Have you ever felt like there are others that have just cracked the code on AI? There are those out there with the right kind of setup where they can put very little in and get their marketing done for free. And on top of that, their content is good, insightful, and doesn't seem like an AI wrote it?

The fact is, that doesn't really exist. Those who put forth the effort are, at best, delivering slightly better AI slop than putting forth little effort. And those who are using AI to help produce actually-good writing are putting in the effort.

This blog post is a testament to the point. In the original version, I was testing out a skill that I had written for blog writing. It followed a process that built it and checked it in a non-linear fashion with the hopes of eking out some text that didn't sound halfway like the next AI-written blog. I even ran it over at the end with a humanizer skill and Gemini 3.1 Pro (which I recently learned will write more like I want without much direction). To skip to the end, I mined some of the organized ideas back out of it and started over.

Basically, there's no free lunch to quality.

So, why then do people feel like there is? Why is it that when I spend time cycling through iterations that I wonder if I'm just not doing it right?

To unpack that, it's the quantity that we can get from AI that tricks us into sometimes thinking that it should be easy. But quantity is not just a strength of AI, but increasing the length (or shortening, for that matter) is literally how it works.

How the model technically works

In a language model, meaning lives in vector directions, and generation is a linear readout of a vector state, so when you ask for "more (or less) of the same", the model simply walks in the same semantic direction, and nothing new has to be decided. (https://arxiv.org/html/2311.03658v2)

But "better" isn't a transformation of the text. It's a comparison against something that was never in the document to begin with: the standard for what the writing should. That standard lives in your head, having been assembled from your taste, your history with the topic, and the decisions you've already made.

What's more, it's been shown that AI models cannot distinguish good quality from mediocre. Researchers from ByteDance asked language models to judge which was better between two pieces of writing. They used pairings where one was clearly the better writing (they had controlled for grammar, factual errors and length), and the AI judges essentially flipped a coin. They chose the better writing slightly more than half of the time. (https://arxiv.org/pdf/2510.14616)

Quality takes time and judgement

So where does quality actually come from? Every session I've had that produced something genuinely good follows the same pattern. I had already poured in my own ideas and principles, often stream-of-consciousness, and the AI fed that thinking back to me in different words. That can really solidify a thought or bring things full circle.

The point being, producing something worthwhile takes time and thought. AI generated writing may be fast and easy, but it's also obvious. That said, it is still more efficient than it used to be, and if you take that saved time and reinvest it back into the work, you get better quality.