At Goethe University in Frankfurt, 4,048 students recently sat down to describe how they actually use artificial intelligence—and their answers reveal something far more complex than the simple embrace-or-reject binary that often dominates the debate. While 89% of them have incorporated AI tools like ChatGPT and DeepL into their academic writing, most are grappling with a profound tension: they see real benefits, but they also sense something precious slipping away.
The nuance begins with what students actually do. They turn to AI primarily for the scaffolding work of academic life—beginning an assignment, brainstorming ideas, polishing language and style. But when asked about deeper work, the pattern shifts. Only about one-third use AI for analyzing research texts or drafting substantive content. This distinction matters because it suggests students retain an instinctive sense of which tasks demand their own intellectual muscle.
Yet beneath those habits lies genuine unease. Study author Nora Hoffmann observed that "the majority of students try to limit their AI use to tasks they consider justifiable," driven by awareness that these systems make errors and by worry about their own dependency. Just over half said they accomplish less independently when using AI, while 45.5% fear losing their ability to think critically. These aren't marginal concerns—they're the majority view. At the same time, 92.2% of students said they felt personally responsible for the texts they submitted, and 78% described writing itself as an essential way to develop and test ideas.
The reality, however, often diverges from intention. Only 10% of students reject AI entirely. Another 18% manage what Hoffmann calls responsible use—treating AI as a "sparring partner" for testing ideas or a "tutor" for methodological guidance. But 72% admit they're tempted, particularly while reading, to let AI take over entire tasks, delegating authorship itself to what they call the "ghost" role. The culprit isn't moral weakness. Time pressure, grade anxiety, struggling motivation, writing difficulties—these are real conditions of student life, and they collide directly with AI chatbots designed to make themselves indispensable. One student captured the interior conflict perfectly: "Sometimes I have to actively motivate myself to think independently instead of handing the task over to AI."
The urgent question now is how universities respond. Hoffmann notes that the moment for intervention hasn't yet closed—many students learned to write before AI became widespread and still remember what they gain from wrestling with ideas on their own. But that cohort is aging out. The recommendations from Goethe's Schreibzentrum are specific: universities should define, discipline by discipline, which skills remain irreplaceable, which can reasonably be delegated, and which AI-related literacies students must develop. Critically, they should weave emotional and motivational dimensions into AI literacy programs, not just technical ones. And they should establish subject-specific guidelines while rebuilding support for reading and writing as learning processes, not just endpoints for assessment.
What's striking about this research is that it refuses both technophobia and technophilia. "The goal is not to ban AI," Hoffmann concludes, "but to enable students to use it critically and thoughtfully." The survey shows students are already trying—what they need now is institutional help to succeed.