When Can GAI be Used Anyway? An Analysis of Engineering Faculty's Generative AI Policies
Plain Language Summary
This paper analyzes how engineering professors and instructors decide when students in their classes are allowed to use generative AI tools like ChatGPT. Because there is currently no single university-wide rule, students often face very different expectations from one class to the next. Drawing on interviews with 169 engineering faculty and instructors across 18 universities, we sorted their policies into three groups: no restriction, some restriction, and full restriction. We found that most instructors land in the middle, allowing generative AI for some tasks (like debugging code or brainstorming) while prohibiting it for others, and that they mostly build these policies on their own rather than following a shared institutional standard.
Contribution
We extend prior institutional-level generative AI policy research by empirically mapping instructor-level classroom policies in engineering specifically, applying a deductive content-analysis codebook grounded in Policy Diffusion Theory to characterize both the restriction level and the underlying motivators/objectives behind 169 faculty members' and instructors' generative AI policies.
Research Questions
- What GAI policies do engineering instructors adopt in their classrooms?
- How do GAI policies for engineering instructors' classrooms vary?
Methods
We recruited 169 engineering faculty members and instructors from 18 universities, selected by undergraduate degrees conferred per ASEE's By the Numbers report, for 45-60 minute semi-structured interviews conducted via Zoom between January and December 2025, asking about their current AI policy, how it was developed, and its goals. We analyzed responses to each interview question independently using content analysis: policies were coded into restriction-level themes (unrestricted, semi-restricted, fully restricted), semi-restricted policies were further coded by permitted use case (e.g., debugging, brainstorming, feedback), and development/motivation responses were mapped onto a deductive codebook derived from Policy Diffusion Theory (e.g., emulation, taken-for-grantedness, policy transfer).
Key Findings
Of the 169 participants, 155 had a GAI policy for their class(es) while 17 did not, and among those with a policy, 120 (77%) fell into the semi-restricted category, compared to 30 fully restricted and only 5 unrestricted. Within semi-restricted policies, instructors most commonly agreed on permitting GAI for syntax/debugging (82% of relevant cases), understanding/clarity (63%), and idea generation/brainstorming (63%), while use cases like draft writing (18%), editing (9%), and sketching/visualization (9%) were far less consistently allowed. Policies were more often self-generated or adapted through emulation and policy transfer than derived from university-level guidance, and instructors most frequently framed their policy's purpose as enforcement (65%) and specifying acceptable use (52%) rather than pure restriction.
Implications
Since semi-restricted policies dominate, but instructors disagree considerably on which specific use cases are acceptable, we see an opening for department- or college-level coordination that could standardize the boundaries of permitted GAI use within a discipline while still allowing course-level tailoring. Because policy development is currently siloed and largely self-driven despite calls for more faculty training and support, we argue that closing this gap, and mapping these classroom-level findings against evolving federal and institutional policy, is a necessary next step for giving students consistent, discipline-specific expectations for GAI use.