A new peer-reviewed study of 38 college students found that writing with artificial intelligence takes more mental effort than writing without it, not less — a conclusion that challenges a common assumption as businesses invest billions of dollars in AI tools and employee training. The research, led by Abram Anders, associate professor of English and the Jonathan Wickert Professor of Innovation at Iowa State University, was published in the journal Computers and Composition and detailed by Iowa State on Monday, June 15.

Anders and co-author Emily Dux Speltz, an assistant professor in the Department of Humanities and Communication at Embry-Riddle Aeronautical University, tracked 38 undergraduates from 22 different majors across two semesters in an experimental course called “AI and Writing.” Students completed structured assignments, then wrote reflections documenting how their thinking changed while working with tools such as OpenAI’s ChatGPT and Anthropic’s Claude.

Most students entered the course expecting AI to do much of the work for them. Instead, they discovered something different. “Writing with AI doesn’t take the work out of writing,” Anders said. “It changes it.”

That finding carries implications well beyond the classroom. As Microsoft, Google, OpenAI, and Anthropic compete to bring AI writing tools into offices around the world, and employers devote significant resources to training their workforces, the study suggests the technology shifts work rather than eliminating it. AI can generate polished text quickly, but the responsibility for judgment, accuracy, and decision-making remains with the user.

Anders put it directly. “AI only handles the surface-level writing, and the real heavy lifting — idea formation, judgment, revision strategy, and quality control — remains with the student writer,” he said. Replace “student” with “employee,” and the finding applies just as easily to today’s workplace.

The researchers identified three ideas students had to understand before AI became a productivity tool rather than a shortcut. The first is that writing with AI is an experiment, not a vending machine. A single vague prompt rarely produces useful work. The second is that strong results depend on the user’s own expertise. Writers must understand a subject well enough to recognize when AI gets facts wrong or produces weak analysis. The third is that the human writer—not the software—must remain responsible for the meaning, direction, and purpose of the final product.

One of the study’s most striking findings involves what the researchers call the “fluency trap.” AI often produces writing that sounds confident, polished, and authoritative even when it is shallow, misleading, or entirely false. Because the writing appears professional, many users instinctively trust it without carefully verifying the information.

Anders and Dux Speltz found that many students initially approached AI much like a search engine, entering a prompt and accepting whatever answer appeared. To challenge that mindset, the course included an exercise called “Create a Fluent Hallucination,” in which students deliberately generated believable but completely false AI content, including fabricated events and invented sources. The exercise was designed to demonstrate firsthand how convincing incorrect information can appear when produced by generative AI.

The lesson extends well beyond education. Businesses increasingly rely on AI to draft emails, marketing materials, reports, proposals, contracts, customer communications, and internal documents. If employees fail to verify AI-generated information, polished errors can quickly become expensive mistakes.

The workforce implications run even deeper. Rather than eliminating effort, the study concludes that AI shifts effort toward the aspects of work that are most difficult to automate: defining problems, exercising judgment, evaluating evidence, making decisions, and revising toward a clear objective. For employers calculating the return on AI investments, that complicates the simple assumption that AI automatically reduces labor. While software may produce a first draft in seconds, organizations still need skilled employees capable of directing, evaluating, and improving that output.

The research also reshapes how writing ability should be viewed in hiring and workforce development. Anders and Dux Speltz argue that as AI becomes embedded in academic, professional, and everyday communication, success will require more than knowing how to operate the software. Workers will need a stronger understanding of how writing and thinking work together.

“AI changes the workflow, but it doesn’t change the fact that writing is thinking,” Anders said. “Students still have to make decisions, set direction and shape meaning.”

The authors are careful not to overstate their conclusions. The study does not claim AI made participants better writers. Instead, it examined how students described changes in their thinking throughout the course. The researchers acknowledge that additional studies involving larger groups are needed to determine whether those changes produce lasting improvements in writing quality. The findings also reflect the experiences of a relatively small group of 38 students.

Even so, the practical message is difficult for employers to ignore. Students who embraced the three core concepts became more deliberate, more skeptical, and more thoughtful in how they used AI. Those who viewed the technology as a shortcut generally produced shortcut-quality work.

As companies continue investing billions in AI software and employee training, the study suggests the biggest competitive advantage will not come from having access to AI—it will come from having employees who know how to question it, guide it, and improve what it produces. AI may generate the first draft in seconds, but the research indicates that critical thinking, sound judgment, and subject expertise remain the qualities that ultimately determine the quality of the final work.

JBizNews Desk | Ames, Iowa

© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.