I Quit AI for 30 Days. Here's What It Did to My Brain

On day 3 I caught myself typing a question into a note-taking app, out of pure muscle memory, waiting for it to answer me. It didn't. That's when I realized how deep this went.

For 30 days I used zero AI. No ChatGPT, no Claude, no autocomplete, no AI summaries, no "fix this email." I worked like it was 2021 again. Here's what got worse, what got better, and what I permanently changed when it was over.

One month, zero AI. The withdrawal was real.

Why I did this to myself

It started with a moment I couldn't shake. A colleague asked me a question in a meeting — something squarely in my area — and my honest first instinct was "I'll ask Claude and get back to you." Not "let me think about it." The chatbot had become the first stop, and my own head had become the backup.

I'd also read the research that's been piling up. An MIT Media Lab study hooked essay writers up to EEGs and found that people writing with ChatGPT showed weaker neural engagement and could barely quote their own essays afterward. Researchers started calling it cognitive offloading: every time you hand a mental task to a tool, the muscle that did that task gets a little less practice. None of this proves AI rots your brain. But I'd been using it for roughly three years, daily, for exactly the kind of work I most wanted to stay good at. So I ran the experiment on myself.

The rules were strict: no chatbots, no AI coding tools, no AI features hiding inside other apps — no summarize buttons, no smart replies, no generated headers. Search engines were allowed (with AI answers scrolled past), spellcheck was allowed. Thirty days.

Week 1: withdrawal is a real thing

I'm not going to dress this up. The first week was bad.

Writing was the worst of it. My process for two years had been: dump messy thoughts into a chatbot, get a structured draft back, rewrite it in my voice. Without that middle step, I sat in front of blank pages for embarrassing stretches of time. A newsletter that normally took 90 minutes took three and a half hours. Code was nearly as rough — I'd gotten so used to autocomplete finishing my thought that plain typing felt like walking through sand. I reached for documentation I hadn't opened in a year because I'd been letting AI paraphrase it for me.

And the reflex. I counted, out of morbid curiosity: on day 2 I started to open a chatbot eleven times. Eleven. For things like "what's a synonym for rigid" and "how many ounces in 750ml." Questions I either knew or could answer with ten seconds of actual thought.

Weeks 2–4: the part nobody talks about

Somewhere around day 10, the discomfort turned into something else.

My first drafts got better. Not faster — better. When you know no tool is coming to save you, you think the idea through before you start typing, and the sentences come out closer to done. By week three my editing passes were noticeably shorter than they'd been in the AI era, because I wasn't spending them sanding chatbot cadence out of my own writing. I wrote about that voice problem in my piece on the words that make text sound AI-generated — turns out the easiest way to avoid them is to not start from AI text at all.

Reading changed too. I'd developed a habit of summarizing long articles and papers instead of reading them. Reading the full thing again was slower and, annoyingly, much stickier — two weeks later I could still recall arguments from papers I'd actually read, while the summarized ones had always evaporated by the next morning. A summary gives you the conclusion without the reasoning, and it turns out the reasoning was the part my brain kept.

The strangest change: boredom came back. Waiting in line, walking, showering — with no chatbot conversations to continue in my head, my brain went back to chewing on problems in the background. Two of my better ideas that month showed up uninvited on walks. I don't think that's mystical; it's just what brains do when you leave them unassigned.

What I kept, what I cut

Day 31, I didn't rush back to everything. I re-added tools one at a time, only when I missed them for a concrete reason. The final scorecard:

That last row saved me real money, which is a whole topic on its own — I broke down the math in which AI subscriptions are actually worth paying for.

The rule I landed on, and the one thing I'd actually recommend copying: I do the first pass myself on anything I want to stay good at. First draft, first design sketch, first attempt at the bug. AI gets invited in afterward, as a critic and an accelerator, never as the author. It's the difference between a tool that extends your thinking and one that replaces it — same apps, completely different long-term effect.

Should you try it?

A full month is a lot, and honestly more than most people need. The two-week version gets you most of the insight: pick the one skill you'd be saddest to lose, and do it unassisted for 14 days. You'll find out within a week whether AI has been amplifying that skill or quietly replacing it. For me the answer was different for every tool — which is exactly why the blanket takes on both sides ("AI makes you dumber" / "refusing AI makes you obsolete") are both useless.

And if a full detox sounds miserable but the dependence feels familiar, start smaller: I once ran the opposite experiment and let AI plan my entire week, which taught me a lot about where the tools genuinely earn their keep.

FAQ

Thirty days without AI didn't turn me into a Luddite. It turned me into someone with opinions about which tools deserve access to my thinking — and that, it turns out, was the thing three years of daily use had quietly taken away.