When Not to Use AI: Real Mistakes Creators Make, and How to Fix Them Fast

Learn the exact moments where using AI quietly makes your work worse instead of better.

People keep asking the same question again and again: “When should I use AI?”

Honestly, that’s the easy question.

If a task is boring, repetitive, low value, or something you keep doing every day without thinking, AI is perfect for that.

But Nitin, what can AI do? Well, it can automate the tedious parts, generate drafts, test variations, break down research, or write the code skeleton so you don’t have to repeat the same things again.

I’ve even written a couple of posts sharing practical use cases for NotebookLM, Nano Banana, ChatGPT Atlas, Google AI Studio, and more to help you how to use AI in your everyday workflow.

But the question no one asks is the one that matters the most: “When does AI quietly make my work worse even if it makes it faster?”

This is where most creators mess up today.

They use AI in places where they shouldn’t, and instead of saving time, they spend hours fixing the mess or publishing something that feels lifeless. And the saddest part is they don’t even notice that their work is slowly losing its identity.

Well, that’s what I’m going to share with you today.

Below are the everyday creative situations where AI looks helpful on the surface but actually slows you down, kills your voice, or feeds you nonsense. And right after each one, I’ll show you the exact workflow you should use instead.

Note: If you found this post valuable or learned something new, consider becoming a paid subscriber to this newsletter. It’s the best way to support my work and keep this kind of content coming.

With that said, let’s get started.

1. Don’t use AI to write anything where your voice matters

Let me be honest here.

I read a lot of posts online from creators on Substack, Medium, LinkedIn, and multiple websites. And after two paragraphs, I can easily tell whether it’s written by a human or an LLM.

The signs are obvious, like:

• Short robotic sentences.

• Random sudden punctuation choices.

• The same predictable words that every AI model loves to throw at you.

And here is what most creators do. They write a vague prompt like “Write a post about {topic} in my style” and expect magic.

And what you’ll get in return is obvious: a generic post with barely any real value.

Look, even after spending more than two years training these models to understand my tone, even after giving detailed instructions, even after telling it to keep my natural rhythm and informal flow, it still slips in those typical AI phrases that break the authenticity.

Here’s an example:

And so, you need to understand that your voice is the only thing that truly separates you. You cannot outsource that to any LLM models no matter how good they are.

What to do instead: use AI around your writing, not inside it.

• Ask AI to give you ten structures or angles.

• Pick one that feels right.

• Write your messy first draft on your own.

• Feed that draft to AI to tighten transitions or remove repetition.

• Rewrite again in your voice.

• Then edit or modify further.

Here’s a simple prompt that works well for me:

Rewrite this but keep my emotional tone and informal rhythm. Do not shorten sentences and keep the flow natural. Please don’t include em dash in the generated sentences, and don’t rephrase it completely.

---

2. Don’t use AI to brainstorm original ideas in your niche

You know, I have been writing online for more than five years now.

I’m also a developer learning AI every single day so I can teach people real workflows that actually help them work faster. And even with that experience, when I ask an LLM for ideas, I get the same generic stuff everyone else gets.

You must have seen it too.

• The overuse of colons in every title.

• The same ten ideas that every YouTuber has already turned into a video.

• The same angles recycled in different words.

Even try writing a detailed prompt, and the ideas will be mostly the same.

It isn’t your fault. AI gives you the average of the internet, not the leading edge.

So instead of that, simply use AI to research, and please:

• Go to Reddit.

• Go to Hacker News.

• Go to YouTube, and even see the comments.

• Go to X, and even see the replies about what people are talking.

And find real frustrations people talk about or find questions people ask repeatedly.

What next? Write based on your own experience and your own experiments after doing the research.

---

3. Don’t use AI to edit emotional or personal writing

This is something I’ve learned the hard way.

Whenever I write anything personal, it’s never perfect. It has messy lines, awkward parts, long sentences, and little moments that don’t sound “clean” but they sound like me.

And the funny thing is, that’s exactly why people connect with me on Substack.

But the moment you paste something emotional or personal into an LLM and say “Make it better”, it doesn’t make it better. It makes it something weird, and the worst part is that it removes your emotions or feelings.

I tried adding one of my Substack Notes directly into ChatGPT and asked it to make it better.

And it actually made it worse.

Why? AI treats your raw feelings or emotions almost like grammar issues, so it trims the emotions, replaces your little quirks with celebrity-quote language, and suddenly the whole content looks different.

I’ve tried this enough times and seen that the posts surely become more accurate and readable, but they lose the emotions.

