How I Use AI to Write Without Losing My Voice

August 12, 2026

I love writing, especially for my blog. My recent drafts started manually. Yes, manually, with paper and pen. When I have a good version, I dictate the text to AI to get the digital version. After that, I start the review part.

At first I reviewed the text the next day, so I had a fresher mind to look at it from another angle. It's a good exercise to improve your writing skills. Later I started using AI, but only for grammar and punctuation fixes. No deep review yet.

The first (painful) reviewer version

Then I decided to test AI for a more detailed review. A few days later, I regretted it. The output looked like a bug audit report: a long list of points to improve, clarity suggestions, sentences that could be misunderstood, the low and high moments of the text, and so on.

Trying to obey the AI report was stressful. I was avoiding sitting at my desk to review the draft. An activity I loved, and used to be a relaxing moment, turned into something boring and exhausting.

Then I realized something. If I started fixing every nitpick the AI found, I'd end up like Theseus. If you don't know the Ship of Theseus paradox, let me tell you quickly:

If I fixed every AI suggestion and replaced word by word, at the end of the day, is it my text or the AI's text?

How I adjusted to make it better

So I stepped back and tried a new way to use AI, which I liked and I'm currently testing. No TODO items to fix, no long lists of improvements, no strong statuses like Should Fix, Recommended, or Optional. Just five cards with brief feedback on each part of the text:

• Title & hook
• Clarity & phrasing
• Logic & organization
• Voice & fairness
• Emotional impact

Each card is short: one or two paragraphs about the AI's thoughts on the text. Here's an example from the review of Your tools should match how you think:

Clarity and phrasing
Good enough: The core metaphor stays understandable across examples.

What's good

The draft returns to the same distinction in clear, concrete forms: what appears on a screen versus the categories and priorities already present in someone's mind. The money example is especially useful because the reader can see the internal correction happening in lines such as "this line should not be here."

What needs attention

The first description of "mental translations" carries several ideas at once: important information, accurate data, decisions, and action. Consider whether the first explanation could give the reader one simpler mental picture before it broadens into all of those benefits. That would make the rest of the post easier to follow, especially for readers who do not naturally think of reports as tools for decision-making.

Simple and concise comments, not a heavy and imposing review. They are not TODO items. They are signals about the direction I'm going. I can ignore the feedback if I want without feeling like I'm leaving a "bug" inside the text.

Conclusion

This way of reviewing is much lighter and more efficient than the previous version. But it's still under validation. I will use it on more future posts and see if it helps as I expect. I will probably post an update here on the devlog about it.

If you want to see the detailed evolution of my "Review Draft Report" skill, check the CHANGELOG.md file, which contains even my real feedback to AI:

"The bloat of status is a mess"
"The reports are heavy and overwhelmed me"
"Honestly, it was stressful."

What about you? Do you like writing too? Have you been using AI for it? Feel free to reach out to talk about your experiences. I will really appreciate it ([email protected]).

See you next time!