Evaluating AI-Generated Student Work
Picture this: You're teaching a 10-week undergraduate class. You assigned a weekly one-paragraph response to the week's readings, and your students are now submitting content that feels oddly correct, grammatically perfect, yet strangely impersonal. You immediately suspect they wrote it with AI.
Help! What do you do?
First of all, realize you're not alone, and it's not the way your class is set up. The Digital Education Council's 2024 Global AI Student Survey, which gathered 3,839 responses from students across 16 countries, found that 86% use AI in their studies. UCLA’s 2025 Senior Survey found that 73% of students reported using it. This scenario is here to stay. Given that, what can you do to keep guiding students toward critical thinking—and what can you do when you suspect the work they submitted is (excessively) AI-generated?
Focus on your class goals
Think about the assignments and tests you designed. What are you trying to test for, and what are you assessing when you grade? This helps you shift your focus from the format of an assignment (which you might have to change) to the aim of it (which you want to preserve).
To help you brainstorm, here are a few common things instructors test for, depending on course content:
Memorization: facts, formulas, events, grammar structures
Comprehension: contextualization and interpretation of written passages or historical situations
Writing ability: fluency with different formats and styles, clarity of expression, precision in formulating arguments
Critical thinking: commenting, critiquing, making connections, noticing gaps
Ask your students openly
To work around heavy AI use, it helps to first understand why students reach for it. Two of the most common reasons—and two different ways of framing the problem—are cheating and creativity. Depending on how you see it, you might choose different paths: find ways to curb the behavior, or find ways to accompany it.
Start by asking students directly. That AI use is widespread is an open secret at this point, so don't be afraid to raise it first. Is it because they're pressed for time? Have too much homework? Don't feel confident in their writing? Find the writing itself too challenging? Don't know where to start?
You could share an anonymous survey in Week 1 asking what leads them to use AI (an anonymous Google Form or a Bruin Learn survey works well). If you do, remind students you're not trying to police their answers—you're trying to understand how they work in your class so you can help them get something valuable out of it.
Rework your assignments
Try reshaping some assignments to focus on feedback and improvement while making AI feel less necessary.
Consider multimodal formats. Ask students to submit audio or video of themselves speaking, presenting, or commenting on a paper they're responding to.
Break writing into smaller iterations (if class size and topic permit). This helps you get acquainted with each student's voice and style, and lets them work on their writing under less time and grading pressure per assignment.
Assign written reflections on the research process. Along with scaffolding research assignments, have students document their process: keywords and search queries, databases searched, permalinks to results, how they decided which sources to use or ignore, and an annotated bibliography of what they chose.
Focus on improvement over one-off performance. For a longer paper or complex final project, scaffold assignments across the quarter so individual tasks build toward—and earn feedback for—the final product.
Include AI in the assignment. In an analysis- or critical-thinking-focused class, have students include an AI response among their sources and explicitly evaluate it: the kind of information it provides, its tone, positionality, and possible biases, and how it compares to other sources.
Encourage students to cite AI and treat it as a source. In a writing- or research-focused class, help students see the difference between piecing knowledge together from varied sources (each with different validity and credibility) and pulling answers from one opaque source that presents a single possible answer. (See our resource on citing generative AI.)
Offer flexibility on deadlines and topics. Empowering students to choose what they work on can disincentivize AI use for assignments they have less time for or interest in. For example, if your class calls for 5 response papers, write 8 prompts for the quarter and let students skip the 3 weeks or topics they least want to cover.
Loop in your Subject Librarian. For research assignments with outside sources, a consultation can surface what students struggle with, where they're tempted to plagiarize, and which library or free resources fit best. You can also request a library instruction session on resources, research methods, and the benefits and pitfalls of AI.
Add conferencing or in-class verbal feedback. Students who didn't spend much time writing will struggle to answer questions about an AI-generated essay.
Assign peer reviews. As Peerceptiv notes, assessing peers' work—AI-generated or not—helps students identify strengths, weaknesses, and nuance, honing analytical skills. Peerceptiv is one good tool for this.
Reward verifiable sourcing. Rather than only penalizing false, fabricated, or "hallucinated" information, give explicit credit for accurate, well-attributed sources—so students see careful verification as part of the work.
Grading work you suspect is AI-generated
Whatever new formats you try, you'll likely still face the question of how to grade work that was meant to be student-generated but feels like it wasn't.
There is no reliable way to tell AI-generated writing from human writing. (See our resource, "The Imperfection of AI Detection Tools.") Turnitin, Grammarly, and others offer AI-identification features, but they are not dependable, because AI-generated text has no intrinsic linguistic markers that distinguish it from human language. A few things every instructor should know before acting on a detector score:
The tools are unreliable in exactly the conditions you're facing. Turnitin won't even flag a document unless a substantial portion (around 20%) looks AI-written, and it needs a minimum length (roughly 300 words) to attempt detection at all. A one-paragraph weekly response is well below that threshold, so a score on it means very little.
They carry a documented equity problem. Detectors disproportionately flag non-native English writers and first-generation students. One widely cited Stanford study found that several detectors flagged roughly 61% of non-native English essays as AI, versus near-zero for native writers—because the statistical patterns these tools key on penalize the simpler vocabulary and structures common among English-language learners. Acting on detector output can therefore harm exactly the students your course should support.
Institutions and vendors are backing away from treating scores as proof. Several universities—including Vanderbilt, Michigan State, and Northwestern—have paused or disabled Turnitin's AI detection over false-positive concerns, and Turnitin's own guidance states its scores should not be used as sole evidence of misconduct.
False accusations carry real consequences. Students have successfully challenged universities over wrongful AI accusations, so a flag should be treated as the start of a conversation, never a verdict.
Be wary, too, of the implicit bias that a particular student "couldn't possibly" have phrased or framed something a certain way on their own. Rather than spending your or your TA's time investigating writing style, focus on whether students seem to have gained a nuanced, improved understanding of the course content—and lean on process evidence over detection. Drafts, outlines, notes, and Google Docs version history show the work actually happening and are far more defensible than any detector score.
❗ If you suspect an assignment is AI-generated, you have options:
Be clear from the start. Address AI use in your syllabus, emphasize the educational (not punitive) role of grades, and say in advance what you'll treat as cheating and what would be grounds for asking a student to redo or resubmit. Setting expectations before a dispute is the single most protective step you can take.
Ask the student to redo the work during office hours, under your or your TA's supervision.
Grade around the doubtful parts. If you have concerns about specific portions, grade as if those parts were missing and ask the student to rework them before you grade the whole.
Ask the student to talk you through it. Have them explain what they wrote, their thinking, and why they chose certain arguments—and to retrace and critique the sources they used.
Follow institutional process. In some cases AI use is a form of cheating or plagiarism. There's no foolproof way to identify it, so refer to UCLA's academic-integrity policies and your conduct office's procedures rather than acting on suspicion alone.
The bottom line: AI is here to stay, and students are using it. The challenge is to find new ways to guide students through difficult or time-consuming writing and critical thinking—without demonizing AI, and without being blind to its growing presence in their everyday lives.
Special thanks to the UCLA Library for their review and contributions.