Is AI Ruining Your Writing Skills? What New Research Means for Authors

A growing body of research is asking a question that matters to every author who drafts with AI: can the tools that make you faster also make you worse? Two new studies, one by Anthropic and one published in Nature and Scientific American, suggest the answer is not a simple no.

What the Anthropic study found

Researchers at Anthropic ran a randomized controlled trial with 52 software engineers. All participants were given a basic coding task. They could search the web and consult instructions. Half were also encouraged to use an AI assistant.

After the task, every participant took a quiz on what they had learned. The engineers who used AI scored an average of 50 percent. The group that worked without AI scored 67 percent. The AI-assisted group did particularly badly on questions that asked them to diagnose errors in the code they had just produced, a sign they had not absorbed the concepts behind the code the AI wrote for them. The study was posted on arXiv ahead of peer review.

Kevin Crowston, an information scientist at Syracuse University who researches how generative AI changes the way developers learn, described the finding in blunt terms. "Now you have this very odd disconnect between performance and learning," he told Nature. "People can perform at a pretty high level, because they are basically borrowing skills from the AI, but they are not developing those skills themselves."

The pattern holds across professions

The coding study is not the only signal. A study of Polish endoscopy specialists, published last October in The Lancet Gastroenterology and Hepatology, tracked physicians who had performed at least 2,000 colonoscopies each. When they were given access to an AI tool that flagged precancerous lesions in real time, their unassisted detection rate dropped from 28.4 percent to 22.4 percent over three months. When the tool was unavailable, they found fewer lesions than they had before the tool existed.

A survey of US healthcare workers, cited in the Scientific American piece from June 18, found that 70 percent of nurses and 77 percent of physicians worry about losing their skills because of over-reliance on AI.

Tapani Rinta-Kahila, an information-systems researcher at the Hanken School of Economics in Helsinki, points out that earlier technologies eroded skills too: GPS weakened navigation abilities. But generative AI, he told Nature, is "the first technology that automates various cognitive faculties around thinking and interpretation, which were long considered unique human skills." His own 2018 study found that accountants who had used an automated system for more than a decade forgot how to do several routine tasks when the system was taken away.

Yuichi Mori, a physician-researcher at the University of Oslo and co-author of the endoscopy study, said plainly that "there is no established solution against deskilling right now." He expects it to become "a very hot research topic in the next decade."

What this means if you write with AI

Nobody has yet run a randomized trial with novelists under the same conditions. But the mechanism described in the Anthropic and endoscopy studies transfers cleanly to writing. When a tool does the sentence-level choosing for you, you spend fewer hours making those choices yourself. Over months and books, the risk is not that AI will steal your voice overnight. It is that you gradually exercise the muscles that produced it less often.

That does not mean you should stop using AI. The same studies show AI raises immediate output quality. The engineers produced working code faster. The endoscopists spotted more adenomas when the AI was on. The problem is not that the tools fail during use. It is that the user's own capability dips when the tool is not there, and the dip grows over time.

For authors, the practical answer is not abstention. It is routine that keeps your own editorial judgment in the loop. Three habits emerge from the research:

• Revise every AI-generated passage yourself, word by word when the voice matters. The act of rewriting is where you make craft decisions you otherwise outsource.
• Before accepting an AI suggestion, ask why it made that choice. Diagnosing the output, the way the Anthropic study quiz asked engineers to diagnose code, is the learning part.
• Write some sections from scratch, without AI, even when a model would produce them faster. The endoscopy study suggests that continuous exposure to AI assistance can make clinicians "less motivated, less focused, and less responsible when making cognitive decisions without AI assistance." Writing without AI for part of every session keeps that muscle warm.

The authors who stay sharp will not be the ones who never touch AI. They will be the ones who treat it as a drafting partner that speeds up the typing, and keep the thinking for themselves.

Our guide on how to edit and humanize AI-generated book content covers the revision process that turns a generated draft into your own prose. The current disclosure rules and quality expectations for platforms like Amazon KDP and Kobo are covered in how to publish an AI book on Amazon KDP.

PageWriter Studio is built around that loop. The AI generates outlines, drafts, and full chapters. You steer every revision with a style profile tuned to your own writing. The research says the tool makes the first draft faster. The question that matters is whether you still own the sentences after revision. If you want to try that workflow on your next title, you can start a free trial and stay in control from the first page.