Are your rubrics over-rewarding AI?
Designing criteria that better capture student learning
A well-designed rubric rewards meaningful evidence of knowledge, skills, judgment, and critical thinking. As AI tools become standard writing and thinking companions, traditional rubrics can over-reward work that appears polished or complete but lacks depth or independent thought.
Large language models (LLMs), the generative AI systems behind tools such as ChatGPT, can produce increasingly sophisticated work, especially when students refine it through repeated prompting to add depth, sources, or a personal voice. This makes it difficult, and often unreliable, to distinguish AI-supported work from solely human-authored work. This guide focuses on designing rubrics that reward evidence of student learning without relying on AI detection.
Even a strong rubric can help instructors evaluate the quality of submitted work, but it cannot establish how that work was produced or what a student can do independently. Strengthening a rubric may also require examining whether the assessment itself provides credible evidence of learning.
Strategies
Shift from detecting AI to protecting learning
One way to examine an assessment is through a Form-Signal-Strategy Framework:
| Guiding Question | |
|---|---|
| Form | How resilient is this assessment, or its individual components, to generative AI? |
| Signal | What features of student work are making you question whether the submission demonstrates the student’s own learning? |
| Strategy | What could you change in the design, delivery, or scoring of the assessment to better capture that learning? |
When reviewing student work, you may notice patterns such as unusually polished writing, similar phrasing across submissions, limited critical appraisal, or ideas that appear more sophisticated than expected. These characteristics can raise questions about what students have learned, but cannot reliably establish whether AI was used.
Consider what your rubric rewards, what matters most, and whether the assessment provides meaningful evidence of student learning. You may need to revise the criteria, performance descriptors, weighting, or the assessment itself. Rather than asking, “How can I tell if AI was used?” consider asking, “How can my rubric better reward authentic evidence of the learning I want students to demonstrate?”
Three ways to strengthen your rubric
1. Align your criteria and weighting with your learning outcomes
Start with the knowledge and skills students are expected to demonstrate. Avoid rewarding compliance with assignment instructions, such as page count, number of references, or formatting, rather than the quality of learning demonstrated. Where possible, focus criteria on what students do with their knowledge.
Pay attention to how marks are distributed across criteria. If presentation, organization, and completeness carry substantially more weight than analysis, interpretation, or disciplinary judgment, students may earn high marks for polished work that demonstrates limited understanding. While there may be a place for these elements, the weighting should reflect their importance in the assessment.
Ask yourself:
- What do I actually want students to show me they know or can do?
- Do my criteria and weighting provide meaningful evidence of those learning outcomes?
- Which criteria primarily measure compliance with instructions?
- Could a student satisfy this criterion without demonstrating substantial understanding?
2. Use only as many performance levels as you can meaningfully distinguish
Effective descriptors help distinguish between work that is merely complete, work that demonstrates developing understanding, and work that demonstrates sophisticated application, reasoning, or judgment. These distinctions should be grounded in the learning outcomes and supported by observable evidence in students' work.
Ask yourself:
- Can I clearly explain the difference between adjacent performance levels?
- Would another instructor be able to recognize those differences?
- Are some of my performance levels describing essentially the same quality of work?
- What level of performance should work that demonstrates surface-level competence—but limited understanding—actually receive?
3. Describe quality rather than counting features
Performance descriptors are most useful when they describe the quality of students’ work rather than simply counting how many required elements are present. Consider the difference between a descriptor such as “Includes three relevant examples” and one such as “Uses relevant and specific examples to develop and substantiate the analysis.” The first primarily rewards the presence of a feature. The second asks the grader to judge the quality and function of the evidence.
Ask yourself:
- Can I clearly explain the difference between adjacent performance levels?
- Would another instructor be able to recognize those differences?
- Are some of my performance levels describing essentially the same quality of work?
- What level of performance should work that demonstrates surface-level competence—but limited understanding—actually receive?
Selecting your rubric format
The format of your rubric can also shape what gets rewarded.
Analytic rubrics separate an assignment into several criteria and assign a performance level to each. They can provide students with considerable transparency about expectations and can be particularly useful when instructors want to give feedback on specific skills.
The effectiveness of this rubric type depends on the criteria and descriptors it uses. Highly detailed analytic rubrics can sometimes encourage a checklist approach to completing an assessment, or may provide a detailed blueprint for LLMs to generate a matching submission. When using analytic rubrics, consider using criteria that are more difficult for LLMs to fulfill, such as those that require students to:
- Use specific examples from in-class discussions or lab data and clearly link them to a core reading.
- Identify the project’s most challenging phase and assess how effective the chosen strategy was in overcoming it.
