Finished copy. Unfinished thinking
Several of the responses to my last post made versions of the same argument: train the AI model on your voice. Give it clear instructions.
That’s good advice, but it only solves the surface problem. It makes the writing sound more like the writer or the organisation behind it. And that’s where the trouble starts because it can also make the draft sound finished before the thinking earns it.
Before AI, a rough draft usually looked rough. You could see the gaps – the weak claim, the missing example, the paragraph that hadn’t decided what it was trying to say. All the indications of thinking in progress.
AI changes that.
It gives unfinished thinking the appearance of finished copy.
The headings are in place. The tone is reasonable. Sentences move into each other without any indication of friction. The draft looks as if it has already passed through some kind of editorial process.
That changes how organisations handle it.
Instead of being treated as raw material, the draft now enters the workflow as near-final copy. People tweak it. Shorten it. Make it “warmer”. Make it “sharper”. Move it into slides. Send it for approval. Paste it into LinkedIn.
But some of the central questions may never have been answered.
What are we actually claiming?
What evidence supports it?
Who is this for?
What could we defend if challenged?
What has the draft made sound settled when it is not?
The issue is not only whether the copy is good – it is whether the draft has been given the right status. Some AI drafts are useful raw material. Some are useful structure. Some are not useful at all. None of it should be mistaken for judgement.
As my colleague (@)Sangeeta put it, the surface tells are the easy part. The harder question is what the fluent prose has been allowed to get away with.
Because when unfinished thinking looks finished, the organisation starts editing surface instead of substance.
The editorial work remains undone – because it was quietly handed to the tool least able to do it.