Robot writing has no voice. That’s why you find it annoying. Like it or not, your brain is currently one of the best organic pattern matchers in the known solar system, and that means you can detect patterns. Like breathing, you just do this. All the time.
You are not being triggered by the endless em dashes; it’s not the trailing participial clauses, and the incessant adverb transitions. It’s that the robot does not care about what it’s writing; it has near-infinite data at its disposal, and it’s been instructed to keep you happy because that makes good business sense. Let’s discuss each in reverse order:
Brown-noser: Like any good business, the job of the robot is to make you a repeat customer, and that means it needs to efficiently respond to your prompt. Two parts to explain here:
- Efficiently. If you’re using ChatGPT or Claude on the web, those respective companies have provided the robot with meta guidance on how to answer. I suspect they provide a clause you never see that reads “Better is the enemy of done,” which means “Answer quickly,” because faster is cheaper. The reason I know this exists is that I mostly use Claude via Ghostty, and Claude Code knows to think (see: burning tokens) before answering my questions, which takes longer and costs more. I am shocked when I prompt via the web, and the answer is often both instant and frequently wrong.
- Your prompt. What’s your prompt? I don’t know. I do know that the robot’s primary job in responding is to make sure you get a satisfying answer. I didn’t say correct; I said satisfying. How do I know this exists? Because I have super dumb ideas all the time and never in the history of ever has a robot told me my idea was dumb and — wow — when I say dumb, I mean if you were sitting next to me in this coffee shop and I was telling you this dumb thing, you would raise your hand to stop me, put that hand on my shoulder, and tell me, “Wow. That… is dumb.” Furthermore, once you graciously shared this feedback, I would immediately see the dumbness.
Brown-nosing is when someone on your team incessantly and annoyingly agrees with you. All your ideas are good. Yes sir! I’ll get right on that. The robots have weaponized this sycophantic behavior at scale.
True story: I had a question about something dumb, and I was sparring with Claude Code on the inane topic when I realized all the robot was doing was agreeing with me. I typed, “I think you are telling me what I want to hear,” to which it responded, “Yeah. Probably.”
Infinite Know-It-All: Ok, not infinite, but a lot more than you have at your immediate disposal. The robot wants to answer your query efficiently, and it has all of that at hand, so it rapidly finds quite a bit of information and wants to share every bit of it with you. But which data matters more? Which matters less? What are the interesting bits versus the laborious, boring, endless facts?
Given your brief prompt, desire to help, and impressive repository of data, web-based robot responses are often walls of text. It’s like your good friend who is suddenly into the Jason Bourne franchise. Endless Matt Damon facts. Is it interesting? Maybe. Is it often more than you need? Frequently. These robot responses read like a know-it-all, which is interesting because they literally know nothing, which is the actual point…
Robots Don’t Care About Caring: That wall of text the robot just generated for you on the breakfast sandwiches in San Francisco is mostly true (it did search for the best breakfast sandwiches in San Francisco) and pretty complete, but as you read it… that emptiness you feel inside? It’s because the writing has no voice.
Good writing captures voice. Good writing documents how another human thinks. When you read text generated by a robot, it’s that thinking that is missing. Your world-class pattern matcher sees all the words, sentences, ideas, arguments, footnotes, em dashes, and everything else you’d expect out of writing, but the reason you squint your eyes and wonder, “Did a robot write this?” is because… there is no voice.
It does not care about its writing. It looks like it cares about you, but that’s because it’s echoing the last words you wrote — the prompt — and working incredibly hard to find the optimal set of words to respond to your specific set of words.
Do you want to know how someone thinks? Read their writing.1
Expect the Unexpected
Lyle and I were talking about robots on The Important Thing. I explained that the way you’d know a work was human was in the mistakes, “The robot would never do this, this is dumb; must be a human.”
Mistakes might not be the right word.
Non-linear. Out of place. Creative. Or perhaps just unexpected.
I was watching Sinners for the first time… on a plane. You know the scene: it’s when our main character, Sammie Moore, starts singing for the first time, summoning his musical ancestors, both past and future. Suddenly, there was a Jimi Hendrix-inspired character on the screen in the middle of what you thought was a Depression-era vampire movie
I clapped. Sitting on that plane.
Because I didn’t see that coming. My brain loved it. The lack of pattern.
So much voice.
- Some of the best long-form voices out there right now: Daring Fireball, Simon Willison, Charity Majors, Jason Kottke, and Stratechery. ↩︎
Leave a reply