The EFL Language Learning AI That Is Also Accidentally Teaching Creative Writing Better Than Any Human Tutor

A group of intermediate English students at a language school in Tehran were given a seemingly simple assignment: write a brief essay in response to a picture prompt. An AI writing assistant was available to them. The task was presented as a language exercise. Grammar. vocabulary. The standard. Slowly and then with some surprise, the researchers observing them noticed that the students were doing more than simply fixing their sentences. Their voices were growing. stories. a level of control over the page that was not demonstrated in the same manner or at the same rate by the control group, which was receiving instruction in a traditional classroom.

Researchers Xinqiao Cen and Goodarz Shakibaei’s study, which was published in Scientific Reports in January 2026, is based on that observation. They separated 92 male intermediate EFL students into three groups: a control group that received traditional instruction, a group that received extensive AI-generated feedback on their creative writing, and a group that received minimal AI support. The outcomes were clear-cut. The two AI-assisted groups performed better than the control group. The biggest gains were seen in the high-AI support group. Crucially, the improvements weren’t merely technical; they also manifested in the students’ confidence, engagement, and willingness to take innovative chances that cautious, grammar-first instruction tends to discourage.

When you consider it, the explanation is almost elegant in its simplicity. One of the biggest obstacles to creative writing is grammar anxiety. The issue is well known to any language instructor who has witnessed a student freeze over a blank page, paralyzed not by a lack of ideas but by a fear of making a mistake. In reality, ChatGPT and QuillBot are examples of AI tools that absorb that anxiety. They swiftly and impartially handle the mechanical layer, which includes conjugations, prepositions, and sentence structure, at any time of day and as often as necessary. And something else becomes available once that layer is taken care of. All of a sudden, the student is free to consider what they truly want to say.

Under the direction of researcher Aliakbar Tajik, 65 intermediate students participated in a four-month writing program at Islamic Azad University’s Varamin-Pishva Branch. Half of the students used QuillBot, and the other half used traditional classroom techniques. In addition to reporting higher motivation and more autonomy, the QuillBot group demonstrated statistically significant improvements in all of the writing dimensions the researchers assessed, including coherence, vocabulary range, grammatical accuracy, and task completion. In the follow-up interviews, students talked about something that is rarely adequately captured in research papers: the sense of being accepted where they are, free from shame and the social pressure of a teacher observing them struggle.

Reading this research gives me the impression that the discovery of creative writing is truly coincidental; no one who created QuillBot or ChatGPT for language learning took the time to plan a course on narrative development. The tools were designed to identify mistakes, recommend synonyms, and reword awkward phrases. However, by carrying out those tasks consistently and smoothly, they made it possible for something more challenging and fascinating. According to the Cen and Shakibaei study, students in the high-AI support group demonstrated greater creativity and engagement because the AI managed cognitive load, or the mental burden of worrying about technical accuracy, while leaving the expressive layer up to the student.

It should be noted that this is not a clear-cut case for substituting human writing instruction. Regarding this, the researchers are cautious. Human judgment is still needed for thematic organization, cultural nuance, and the kind of deep structural feedback that helps a writer understand why their story isn’t working. The particular resonances and references that are important in an Iranian classroom, a Turkish university, or a Japanese language school may be missed by AI systems that use training data primarily from Western, English-language sources. What the student is reaching for is unknown to the tool. Only a teacher who is familiar with the pupil does.

However, the AI is doing something that the classroom frequently cannot for the students in these studies, who are intermediate learners sitting in resource-constrained settings, frequently without access to native speakers, and frequently managing significant anxiety about their own linguistic inadequacy. It is showing up at eleven o’clock at night, patient and impartial, and saying things like “try again,” “here’s another way to phrase that,” and “this is how a more confident sentence sounds.” Unexpectedly, some of them are learning how to tell stories in that area.