AI Writing Tools Create Fluency Trap Where Polished Output Masks Shallow Thinking, Study Finds

Are you using ChatGPT or Claude for writing work? A study says you may be landing in a fluency trap

Researchers followed students using AI writing tools for two semesters and found that polished output is the problem, not the solution. If you've been relying on ChatGPT or Claude to help you with your writing, a new study suggests the polished output you're getting may be giving you false confidence. Research published in the Computers and Composition journal found that AI writing tools create a "fluency trap," where refined, confident-sounding output masks shallow thinking and gives writers a false sense that the work is done. Fluent doesn't mean finished Abram Anders, associate professor of English at Iowa State University, and co-author Emily Dux Speltz, assistant professor at the Department of Humanities and Communication at the Embry-Riddle Aeronautical University, followed 38 undergraduate students across two semesters in an experimental "AI and Writing" course. Students came in expecting AI to cut their workload, but it didn't. The research explains that the fluency trap sets in because AI produces text that reads as confident and clean, leading writers to trust it even when the content is wrong, shallow, or off-point. Many students initially treated AI like a search engine, entering a vague prompt and accepting whatever came back. Over time, they learned that effective prompting required planning, clarity, and rhetorical awareness, the same skills strong writers already use without AI. Recommended Videos "AI writes in confident sentences, uses the right tone and sounds smart," Anders said. "But that polish can trick students into trusting it, even when it's wrong, shallow, or missing the point entirely." What good AI-assisted writing actually looks like The researchers identified three things writers need to understand before they can use AI effectively. First, working with AI requires genuine trial and error, not a single prompt and accept. Second, AI output still needs human judgment to check claims, refine logic, and match the expectations of a given context. Third, AI can generate text, but it cannot generate purpose. Only the writer can decide what the writing is arguing and why it exists. Students who worked through those three thresholds stopped treating AI as a shortcut and started using it to test ideas, evaluate options, and sharpen their arguments. Anders and Dux Speltz describe this shift as moving from outsourcing your writing to orchestrating it. "AI changes the workflow, but it doesn't change the fact that writing is thinking," Anders said. That distinction matters more as AI-generated text becomes harder to tell apart from the real thing.

Writing with AI demands more thought from students, not less | Newswise

Abram Anders (above, standing), associate professor of English and the Jonathan Wickert Professor of Innovation at Iowa State University, talks with undergraduate students during an experimental "AI and Writing" course at Iowa State. AMES, Iowa -- Writing with AI can look deceptively simple. Effortless, even. Type in a prompt and a polished paragraph appears in seconds. Tidy, confident, clean. But that apparent ease is also deceiving, says Abram Anders, associate professor of English and the Jonathan Wickert Professor of Innovation at Iowa State University. "Writing with AI doesn't take the work out of writing," he said. "It changes it." In a new study published in Computers and Composition, Anders and co-author Emily Dux Speltz, an Iowa State alum and assistant professor in the Department of Humanities & Communication at Embry-Riddle Aeronautical University, suggest the biggest hurdle in teaching students to write with AI isn't the technology -- it's the students' assumptions about what writing is. "Students often expect AI to function as a shortcut, but the truth is, AI-assisted writing demands more thought from students, not less," said Anders, who also serves as associate director of the Student Innovation Center at Iowa State. "As a tool, 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." Crossing threshold concepts To conduct the study, Anders and Dux Speltz designed an experimental "AI and Writing" course that followed 38 undergraduate students from 22 majors as they learned to collaborate with generative AI tools over the course of two semesters. The students completed structured assignments, reflected on their process and documented how their thinking changed as they experimented with AI tools. At the start of the course, Anders said students carried a variety of assumptions, including "better tools should require less effort" and "AI will do the work for me." But reality quickly challenged those beliefs, he added, with one student reflecting, "I had to learn how to think about my thinking." What also emerged, the researchers found, were three "threshold concepts" -- or big ideas -- that students need to understand before they can write effectively with AI. The first? Writing with AI is experimental, and students must learn to try, test and tinker. "AI isn't going to provide a 'perfect' answer or automatically spit out what you need," Anders said. "It requires trial and error -- trying, testing, revising and trying again." The researchers said some students reported they initially treated AI like a search engine: enter a vague prompt, accept whatever comes back. But as the course progressed, they learned that effective prompting required planning, clarity and rhetorical awareness -- the same skills strong writers use without AI. Which brings us to the second threshold concept: writing with AI still requires human expertise. "AI writes in confident sentences, uses the right tone and sounds smart," Anders said. "But that polish can trick students into trusting it, even when it's wrong, shallow or missing the point entirely." This potential pitfall is sometimes to referred as the "fluency trap," Anders said. However, once students learn to read AI content critically and question it, they begin to see that fluency is not the same as understanding. "It's crucial that students learn to interrogate what AI produces and not just edit it," Anders said. "This means checking claims, refining logic and ensuring the writing aligns with different expectations related to different disciplines -- all work that requires human judgment." This also leads into the idea of ownership, which Anders and Dux Speltz address with a third threshold concept: writing with AI should ultimately augment human agency, not replace it. "Students must recognize that while AI can generate text, it can't generate purpose -- only the writer can do that," Anders said. "Generative AI can't decide what it's arguing, what matters or why the writing exists. It's a tool that requires human direction, judgement and boundaries." The researchers describe this as a shift from "outsourcing work to orchestrating it." "After crossing the third threshold concept, students are using AI to explore possibilities, test ideas and refine thinking rather than to avoid the cognitive load of writing," Anders said. Why this research matters As AI tools become more common in academic, professional and everyday writing, Anders and Dux Speltz say students will not only need technical proficiency, but also a deeper understanding of how writing works. "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." Students who moved through the thresholds as part of the "AI and Writing" course reported becoming more reflective, more critical and more intentional about their choices, the researchers said, and instead of treating AI as a shortcut, they began using it to evaluate ideas, explore alternatives and strengthen their arguments -- a shift that mirrors the demands of real-world writing. "When students learn to direct AI rather than depend on it, they become stronger writers, and that's the skill that will matter long after the tools change," Anders said.