I teach creative writing and I’ve been watching students try to use AI to produce work that sounds like them for about two years now. The results have taught me something I didn’t expect: voice isn’t just a stylistic pattern. It’s an accumulation of choices made under pressure, uncertainty, and genuine subjective investment in a particular way of seeing.
When you ask an AI to ‘write like me’ and give it examples of your work, what you get is a statistical approximation of your surface patterns. The sentence lengths, the vocabulary range, the punctuation habits. It can reproduce those with reasonable fidelity. What it cannot do is reproduce why you made those choices. The hesitation before an unusual word. The decision to end a section abruptly because the silence says more than another sentence would. Those aren’t patterns. They’re positions.
For students who are early in developing their voice, the AI approximation is actually a trap. It gives them back an image of their style that’s already flatter than the original, and then they start writing toward that image rather than toward the actual work. The feedback loop produces increasingly generic versions of something that was never that distinctive to begin with.
For more experienced writers, I think the picture is more complicated. Using an AI draft as a pressure surface to push against, to see what you disagree with or want to say differently, that seems like a legitimate creative use to me. Not ‘write for me’ but ‘give me something wrong that I can correct into what I actually mean.’
Has anyone found uses that feel genuinely authentic rather than like outsourcing something that was supposed to be yours?
For SEO content, individual voice isn’t really the goal. Brand voice is, and consistent brand voice is a pattern problem the tools handle reasonably well. But I do notice that AI-drafted content across my clients starts to converge over time toward a similar register even when I’m prompting for distinct brands. The tools have a center of gravity they return to and fighting it requires ongoing effort.
The ‘pressure surface to push against’ framing is the most useful way I’ve heard it described. I use it similarly when I’m coaching writers through a draft. The tool gives you something concrete to react to. Reacting is often easier than initiating, and the reaction is yours even if the starting point wasn’t.
here’s the thing: I use AI drafts regularly for client work and I’d say the voice-matching problem is real but overstated for professional writing contexts. when the goal is brand voice consistency rather than individual creative voice, the tool performs better because brand voice is more pattern-based and less position-based. what you’re describing matters more for literary work than for content work.
The ‘flatter than the original’ observation is the most precise description I’ve encountered of what happens when you ask these tools to match a voice. In my experience editing manuscripts, you can often tell when a writer has been using AI to assist because the passages feel technically competent in a way that’s missing something. They’re doing what good prose does without knowing why. That gap is exactly what you’re naming.
ngl as someone who uses these for school writing I recognize what you’re describing. the output sounds like me enough that I don’t think about it, but reading this I wonder if that’s actually the problem. it’s close enough that I stop trying to get closer.