So instead of letting AI rewrite your emotions, use it only for two things that don’t touch the soul: fixing confusing lines or grammar issues.

A simple rule I follow now: if a paragraph came from my heart, I don’t let AI rephrase it. Sometimes I just ask it to clear grammar mistakes or fix the sentences, but not rephrase it.

---

4. Don’t use AI to respond to clients/customers directly

This one is way more practical than people think.

I can see that a lot of creators and even individuals are connecting AI to Gmail so it replies automatically.

But here’s the problem: clients don’t write the same kind of emails every day, and not every client is the same.

Sometimes they’re stressed, sometimes they’re confused, sometimes they’re asking something small but the tone tells you they’re worried about something bigger underneath.

Or sometimes they are asking for help, partnerships, or something different that AI cannot catch, and it often gives robotic responses.

And one wrong word can look careless or rude or overconfident or too formal. So I use AI sometimes to suggest what I can reply, but I never automate the complete process no matter how much better the LLMs become.

Now here’s what I do, and it works really well.

Whenever I get a long or tricky email, I don’t ask AI to “reply to this”. That always gives a robotic answer. Instead, I write 2–3 messy lines explaining what I want to say and ask AI to turn those lines into a cleaner version.

For example, here’s an email that I’ve received:

And then I gave it to ChatGPT asking for some info before providing the exact pricing.

You can see the response is good, but it uses an em dash everywhere, and I don’t like the ending sentences so want to modify before sending.

Yes, the final edit is always mine. I read the email again, feel the situation, and then rewrite AI’s output the way I would actually say it if I was talking to the person.

This way I get the speed of AI, but the voice is still mine.

But why make the process so tedious? Well, I’m working with clients, and they are paying me to get the work done.

And I believe sending an email is the first impression, and I don’t want them to read it and feel like “This sound like an AI”.

---

5. Don’t use AI for private documents, contracts, letters, or screenshots

Let me be honest, I don’t upload private stuff because I don’t trust any of these AI tools with sensitive data.

I know that every model is getting trained on huge datasets, and every company is hungry for more data.

In India, most popular AI companies (OpenAI, Anthropic, and even Google) are literally giving paid plans for free just to get more input from users. That should tell you enough.

So here’s the practical version of what I do: if something has bank details, my passwords, legal numbers, client info, screenshots with emails or phone numbers, or anything I would never show a stranger, I don’t upload it.

I handle it myself or rewrite the sensitive part before asking anything.

I only upload text that I’m comfortable losing control over. It’s a simple rule, and it may save me with a lot of trouble.

---

6. Don’t use AI when instructions are vague

This is the mistake almost everyone makes.

They write vague prompts into ChatGPT, press Enter, and expect a miracle. And then they complain that “AI doesn’t understand”.

I’ve seen and experienced that AI can only work well when you think well and prompt well.

If your instructions are vague, the output will obviously be vague. If you don’t know what you want, the model won’t magically figure it out for you.

I’ve tried a couple of examples above, and you’ve seen that we usually get responses that are not as good as we want.

Here’s one more example:

But when I added clear instructions, detailed context, and my writing style, the responses became significantly better.

So before asking AI, write down:

• What you want

• Why you want it

• What the final format should look like

• Any constraints

• The tone you prefer

And that’s how you a proper response, and what you want.

---

Let’s wrap up

If you’ve read this far, here’s the one thing I want you to take away:

Using AI isn’t the real skill anymore because almost everyone can do that now, but the creators who actually stand out are the ones who know when to step back, trust their own mind, and avoid using AI in places where it quietly washes away their voice and turns their work into yet another piece of forgettable, lookalike content.

And most creators don’t even realise this is happening.

No doubt, AI is great for the boring parts as it saves time, reduces friction, and helps you move faster. But it cannot understand your story, your taste, your intent, or the small moments where you choose one sentence over another because it feels right.

And that’s why you have to be very clear about what you want it to do.

In simple terms,

• If the task has no emotional weight, give it to AI.

• If the task needs your brain, your voice, or your judgement, do it yourself first and let AI support you after.

That’s the whole point of this post.

Once you understand this, everything else becomes easier.

• You know where to delegate.

• You know where to slow down.

• And you know where to let AI do the heavy lifting without touching the parts that define you.

---

Hope you like it.

That’s it—thanks.

Lastly, if you found this post valuable or learned something new, consider becoming a paid subscriber to this newsletter. It’s the best way to support my work and keep this kind of content coming.