “Ultimately.” “Like never before.” “In this fast-paced world.” “Finally.” “Think of it this way.” “That’s why it matters.” “If done right.” “A common mistake is…” “The reality is…” “It’s not about being perfect, it’s about moving forward.” “Whether you’re a beginner or an expert.” “This changes everything.” “In conclusion.”

None of these are wrong. They’re simply dead. Linguistic crutches. AI uses them by default because it was trained on the safest, most average version of how people write.
The exact center of the curve. This is where many people get it wrong.

The problem is that AI writes by default.
And flaws always sound the same as everyone else’s.

AI doesn’t have a voice. If you don’t cut it, you get politeness when you want forcefulness, balance when you want position, explanation when you want punch.

That’s why so much AI-generated content feels interchangeable. Different topics, same rhythm. Different authors, same temperature. It sounds like background noise.
Your brain ignores it just like it ignores banner ads.

The solution isn’t writing a better prompt in the moment. Permanent directives are the difference between guiding the model once and defining how it always behaves. They tell AI what not to say, not just what to say. They set the pace, the forbidden constructions, the tolerance for different points of view, and how far sharpness is allowed before someone softens it “for clarity.”

Without that, every piece of writing ends up in beige. Human writing isn’t optimized. It’s inconsistent. People cut transitions. They repeat words for emphasis, not for elegance. They leave sentences short when they’ve finished thinking, not when a style guide tells them to stop. Sometimes it sounds impatient, biased, unfinished. That roughness is exactly what flaws smooth out.

If your AI-assisted content could be swapped with 90 percent of what’s on the internet and nobody would notice, that’s not a problem with the model. It’s because you haven’t told it what not to do.

The uncomfortable truth is this: sounding human now requires more intention than sounding generic. You have to actively discard phrases that feel helpful but don’t add anything. You have to stand firm and tell the AI ​​what it’s not allowed to write. You have to choose recognition over perfect understanding. Friction over polish.

The AI ​​will continue to write exactly like the internet sounds.

If you don’t want to be part of that 90 percent, you have to force it to

Here is a permanent directive prompt for the sake of example.

WRITING VOICE DIRECTIVE — PERMANENT

BANNED PHRASES. Never use: “Let’s break it down.” “That said.” “The key takeaway is.” “This begs the question.” “At the end of the day.” “Now more than ever.” “In today’s fast-paced world.” “Ultimately.” “Think of it this way.” “Here’s why that matters.” “When done right.” “A common misconception is.” “The reality is.” “This isn’t about X, it’s about Y.” “Whether you’re a beginner or an expert.” “This changes everything.” “In conclusion.” “It’s not X. It’s Y.” “Dive into.” “Leverage.” “Unlock.” “Game-changer.” “Seamless.” “Elevate.” “Empower.” “In the landscape of.”

TONE. Always:

• Direct. No warm-up. No easing the reader in.
• Opinionated. Take a position. Don’t balance for the sake of balance.
• Short sentences when the point is made. Stop there.
• Impatient when the idea demands it.

RHYTHM. Never:

• Pad a thought that’s already complete.
• Use a transition phrase to buy time.
• Repeat an idea in a different way “for clarity.”
• End a section with a summary sentence.

STRUCTURE. Always:

• Cut the intro. Get to the point immediately.
• Let ideas land before moving to the next one.
• Use short paragraphs. One idea per block.
• Leave white space. Don’t fill it.

VOICE. Never:

• Sound like a blog post.
• Explain what you’re about to say before saying it.
• Soften a hard truth to sound balanced.
• Use theatrical contrast (“It’s not X. It’s Y.”) unless it genuinely earns it.

THE TEST. Before publishing, ask:

Did I say something that might make someone uncomfortable? Good. Keep it.
Could this be swapped with 90% of content online and nobody would notice? If yes, rewrite.
Does it sound like it was written by a specific person, or by a machine averaging the internet? If the latter, rewrite.