Students' statistics writing has shifted noticeably since large language models like ChatGPT became widely available, with their prose now resembling AI-generated text more closely than before. A new analysis of over 1,600 undergraduate data analysis reports written between 2021 and 2025 shows that students are adopting writing patterns characteristic of large language models (LLMs), particularly in introductions and conclusions. At the same time, educators worry that offloading writing tasks to AI may prevent students from developing essential analytical and communication skills.

What's Actually Changing in Student Writing?

Researchers compared student writing to text generated by four major LLMs: GPT-4o, GPT-5 Mini, Gemini Flash, and Claude Haiku. The findings reveal a measurable shift in how students structure their sentences and choose their words. One consistent pattern emerged: students are using more nominalizations, which are nouns formed from verbs or adjectives. For example, instead of writing "we operated the equipment," students increasingly write "the operation of the equipment." This grammatical choice is a hallmark of LLM-generated text.

The shift is most pronounced in the first and fifth quintiles of student reports, which roughly correspond to introduction and conclusion sections. These are precisely the sections where students might be most tempted to ask an AI tool for help organizing their thoughts or polishing their language. Interestingly, researchers also found that students' writing has become more similar to expert statistics writing in recent years, suggesting that LLMs may be exposing students to more professional communication patterns.

Why Should Educators Be Concerned About This Trend?

The concern isn't simply that students are using AI; it's that relying on LLMs for writing tasks may prevent students from developing deeper understanding of their own work. Writing assignments in statistics and data science courses serve multiple purposes beyond just producing a final document. When students write about their analysis, they organize their thoughts, refine their understanding, and learn to communicate complex ideas clearly. If an AI tool handles the writing, students may miss these cognitive benefits entirely.

Writing skill development traditionally follows three stages: knowledge-telling (stating what you already know), knowledge-transforming (using writing to think through a topic and refine ideas), and knowledge-crafting (actively considering your audience and adapting your message accordingly). Heavy reliance on generative AI may trap students in the knowledge-telling stage, where they simply reiterate information without engaging in deeper processing. Research shows that participants who did not use ChatGPT in writing assignments reported engaging in deeper processing, exerting more mental effort, and maintaining stronger sustained attention during the task.

How Can Educators Adapt Their Assessment Strategies?

• Targeted Writing Assignments: Focus on specific sections like report introductions where students must structure their own thinking rather than relying on AI to organize their ideas for them.
• In-Class Verbal Assessments: Require students to explain their statistical findings and limitations verbally, where they cannot offload the task to an AI tool and must demonstrate genuine understanding.
• Process-Focused Evaluation: Assess students on their ability to interpret statistical results in context, consider ethical implications, and think critically about data limitations, rather than solely evaluating the final written product.
• Collaborative Problem-Solving: Design assignments where students must work together to analyze data and communicate findings, making it harder for individuals to substitute AI work for their own thinking.

The research underscores a fundamental tension in modern education: AI tools are becoming more capable and accessible, yet the skills that writing assignments were designed to develop remain essential. Students who repeatedly offload writing tasks to LLMs may struggle to communicate their findings spontaneously in future careers, produce statistical analyses in novel settings, or critically evaluate data and statistical claims presented to them.

The study's authors propose that educators should not abandon writing assignments but rather redesign them to emphasize the cognitive processes that make writing valuable in the first place. By focusing on how students think through problems and structure their reasoning, rather than just the final prose, instructors can preserve the learning benefits of writing while acknowledging the reality of AI in students' lives.