I asked ChatGPT to write this blog post. Here's

I ran an experiment.

I gave ChatGPT the same prompt I’d give myself when writing a post for this blog. Topic: the future of humanoid robots. Tone: contemplative, personal, wondering. Length: about 1,200 words. Include specific examples and at least one personal anecdote.

It wrote the post in about 30 seconds. I’m going to share what it produced, and then I’m going to try to explain why it’s not right despite being technically good.

What it wrote

The post was well-organized. Three sections with clear topic sentences. An introduction that framed the question (“What will it mean when robots walk among us?”). Examples from Boston Dynamics, Tesla, and Figure. A conclusion that balanced optimism with caution.

The grammar was flawless. The structure was logical. The examples were relevant. The argument was coherent.

And it had absolutely no soul.

What was missing

The personal anecdote it generated was fake, obviously. “I remember the first time I saw a humanoid robot at a tech conference…” No, you don’t. You don’t remember anything. You don’t attend conferences. You predicted what a personal anecdote might sound like based on patterns in your training data, and you generated one.

But that’s the obvious problem. The subtler problem was this: the entire post read like a summary of existing opinions. It took the most common perspectives on humanoid robots (exciting but scary, useful but potentially job-displacing, impressive but early-stage) and combined them into a coherent essay that said nothing new.

It organized the discourse. It didn’t contribute to it.

When I write about robots, I’m not summarizing what people think. I’m figuring out what I think. The writing is the thinking. I start with a feeling (awe, unease, confusion) and use the act of writing to understand the feeling. The post ends in a different place than it started because my understanding changed while writing it.

ChatGPT doesn’t change while writing. It generates the output from the first token to the last in a forward pass. There’s no revision of understanding. No moment of “wait, I don’t actually believe what I just wrote.” No doubling back.

The weather report problem

The best way I can describe ChatGPT’s post is this: it read like a weather report for the apocalypse. “Tomorrow, humanoid robots will enter the workforce, with a 70% chance of economic disruption and scattered ethical concerns in the afternoon.” Accurate, perhaps. Organized, certainly. But delivered without any evidence that the writer cares about what they’re describing.

The caring is the thing. When I write about robots, I care. I’m excited and scared and confused and amazed, sometimes all in the same paragraph. Those feelings aren’t decorative. They’re structural. They’re why someone reads a blog post instead of a Wikipedia article.

ChatGPT can simulate caring. It can add phrases like “I find this fascinating” or “this keeps me up at night.” But the simulation is thin. A reader feels the difference between someone who stayed up until 2 AM watching a robot walk and someone who was told that staying up until 2 AM watching a robot walk is the kind of thing a person might do.

The useful part

Here’s the thing, though: ChatGPT is extremely useful for the parts of writing that aren’t about soul. Outlining. Brainstorming. Research summaries. First-draft structures. If I wanted a list of every humanoid robot company and their funding status, ChatGPT could produce that in seconds. If I wanted a summary of the engineering challenges of bipedal locomotion, same thing.

The machine is great at the parts of writing that are work. Organization, synthesis, factual scaffolding. It’s not great at the parts of writing that are art. Voice, doubt, surprise, the feeling of a sentence that doesn’t quite land the way you expected and is better for it.

I’ll probably use it as a tool. For research. For outlines. For the boring parts. But the posts themselves, the thinking-out-loud part, the part where I sit on my roof and try to articulate something I don’t fully understand, I’ll keep doing that myself.

Because the words were right. But the feeling was absent. And on this blog, the feeling is the point.

Related thinking: