AI changed how I write, but probably not in the way people assume.

I don't sit down, give ChatGPT a topic, and ask it to write an article.

Most of my writing starts long before I open ChatGPT.

1. Capture signals

It starts with everyday information consumption.

Books. LinkedIn posts. Instagram. Conversations. Something someone says over coffee. Occasionally, just an idea that appears while I'm doing something completely unrelated.

I think of these as signals.

When something makes me stop and think, I capture it immediately. Usually in Notion. Sometimes I type it. Increasingly, I record a voice note. If Notion isn't available, Apple Notes works.

The important part isn't where it goes. It's reducing the distance between noticing something and saving it.

This is actually one part of the workflow I still want to improve, and potentially build something around. Discovery should probably be more ambient and automatic than it is today.

2. Decide what is worth thinking about

Later, usually on a weekday evening or weekend morning, I go through what I've collected.

I'm not really asking:

What should I write about?

I'm asking:

What do I actually feel like thinking about?

There's a difference.

I also don't spend much time doing traditional topic research unless I need to verify something specific. Factual information is abundant now. If someone wants a summary of a paper, the history of an idea, or an explanation of a concept, they can ask AI.

I don't think my value comes from repackaging information that is already available.

The reason I write is more selfish than that.

I want to organize what I think so that six months or five years from now, I can return to it, understand what I was thinking, and hopefully become a slightly better version of myself.

Writing is partly publishing.

But mostly, it is archiving thought.

3. Talk before I write

Once I choose a signal, I open ChatGPT or sometimes Claude.

Then I talk.

I use speech-to-text inside ChatGPT or Typeless and do a massive thought dump. Sometimes it's structured. Usually it isn't.

I might talk for ten minutes about one idea, jump between examples, contradict myself, discover another idea halfway through, and eventually circle back to the original point.

This is probably the most important part of my writing process.

I'm not asking AI to generate an opinion.

I'm trying to get my opinion out of my head.

Voice works particularly well because typing introduces friction. When I type, I naturally start editing myself. I worry about sentences, structure, grammar, and whether something sounds intelligent.

When I talk, I just think.

4. Let AI organize the mess

Only after the thought dump do I ask AI to organize it.

This is where AI is extremely useful.

It can identify themes, remove repetition, create structure, clarify an argument, and point out places where my reasoning doesn't quite connect.

But the first organized version is almost never the final version.

Now I read.

And I ask myself:

  • Does this actually make sense?
  • Is this worth archiving?
  • Is there enough of an opinion here?
  • Does this actually sound like me?
  • Does it reflect my taste?

That last question matters more to me than grammar.

Because as AI makes competent writing abundant, competent writing becomes less interesting.

Taste becomes more important.

5. Talk again

Then the process becomes conversational.

I read something and say, "That's not quite what I mean."

I explain why.

AI rewrites it.

I read it again.

Sometimes I push an argument further. Sometimes I delete an entire section. Sometimes I realize the interesting idea isn't the original topic at all, but something that emerged while talking about it.

Most of this still happens through voice.

So the process looks less like traditional writing and more like an extended conversation with an editor who can continuously reorganize what I'm saying.

The loop is simple:

Signal Capture Think Organize Read React Refine

And I keep going until the piece feels like something I would want to encounter again in the future.

6. Publish the archive

At the end, AI helps with the boring parts.

Grammar. Repetition. Basic fact checking. Sentence cleanup. Formatting.

Then I publish it on my personal website.

That's basically how I write now.

AI certainly makes me faster. But speed isn't the part I find particularly interesting.

The bigger change is that the cost of turning an unfinished thought into something structured has collapsed.

Which means the bottleneck has moved.

The hard part is no longer producing words.

It's noticing things worth thinking about.

Having an opinion.

Knowing when the AI-organized version isn't quite right.

Developing enough taste to recognize yourself in the final result.

AI handles more and more of the writing mechanics.

I still have to decide what is worth saying.