George David Gopen

Building Writing Confidence to Reduce Reliance on AI

By George David Gopen, JD, PhD

Why do so many capable researchers hand their writing to a machine? The usual answer is speed. The deeper answer is writing confidence. Deadlines are real, grant cycles are relentless, and a draft that appears in thirty seconds is hard to resist. But speed does not explain why a scientist who can design an elegant experiment would distrust her own account of it. Something else is at work.

After four decades of teaching writing to scientists, lawyers, and physicians, I believe that something is writing confidence. Most of the researchers I have taught do not lack ideas. They lack confidence that a reader will receive those ideas the way they intend. Generative AI offers to close that gap. It cannot.

What the research suggests

Recent studies point in the same direction. In 2025, researchers at the University of Southern California’s Center for Generative AI and Society surveyed 1,000 American college students. Students who felt less confident in a course, or who saw themselves as less competent, were more likely to turn to generative AI. Students who considered themselves competent were less likely to use it.

A second study, presented at the 2025 ACM Conference on Human Factors in Computing Systems by researchers from Microsoft Research and Carnegie Mellon University, surveyed 319 knowledge workers. Those with greater confidence in the AI applied less critical thinking to their work. Those with greater confidence in themselves applied more.

These studies show associations, not causes. Time pressure, workload, and trust in the tool also drive AI use. But the pattern is consistent: the less writers trust their own judgment, the more of that judgment they hand to the machine.

Fluency is not clarity

AI produces fluent sentences. A sentence can be grammatical, polished, and confident in tone while still leaving its reader unsure what matters most.

Readers of English do not read neutrally. They hold firm, largely unconscious expectations about where in a piece of writing they will find particular kinds of information. When writing honors those expectations, readers understand it the first time. When it violates them, readers misread, reread, or give up.

A language model has no idea which piece of your research your reader most needs to grasp, or where that reader will look for it. Only the writer knows what the science means. The writer’s task is to put that meaning where the reader expects to find it. That is a skill, and like any skill, it can be learned.

The stakes have changed

For researchers who depend on federal funding, the shortcut now carries real risk. Beginning with the September 25, 2025 receipt date, the National Institutes of Health no longer considers applications “substantially developed by AI,” to be the original ideas of the applicant. AI use detected after an award may be referred for misconduct review.

NIH also now accepts no more than six applications per principal investigator per calendar year. Each proposal therefore carries more weight than before. Volume can no longer compensate for clarity. Each application must persuade reviewers on its own, in the investigator’s own words.

Writing confidence can be taught

The link between confidence and writing is not new. More than a decade before ChatGPT, a team at the Indiana University School of Medicine identified “underconfidence” among the obstacles that keep faculty from writing, alongside time constraints, and the anxiety and procrastination that can accompany a fear of failure.

That team then measured what happened when faculty learned to write from the reader’s perspective. Between 2006 and 2009, 230 faculty members attended a two-day Academic Writing Program built on the Reader Expectation Approach. I designed and taught the program, and I was one of nine authors of the published evaluation. One hundred-and-fifteen faculty members responded to a follow-up survey.

Survey itemShare of respondents
Agreed their confidence in communicating concepts in writing increased87%
Agreed their manuscripts and grant proposals would be more competitive72%
Reported an increase in grant proposal submissions33%


Increased confidence in communicating concepts in writing scored highest of every item the survey measured. In their written comments, participants described clearer writing, new ways of seeing their own prose, and greater confidence. One reported a paper “accepted with extremely minimal text changes on the first shot.” The study’s authors concluded that the program reduced the anxiety participants associated with academic writing, and they inferred that “greater confidence may lead to increased submissions of scholarly works.”

These are the faculty members’ own assessments of their work. They show what changes when writers understand how readers read: they stop guessing, and they become willing to put their work forward.

The tool is not the problem

AI is not going away, and I do not suggest that researchers pretend otherwise. But a tool used to cover a lack of confidence leaves the lack in place. The writer remains unsure, the reader remains unpersuaded, and the work no longer sounds like the scientist who researched it.

The better investment is in the writer. Writers who understand where readers look for meaning no longer need a machine to guess for them. They can say what they mean, and readers will understand it the first time. This skill and confidence ripple outward to improve all areas of the individual’s professional and personal communications.


Learn about Reader Expectation Approach training for research institutions, or discover my latest book The New Science of Scientific Writing: Using Reader Expectations to Transform Scientific Communication.

References

Posted in: