Students using artificial intelligence for writing may need to think more carefully about their work, not less, challenging the assumption that AI simply makes writing easier, an Iowa State University study found.
The peer-reviewed study examined an experimental “AI and Writing” course that followed 38 undergraduate students across 22 majors in fall 2023 and fall 2024 as they learned to collaborate with generative AI tools. The course included students from 22 majors, including engineering, humanities, business, and science.
Researchers Abram Anders of Iowa State University and Emily Dux Speltz of Embry-Riddle Aeronautical University analyzed students’ reflections after they completed AI writing exercises and self-directed projects.
The researchers identified three broad concepts they said were essential for productive AI use in writing: that writing with AI is an experimental process, that it requires human expertise and dialogue, and that it should strengthen, rather than replace, the writer’s own agency.
This complicates the common AI in education debate. It is not necessarily a threat to academic integrity, but only when students are trained to prompt it and use it effectively.
AI-generated writing can often appear reliable when it is not
One of the central findings of the course was that AI-generated writing can often appear reliable when it is not. This is described by the researchers as a problem of linguistic fluency, that AI can produce polished and confident prose, while still missing factual accuracy.
To confront that problem, students completed an exercise called “Create a Fluent Hallucination,” in which they were asked to generate deliberately false but plausible AI outputs, including fabricated events and invented sources. Other assignments included a prompt competition, an “AI Ethics Tutor” exercise, and a task in which students designed their own AI assistant by deciding which parts of a process should be handled by a human and which could be supported by AI.
The study brought examples from specific student experiences. A journalism student found that ChatGPT could help with writing leads, but did not follow the expected structure of a journalistic lead or journalism rules. The student later received more useful feedback after giving the AI context on the rules for writing a lead and then asking it to respond to a lead the student had already written.
AI still requires students to retain control over significant parts of writing
That example, the researchers suggest, points to a broader finding. AI output improved when students brought their own expertise to the exchange. Rather than treating AI as a search engine, successful students learned to use it as a tool that required context and correction to achieve desired results.
This suggested to the researchers that AI still requires students to retain control over the most important parts of writing. While it can assist with brainstorming, structure, feedback, and even the wording itself, students still need to decide what they are saying, and whether the output meets the standards of their discipline or audience.
In practice, the study suggests that AI may shift where student effort is directed. Rather than writing, students may need to spend more time and effort planning, evaluating, revising, and crucially, deciding how much of the writing process they can delegate to AI.
The study does not claim that AI made students better writers; instead, it examines how students described changes in their own thinking. The researchers acknowledge that further studies are needed to test whether these changes will last over time, and whether they translate into stronger writing or better academic performance.
