The pitch for generative tools is that making things is now free. Type a sentence, get a result, move on. And the generation genuinely is nearly free. But shipping that result raw — unedited, unexamined, straight from the model to the world — is not free at all. It just moves the cost somewhere you don’t see until later.
This post is about that hidden bill, because most people paying it don’t know they are.
Free to make, expensive to ship
The trap is that the cheapness of generation feels like the cheapness of the whole thing. It isn’t. Generation is one line item; everything downstream of it is where the real cost sits, and raw output front-loads all of it onto other people and your future self.
| What raw output seems to cost | What it actually costs |
|---|---|
| A few cents and thirty seconds | Hours of someone else noticing it’s generic |
| Nothing — it looks fine | Your credibility, the moment a reader clocks it as unedited |
| One quick result | The subtle factual errors you didn’t check because it read confidently |
| Time saved now | Your own judgment, which only sharpens when you actually edit |
That bottom row is the one that compounds. Every time you ship something without deciding what’s wrong with it, you skip a rep of the only skill that was ever yours to build. Raw output doesn’t just cost credibility once — it quietly stops you from getting better.
The “confident and wrong” tax
Generated output has a specific, dangerous quality: it’s fluent. It reads as though someone knowledgeable wrote it, whether or not anything in it is true. A hesitant human draft signals its own uncertainty; a model’s draft signals confidence at all times, including when it’s wrong. That mismatch is a tax you pay in credibility every time an error ships wearing a confident voice — and it’s why “it sounded right” is never the same as “it was right.”
Editing is the part you’re actually paid for
None of this is an argument against using the tools. It’s an argument for treating generation as the first step, not the last. The value you add — the thing a client or a reader is actually paying for — lives entirely in the gap between the raw result and the shipped one. Skip the gap and you’ve automated away your own contribution and kept only the commodity part. (I’ve written about the editing skill itself in Twenty Images, One Keeper, and the bigger shift in AI Made Creation Cheap. Taste Got Expensive.)
A pre-ship pass for anything AI-assisted
- Before publishing, find one thing that's generic about it and make that part specific. Raw output is average by construction; your job is to make it particular.
- Fact-check anything stated confidently. Fluency is not accuracy, and the model's tone gives you no signal about which claims are true.
- Ask what a knowledgeable reader would clock as "unedited," and fix exactly that. If you can't spot it, you haven't looked hard enough yet.
- Cut what the model over-produced. Generated work tends to be too long and too even; the edit is usually subtraction, not addition.
- If you changed nothing before shipping, stop and ask what you actually added. If the answer is "nothing," you shipped the commodity and kept none of the value.