These criteria ask students to interpret evidence and justify their reasoning, rather than simply include required elements. Their effectiveness still depends on the assessment task and the evidence students provide.
| Criteria |
Exceeding 4 |
Proficient 3 |
Developing 2 |
Beginning 1 |
|---|---|---|---|---|
| Criterion 1 | Describe how a student exceeds target expectations. | Describe how a student meets expectations. | Describe how a student misses some target expectations. | Describe how a student fails to meet basic expectations. |
| Criterion 2 | Describe how a student exceeds target expectations. | Describe how a student meets expectations. | Describe how a student misses some target expectations. | Describe how a student fails to meet basic expectations. |
| Criterion 3 | Describe how a student exceeds target expectations. | Describe how a student meets expectations. | Describe how a student misses some target expectations. | Describe how a student fails to meet basic expectations. |
| Criterion 4 | Describe how a student exceeds target expectations. | Describe how a student meets expectations. | Describe how a student misses some target expectations. | Describe how a student fails to meet basic expectations. |
| Comments: Insert optional narrative comments here. | ||||
Holistic rubrics describe different levels of overall performance, considering how elements such as reasoning, evidence, and organization contribute to the quality of the work as a whole. They can help instructors make integrated judgments about student performance, although they provide less detailed feedback on individual skills.
| Level | Description of Characteristics |
|---|---|
|
Exceeding |
Write a descriptive paragraph that combines ALL characteristics of exceptional work. |
|
Proficient |
Write a descriptive paragraph that describes ALL target expectations. |
|
Developing |
Write a descriptive paragraph describing work that misses some target expectations. |
|
Beginning |
Write a descriptive paragraph that describes work that fails to meet the basic criteria. |
| Comments: Insert optional narrative feedback here. | |
A hybrid approach combines elements of analytic and holistic rubrics. For example, instructors might assess a small number of meaningful criteria separately while also considering the overall quality and coherence of the work. This allows for feedback on specific skills without losing sight of how they contribute to the submission as a whole.
| Needs improvement | Target Expectation | Exceeds expectations |
|---|---|---|
|
Write specific comments explaining what fell below the target criteria. |
Criterion 1 Name: |
Write specific comments on how the student went beyond target expectations. |
|
Write specific comments explaining what fell below the target criteria. |
Criterion 2 Name: |
Write specific comments on how the student went beyond target expectations. |
|
Write specific comments explaining what fell below the target criteria. |
Criterion 3 Name: |
Write specific comments on how the student went beyond target expectations. |
|
Write specific comments explaining what fell below the target criteria. |
Criterion 4 Name: Clear, concise description of what target expectations look like. |
Write specific comments on how the student went beyond target expectations. |
No rubric format is inherently more resistant to AI-generated work. Consider which structure best supports valid, consistent judgments about the learning your assessment is intended to measure.
Try it: Stress test your rubric
You do not need to wait until the next offering of your course to determine whether a revised rubric works as intended. Follow the steps below to stress test your rubric.
- Choose an assignment where you believe you have a high signal of AI use.
- Select several past submissions representing different levels of performance.
- Score the submissions using your existing rubric.
- Score the submissions using your revised rubric.
- Compare the results. Identify where scores changed and consider whether those changes are justified by meaningful differences in the evidence of learning.
- Consider these three questions:
- Alignment: Does the revised rubric better reflect the intended learning outcomes?
- Distinction: Does it better distinguish between superficial performance and more sophisticated understanding?
- Consistency: Could another instructor use the descriptors to make a similar judgment?
- Revise and iterate as needed, paying particular attention to the criteria, performance indicators, and weighting.
Putting it into practice: Three permissions
There is no single approach to assessment that will address every challenge posed by generative AI. As you review and revise your assessments and rubrics, keep these three permissions in mind:
- Permission to prioritize: Every assessment approach involves trade-offs. Consider which compromises you are comfortable with and which elements of your assessment matter most to preserve. If a revised rubric still cannot show what students can do independently, the task itself may need to change, for example, by asking students to explain key decisions or by adding a brief oral or practical component. Consider which compromises you are comfortable with, what matters most to the learning outcomes, and which elements of your assessment matter most to preserve.
- Permission to diverge: Solutions are not one-size-fits-all. You know your students, discipline, context, and learning outcomes best. Your approach may look different from that of your colleagues. Choose approaches that reflect your learning outcomes, disciplinary expectations, and students' needs.
- Permission to iterate: Teaching has always required revision, and in many ways, generative AI has increased the pace at which some of that revision may be necessary. Give yourself permission to iterate incrementally: Try an approach, examine what happens, and adjust in response to what you learn.
Learn more
Corbin, T., Bearman, M., Boud, D., & Dawson, P. (2026). The wicked problem of AI and assessment. Assessment & Evaluation in Higher Education, 51(4) 736-752.
DePaul University. (n.d.). Types of rubrics. Teaching Commons.
Estaphan, S., & Zulfiqar, T. (2026). Educators’ experiences transitioning from a traditional to an AI-aware assignment rubric: a qualitative study in a health and medicine course. Frontiers in Education, 11.
Hostetter, A. B., Call, N., Frazier, G., James, T., Linnertz, C., Nestle, E., & Tucci, M. (2025). Student and faculty perceptions of generative artificial intelligence in student writing. Teaching of Psychology, 53(3), 319-329.
Murray, N. & Tersigni, E. (2024). Can instructors detect AI-generated papers? Postsecondary writing instructor knowledge and perceptions of AI. Journal of Applied Learning & Teaching, 7(2), 155-167.
Licensed under a Creative Commons Attribution-Noncommercial-Sharealike 4.0 International License.