Boston University Study Reveals How College Students Use Generative AI in Writing Courses

Boston University piloted ChatGPT Plus in introductory courses. Led by program instructor Sarah Madsen Hardy, instructors allowed up to 50 % machine-generated text if students highlighted AI passages in blue. After the semester, 50 volunteers submitted essays; 34 provided logs.

Psypost

Since ChatGPT’s 2022 release, educators have debated whether generative AI threatens academic integrity or can serve as a personalized tutor. The binary framing prompted Boston University to collect data on student behavior.

Psypost

The study, published in the *Journal of Writing Research*, involved introductory writing students, 25 % of whom were English-as-a-foreign-language learners. Researchers anonymized essays and logs, then used human raters and a large language model to categorize 290 prompts into functions such as brainstorming, revision, and direct writing.

Psypost

Only 18.6 % of prompts asked ChatGPT to generate original text; a quarter (?25 %) sought revision help, such as shortening sentences or adjusting tone. Most requests involved explaining concepts, defining terms, or locating sources, focusing on background work. Direct generation appeared mainly in the final quarter of a chat session. In the 50 final essays, only 8.2 % of words were flagged as AI-authored. When AI text was included, about 6 % appeared as whole paragraph blocks; short phrases woven into students’ own prose.

Psypost

Native English speakers requested clarification seven times more often than international peers, while foreign-language learners were about twice as likely to ask for revision feedback and rarely used the tool for summarizing sources. Consequently, international students incorporated fewer AI passages than native speakers.

Psypost

The findings suggest that, with structured guidance, students treat generative AI as a collaborative assistant rather than a substitute, challenging assumptions that AI will automatically erode original thinking and highlighting the potential for responsible instructional models.

Psypost

The research team emphasized that “students who are offered sustained instruction are capable of engaging GAI robustly and adopting its outputs selectively, in ways that suggest judgement about its effects and investment in their own learning.” Instructors noted that draft reviews reinforced self-reporting of AI use.

Psypost

Some educators remain concerned that AI tools could facilitate plagiarism or diminish critical thinking, especially if policies are lax or enforcement inconsistent.

Psypost

Participation was voluntary, so heavy AI users may have opted out. The self-highlighting requirement may also under-report usage. Researchers call for larger, more diverse samples and direct student interviews to confirm these patterns.

Psypost

• “Our findings show that students who are offered sustained instruction are capable of engaging GAI robustly and adopting its outputs selectively, in ways that suggest both judgement about its effects and investment in their own learning,” — Sarah Madsen Hardy

Psypost

Future work will expand the participant pool, add qualitative interviews, and test AI-augmented writing instruction across disciplines.

Psypost