When people edit AI drafts, are they getting better at writing, or is the AI quietly doing the practice for them?

This explores whether editing an AI draft works as writing practice, or whether it quietly replaces the practice that builds skill. The corpus has no study that tracks a person's writing ability over time while they edit AI drafts, so it can't settle the question directly. What it does show is the step the question assumes, people actually editing, often doesn't happen. In one study, writers edited AI-generated paragraphs only 23% of the time, and the edited versions stayed about 96% similar to the original Do writers actually edit AI-generated text before publishing?. If editing were going to build skill, most of that editing isn't taking place.

When people do put real effort into editing, it seems to pay off, at least in results. On Freelancer.com, the same worker got hired more often when they spent more time editing an AI-drafted cover letter, even though most workers sent drafts with barely any changes Does editing time on AI drafts predict hiring success?. That's a link between effort and outcome, not proof that anyone's skill improved. Still, it suggests the value lies in how actively you engage with the draft, not in whether AI was involved. Admissions offers a sharper warning. At one master's program, a majority of 2025 applicants submitted essays that were likely mostly AI-written. Those essays were better written, yet those applicants were admitted at lower rates than similar applicants who didn't use AI Does AI essay use hurt admissions chances despite quality gains?. Polish that comes from the draft rather than the writer doesn't get the same credit.

A less obvious risk is that light editing doesn't just skip practice. It can also train your sense of what good writing sounds like toward the model's default voice. In a controlled experiment, GPT-4o's autocomplete pulled Indian writers' essays toward Western phrasing and cultural references, and American participants got larger productivity gains Do AI writing assistants push non-Western writers toward Western styles?. So someone who mostly accepts suggestions may be learning to write like the model rather than developing their own voice.

The studies that look more hopeful change the writer's role. Writers who set up a proactive AI partner in advance, deciding its role and how often it should step in, used it to generate ideas and to monitor their own writing. The AI didn't produce the text Can writers benefit from configuring AI writing partners in advance?. In shared editors, writers wanted to see when and how their collaborators had prompted AI, partly so they could check the AI-generated text Do writers want to see each other's AI prompts in shared editors?. Both setups keep the human doing the judgment work, which is the part that builds skill.

An unexpected parallel comes from research on AI agents that rewrite their own instruction documents, called "skills". Letting an agent freely rewrite its instructions turns out to be unstable. Improvement is steadier when each edit has to pass a held-out test and the rejected edits are kept as negative examples Does constraining edits make skill learning more stable? Can skill documents be optimized like neural network weights?. That's machines, not people, but the lesson carries over: editing improves skill when each change is checked against a standard and the failed changes count as feedback. For a human writer, editing an AI draft becomes practice only when you can say why one version is better than another. Accepting the draft as it is doesn't give you that.