For faculty, one of the most immediate questions raised by generative AI is also one of the most familiar: How do I know what my students actually learned?
That question has always been central to assessment, but AI has changed some of the evidence faculty have traditionally relied on to answer it. A polished essay, completed problem set, working piece of code, or well-organized presentation may no longer reveal as much about the process that produced it.
Faculty are responding by taking a fresh look at how students demonstrate learning.
NCFDD’s Faculty Perspectives on AI survey report, based on responses from nearly 1,700 faculty members and academic administrators, found that 86.9% of respondents have already changed at least one aspect of their assessment practice. Nearly three-quarters say AI has forced them to fundamentally rethink what they assess, while 77.6% say it has become harder to determine what students actually know.
The changes underway suggest that assessment in the age of AI may increasingly depend on making more of the learning process visible.
How AI Is Changing Student Assessment
AI has made it easier to produce certain kinds of finished work. That does not necessarily make the assignment itself obsolete, but it can change the question faculty need the assignment to answer.
Consider a student who turns in a strong analytical paper. Depending on the course, the instructor may need evidence that the student can identify credible sources, interpret them, develop an argument, make connections between ideas, and explain their reasoning.
If AI can assist with several stages of that work, evaluating the final paper alone may give the instructor an incomplete picture.
The same issue can appear very differently across disciplines. A language instructor may need students to demonstrate grammatical or conversational proficiency they have developed themselves. A faculty member teaching programming may need to know whether a student understands why a piece of code works rather than whether they can produce functioning code. In a clinical program, students may eventually use AI-supported tools professionally while still needing to demonstrate the independent judgment required to use those tools responsibly.
The report shows substantial disciplinary differences in how faculty are experiencing AI, but concern about student learning crosses those boundaries. For example, 85.5% of humanities faculty, 78.9% of social sciences faculty, 78.4% of STEM faculty, and 67.1% of faculty in professional fields say AI makes it harder to assess what students actually know.
That makes the educational purpose of an assignment an increasingly important starting point.
Before changing an assessment, faculty might ask: What knowledge, reasoning, skill, or judgment is this assignment supposed to reveal?
The answer can help determine what students need to demonstrate independently, where AI use might be appropriate, and what additional evidence would make their learning easier to evaluate.
How Faculty Are Adapting Assessments for AI
Many of the assessment changes reported in the survey move beyond evaluating a single finished product.
Among respondents, 57.1% are redesigning assignments to require critical thinking, 55.7% have added more in-class or oral assessment, and 37.9% now ask students to document their use of AI. By comparison, only 15.9% report using AI detection tools, while 80% say they lack confidence in AI or plagiarism detection tools.
Those approaches can give faculty multiple points at which to see students engage with the material.
A paper might include an outline, source notes, drafts, or a short reflection on the choices a student made. A project might conclude with an oral explanation or follow-up questions. Students who use AI might submit their prompts and describe what they accepted, rejected, verified, or changed.
In some cases, AI itself can become part of the assessment. Students might evaluate an AI-generated response, identify weaknesses in its reasoning, verify its claims, or improve on its analysis.
Doug Peterson, PhD, Chair and Associate Professor of Psychology at the University of South Dakota, described approaches in which students use AI for one stage of an assignment and then have to show what they can do with the result.
“It’s your knowledge, it’s your understanding, it’s your job to prove you have acquired it,” Peterson said.
That responsibility remains central regardless of whether students use AI in completing an assignment. Faculty still need to see evidence that students understand the material and can apply what they have learned.
Setting Clear Expectations for AI Use in Assignments
Changing an assignment often raises another question: What kind of AI use is acceptable here?
The answer may reasonably differ from one course to another. Students practicing foundational writing skills may have different boundaries than graduate students critiquing AI-generated research summaries. A student learning to code may encounter different expectations at the beginning of a sequence than in an advanced course where AI-supported development tools are part of professional practice.
That makes some variation useful. But faculty are currently making many of these decisions independently. In the survey, 71.2% of respondents say they set their own AI policy for their students, while only 8.5% follow a university-wide policy.
Clear course-level expectations can help students understand those distinctions.
Rather than relying only on a broad statement that AI is allowed or prohibited, faculty can connect expectations directly to the learning goals of an assignment: which parts students should complete independently, where AI may assist them, what they need to disclose, and what they should be prepared to explain or defend.
Lisa Hanasono, PhD, NCFDD’s Academic Director of Training and Content and a professor at Bowling Green State University, emphasized the importance of allowing for different approaches to AI use depending on the learning goals.
“When we move to this kind of all-or-nothing approach, then we lose the nuance,” she said.
That nuance matters when faculty are establishing expectations for individual assignments. Students need to understand when AI use is prohibited, when limited assistance is appropriate, and when they may be expected to use and evaluate the technology as part of their learning.
Assessment redesign takes time
The breadth of these changes matters.
Redesigning an assessment may mean revisiting prompts, rubrics, class activities, grading practices, student instructions, or the sequence of an entire assignment. Moving more work into class can also affect course schedules. Adding presentations or oral questions may require substantially more faculty time, particularly in large courses.
One humanities respondent in the survey described the challenge directly:
“I’ve had to redesign almost every assignment and assessment so it can be completed and evaluated in the context of an AI environment. So. Time. Consuming.”
That work is part of a much broader transition happening across higher education. Faculty are making decisions about assessment while AI tools and student practices continue to evolve, which means a redesign that works well today may still need adjustment in a future semester.
For departments and institutions, supporting this work means recognizing assessment redesign as substantive faculty work. Time to experiment, opportunities to compare approaches with colleagues, discipline-specific conversations, and clearer shared principles can all help faculty make thoughtful decisions without requiring every instructor to solve the same questions alone.
Keeping Student Learning at the Center
There is unlikely to be one assessment model that works across every discipline, course, and stage of learning.
The survey points toward a broader reconsideration of what faculty need students to demonstrate. What should students know? What should they be able to do? Which skills require independent practice? Where can technology support learning? And what evidence will help faculty determine whether students have developed the knowledge and judgment their courses are intended to build?
These questions are fundamental to teaching, and faculty are already responding to them through changes in their courses and assessments. Their experiences also show why institutions need to provide the time, resources, and guidance that make this work sustainable.
Explore more faculty perspectives on AI in higher education.
NCFDD’s new report, Faculty Perspectives on AI, examines what 1,659 faculty members and academic administrators shared about student assessment, workload, disciplinary differences, institutional guidance, and the changing role of AI in higher education.
