If you’ve ever asked AI to “clean up” your writing and felt both impressed and slightly deflated afterward, you’re not alone. Most people don’t talk about it, but there’s an insecurity that creeps in when the model’s rewrite looks smoother than your original draft, and you begin to start questioning your abilities.

I think it's important to address what’s actually happening:

Cause → effect:

• You give the model a paragraph.
• It hands you back something more polished.
• You accept the change because it feels “better.”
• That tiny approval nudges you to trust the model more than yourself.
• Over time, you stop trusting your own instincts at all.

The problem with this is people tend to draw conclusions like “I guess the AI version is the right one” or "I guess my writing isn't that good after all"

But that’s a false signal. The model isn’t producing “better,” it’s producing “statistically typical.” And when you treat typical as superior, your writer’s intuition slowly atrophies. You begin writing not from conviction but from fear of being “wrong.” You start drafting in ways that appease the model instead of ways that express you. This then hinders your writing, harms your SEO (because do you honestly think search engines are okay with an entire internet built on regurgitated AI slop?) and ultimately helps no-one.

The Collective Downside of the Generative Feedback Loop

Zoom out, and there’s a broader problem forming.

Every AI-polished piece of writing eventually ends up on the internet. And what does the next generation of AI train on?

The internet.

This causes a loop. Cause → effect → consequence:

1. People rely on AI to smooth their writing.
2. Smoothed, generic writing floods the web.
3. Future models train on that content.
4. Models get even more generic.
5. Writers rely on them even more.

We’re building a recursive loop where AI writes like AI because it’s trained on AI written by people who are trying to write like AI.

This is how human texture gets erased at scale. When everyone optimizes for “clean,” we lose the weird, specific, slightly imperfect edges that signal there’s an actual person behind the words.

The Imperfection Advantage

I really want everyone to stop underestimating their abilities. Your imperfections are what makes your writing unique and amazing and the only thing AI can’t replicate.

Your natural phrasing, the unexpected analogy, the slightly unconventional flow are markers of originality. They’re what make readers pay attention. They’re what make your writing recognizable. You can 100% tell an AI generated analogy or story over a lived human experience.

AI can reproduce patterns, and pretend to have a perspective. It can borrow tone, but even with the world's knowledge, it can't produce something that sounds genuinely human.

Which is why the most valuable thing you can preserve in an AI-saturated world is the thing AI is worst at copying: your distinct, imperfect, specific voice.

Why “Improve This” Always Turns Into a Rewrite

At this point, the natural question is: Why does AI keep completely rewriting my stuff when I only asked for an improvement?

LLMs are trained to generate, not preserve. They don’t know which sentences are intentional, which phrasing carries your personality, or what rhythm represents your voice. To the model, everything is editable because nothing is sacred.

The model’s job is to predict the next most statistically probable token.
Your job is to have a point of view.
These objectives are fundamentally different.

So when you say “improve this,” the model interprets it as:

“Rewrite this in the safest, smoothest, most statistically common style possible.”

Maintaining your voice and personality was never part of the model’s optimization target.

If you want to really see this dynamic play out, try this:

Mini-exercise:

1. Write a paragraph that sounds like you.
2. Paste it into an LLM and say: “Please improve this.”
3. Watch how much gets rewritten vs slightly polished.
4. Take what it "improved" and start a new chat. Say "please improve this."
5. It will continue rewriting and rewriting, improving on its own "improvements" forever.

This is why people end up trusting AI more than themselves. The model isn’t actually rewriting anything because it's inherently bad. It’s just doing exactly what it was trained to do: overwrite everything.

The Fix: Be Brutally Specific About the Job

If you want AI to support your writing instead of replacing it, you need to assign the model a job with a clear boundary. Never (seriously, never) say “make this better” or “improve this.”

That’s handing over the steering wheel.

Instead, constrain the scope:

• “Critique this for clarity. Point out confusing sentences only.”
• “Check for passive voice and flag only what hurts readability.”
• “Suggest 3 stronger verbs for this section. Do not change structure.”
• “Highlight sentences that feel too generic.”

(Many more prompt suggestions shared further in this article)

That level of specificity forces the model to behave like a specialist, an editor, a clarity checker, or a strategist over a ghostwriter intent on writing everything from scratch. This is how you stay in control of the writing while still using AI as a partner.

A Real-World Workflow

The most effective (and least voice-destroying) way to use AI looks like this:

Use AI to sharpen your strategy → YOU write → Use AI to critique and refine.

AI should never set your direction, your strategy, or your topic.

It’s here to help you think more clearly about the direction you’ve already chosen.

