I started using ChatGPT the way a family takes in a new member. I asked it questions. I let it draft paragraphs I would once have written myself. Knowledge that used to cost an afternoon of searching now cost a prompt and a few seconds. I was enthusiastic, and for a while that enthusiasm felt earned.
The hangover came a few months later.
What caught up with me was the language itself. Every answer arrived fully formed: no hesitation, no rough edge, every sentence already polished. That fluency was seductive, and I let it do work I used to do myself. I think of the capacity I lost as an inner knife, the tool that cuts through a thicket of information until something solid is left. Mine went dull. I no longer had to push through a problem to get to the other side of it. The machine delivered the finished thought, and I nodded along.
After a while I noticed that everything the tool produced sounded the same. The same phrases, the same pseudo-clever turn of phrase, a faint residue that seemed to say look how smart I sound. Then I noticed the same residue in my own writing. My sentences had started to sound like its sentences.
At that point I stopped handing my language over to the machine.
I still see the residue everywhere now, mostly in places that make claims about themselves: about pages, company positioning statements, personal introductions. Sentences polished to a shine no person behind them would produce on their own. Only a small part of it came straight out of a machine. The language itself has spread, and people repeat it without a script in front of them, the way a phrase from an advertisement outlives the campaign. Post-buzzword speech, I have come to think of it, is AI babble that no longer needs an AI to produce it.
Plenty of it reads well. What troubles me is that I can no longer tell whether a person thought something through before writing it down, or typed a prompt and copied out whatever came back. Those are different acts, and a smooth sentence erases the difference between them.
That difference is what trust runs on. I trust a sentence when I can feel a person behind it, wrestling with an idea, even if the result is rough round the edges. Take away the visible sign of that effort and the basis for trust goes with it, however well the sentence still reads.
I am not accusing anyone of this. I noticed it first in my own writing, and I understand the pull towards it, because I felt it myself for months. What I have chosen instead is slower: write it myself, think it through myself and accept that the result will sometimes be rougher than what a model would hand me in ten seconds.
I still use the tools. I write in German and run the text through a translation tool that renders it into English. The tool does not work word for word. It reads what I meant and finds the right register in another language, useful work worth having a machine do. The thought it translates still has to be mine.