Use AI to Strengthen Your Strategy

Think of this step as “structured thinking with a partner,” this way, you're delegating strategy, you're not making it write for you. You don't want to outsource 90% of your job, only aid 50% of it. This still brings speed and quality while maintaining voice, originality, and maintaining SEO value.

Example: You bring the ideas. AI helps you stress-test them.

AI is useful here for:

• Spotting gaps you didn’t notice
• Highlighting redundancies
• Suggesting alternate angles or reader paths
• Challenging unclear logic
• Helping you decide what order is most persuasive

Don't use AI to invent your strategy. Instead, pressure-test the strategy you already have with the help of AI.

This is important because, yes, LLMs lean heavily on existing patterns in the data they were trained on. If you ask them for ideas from scratch, they’ll often serve you recognizable, overused formats.

So the smarter workflow is:

Bring your rough outline or thinking
Ask AI to analyze it, challenge it, or expand it
Use the outline to write the entire thing in your voice and framing

YOU Write the First Draft

This part is non-negotiable.

Your voice, taste, perspective, and lived experience only show up if you write the draft. You can give AI all the brand voice instructions, documents and samples you want but it will never nail it, and it will not speak from the experience you have.

Use AI as a Critic

Once you’ve written the draft, AI becomes useful again to critique and pressure-test - but I repeat, not rewrite.

AI can help you identify:

• unclear or muddy sections
• arguments that need strengthening
• logical gaps
• tone that slips into generic
• pacing issues
• transitions that feel abrupt

But the key is:
AI points out the issues.
You rewrite the fixes.

This keeps your voice intact while getting the editorial support almost every writer wishes they had. At this point, you can have it draft you 3-5 examples. Then pick from those and rewrite them but keep you're original voice.

Pitfalls: The Fastest Ways to Flatten Your Voice

Even with the right workflow, people still fall into a handful of predictable traps when using AI for writing. They may seem small and inconsequential, but trust me, they add up, and they’re exactly how your voice erodes without you noticing.

Pitfall 1: Asking AI to Do the Thinking For You

If you start with “Generate ideas for me,” you’ll get the same recycled structures and angles everyone else gets. If you want your content to rank and stand out it needs an original POV.

Better: Bring your own angle, even a messy one, and let AI help you pressure-test it.

Pitfall 2: Letting AI Touch the First Draft

This is the quickest way to lose your voice.

Once the model writes a sentence, you’ll subconsciously adapt to its tone instead of writing like yourself.

Your writing becomes a vague imitation of what AI thinks writing should sound like.

Rule of thumb:
If the cursor hasn’t moved under your hands yet, AI shouldn’t be involved.

Pitfall 3: Asking for “Improvements” Instead of Specific Feedback

“Improve this” signals one thing to an LLM:

Erase the human. Make it sound like everything else.

Broad prompts = generic output.
Generic output = voice decay.

Later, your prompt list will fix this by giving you targeted, job-specific asks.

Pitfall 4: Treating AI as a Voice Substitute Instead of a Thinking Partner

If you use AI to avoid writing, it ends up becoming a crutch and it becomes harder and harder to write on your own.

This leads to:

• Overdependence
• Loss of confidence
• Atrophy of intuition
• A writing style that slowly becomes indistinguishable from machine output

AI can support your writing, but it cannot be your identity.

Pitfall 5: Believing Smoothness = Quality

LLMs are masters of fluency, but that doesn't translate to insight or taste. AI can produce many smooth paragraphs that say nothing once you pick it apart. Remember, your voice is defined by the quality of your thinking and not the number of perfect sentences you have.

Pitfall 6: Editing to Match AI Instead of Editing to Match Yourself

This is sneaky.

People rewrite their own sentences to sound more “AI-clean.”
They start mimicking the model’s cadence. Before you know it, you’re writing like a ChadGPT, even though it wasn't your intent.

Pitfall 7: Forgetting That AI Is Trained on the Past, Not the Future

AI is trained on what’s been written - not what should be written.

So if you rely too heavily on its suggestions, you’re anchoring your creativity to yesterday’s patterns. Your goal is to stand out.

Subscriber-Only Prompt List

Try out the workflow above, see how it works for you. When you're ready, use this list of master prompts as a checklist when you write something.

These are the exact prompts that keep AI in its lane and make your process 10× easier.

If you want them, unlock the list below.

Prompts for Writing Improvements

You’ve read the arguments, and you got the workflow.
Now the important part is making AI your partner instead of a replacement of you. Here are quick prompts you can use as a checklist every time you have content you want feedback or revisions on.

These short but helpful prompts will make sure AI stays in its lane and improves and assists, while keeping your originality and voice in tact.