The Need for Rethinking Assessment

Jason Blomquist; Liza Long; Joel Gladd

9 The Need for Rethinking Assessment

How This Chapter Fits in Part 2

Part 2 works as a redesign sequence. It starts with the assessment problem and then moves through the decisions faculty need to make before they revise an assignment.

The Need for Rethinking Assessment identifies the assessment problem. Student AI use, course-design vulnerabilities, and the gap between a finished submission and student learning.
Rethinking Bloom’s Taxonomy for the AI Era names the learning evidence. Bloom’s taxonomy, UnBlooms, and outcome revision help faculty decide what thinking an assessment should make visible.
Inspire and Assure maps the course. The Inspire and Assure approach helps faculty examine how strongly an assessment motivates learning and how readily it can assure learning. The two-lane approach then helps them decide where work should be secured and where AI-supported practice can remain open.
AI Policy Frameworks and the AI Assessment Scale clarifies assignment-level AI expectations. The AI Assessment Scale (AIAS) gives faculty and students a shared vocabulary for what AI use is allowed on a specific task.
The Role of Authentic Assessments and UDL redesigns the task. Authentic assessment and UDL help faculty make assignments meaningful, accessible, and harder to outsource without learning.
Why Process-Oriented Assessment Works Well in the AI Era makes the process visible. Drafts, checkpoints, AI transparency, and reflection show how students arrived at a final product.
AI Detection and Accountability handles detection and accountability. Detection tools, student conversations, documentation, and institutional processes make more sense after the design questions have been addressed.

By the end of Part 2, faculty should have a practical route through assessment redesign: identify vulnerable assignments, clarify the learning evidence, decide which work is secured or open, specify allowed AI use, redesign selected tasks, build process evidence, and communicate accountability clearly.

If any framework term feels unfamiliar, use the Glossary as a quick reference while you read.

What You’ll Reuse from Part 1

Part 1 followed Sarah as she built basic AI literacy: how chatbots work, how to prompt them, how to evaluate their outputs, how to think about agentic AI, and how to disclose AI use responsibly. Part 2 shifts to Marcus, a business faculty member working through assessment design across several courses in the same program.

Several Part 1 habits return here.

R-T-C-I helps you ask AI for useful assignment-design feedback without handing over the decision.
ABC-R helps you evaluate AI outputs for accuracy, bias, context/relevance, and responsible use.
Disclosure and transparency help you communicate your own AI use and ask students to document theirs.
Agentic AI awareness helps you recognize why some old assessment assumptions no longer hold, especially in online and asynchronous work.

Part 2 uses Marcus’s assignment to show how one redesign can develop across the chapters. In your own activities, you can revisit the same assignment or choose a different one each time. Start with a usable first pass on one assignment before attempting a full course redesign.

What Has Shifted

Open your learning management system. Look at recent submissions. If you teach courses with writing assignments, you may have already seen it. Text that is fluent but generic. Responses that hit every requirement without saying anything distinctive. The work can feel different from what students produced a few years ago, even when you cannot prove why.

Generative AI has changed how students can interact with course content. In 2023 the concern was students pasting prompts into ChatGPT. By 2025 the tools had become more capable, more accessible, and more embedded in the software students already use. Marc Watkins argues that if AI itself cannot prevent students from using it to bypass learning, then the structural approach to assessment has to change. The question for faculty is how to respond.

This first chapter examines why students use AI in ways that concern us and what those patterns reveal about traditional course design.

A note before we begin. Not every assessment type is equally vulnerable to AI. Proctored exams, lab practicals, clinical simulations, and observed in-class demonstrations already provide more direct or controlled evidence of what students can do. They may require less redesign than unsupervised written or digital work, where AI can produce a finished submission without making the student’s learning visible. Faculty in nursing, CTE, and the sciences may find that much of what they already do holds up well. Use these strategies to strengthen vulnerable areas while preserving assessments that already provide trustworthy evidence.

How Students Use AI Now

Kevin Rank at Boise State University used a classroom questionnaire in four IT management courses to ask students whether they had ever used AI to cheat. He organized the responses into two patterns. This is a local teaching case, so the categories should not be treated as a general account of how all students use AI.

Outcome-focused use. Fifty-six students answered yes. Their responses included direct answer lookup (37.5%), essay-writing assistance (30.4%), and time pressure as a stated reason (25.0%). Rank described this pattern as transactional because the students emphasized getting answers, saving time, or completing work more efficiently.

Process-focused use. Fifty-four students answered maybe. The most common use was understanding or clarification (42.6%), including help with confusing material or checking work before submission. Rank described this pattern as exploratory because the students were uncertain whether this kind of support crossed a line.

One student in the “maybe” group explained it this way. “I use it to further explain a question I don’t understand. Or check my work before I submit something.”

The two groups used the same general technology for different purposes. Process-focused intent may support learning, but it does not demonstrate that learning occurred. Students still need guidance about when AI support is appropriate and how to check whether they can explain, apply, or evaluate the material without the tool.

A peer-reviewed study by Sperber et al. (2025), which Why Process-Oriented Assessment Works Well in the AI Era examines in detail, found that 58% of survey respondents preferred combined peer and AI feedback, 36% preferred peer feedback alone, and 6% preferred AI feedback alone. In this structured activity, students often treated AI feedback as a complement to human feedback rather than a replacement for it.

A subtler issue appears in metacognition. In two studies using LSAT logical-reasoning problems, Fernandes et al. (2026) found that participants using AI performed above a norm comparison while overestimating their performance. The studies measured performance calibration, not retained classroom learning. They still point to a problem faculty need to consider: a stronger AI-assisted result may not give students an accurate sense of what they can do independently.

The agentic AI escalation. Chatbots generally wait for a user to enter a prompt and then return a response. Agentic AI can carry out a delegated task across multiple steps. Depending on its permissions, an agent may open files, navigate websites, interact with software, and continue working while the user supervises or reviews the result.

In educational settings, practitioners have demonstrated agents navigating learning management systems and attempting multi-step coursework workflows. Anna Mills (2025) documented an agent interacting with Canvas. Marc Watkins (2026) reported on Einstein, a product marketed as capable of completing Canvas coursework on a student’s behalf. An Inside Higher Ed Q&A with Watkins discusses the implications for online courses. These examples do not establish that agents can reliably complete every course or assessment. They show why faculty can no longer assume a student personally completed every step inside an online assignment.

The assessment question is whether the course produces evidence that can reasonably be attributed to the student. Unlimited attempts, automated feedback, asynchronous submissions, and final-product-only grading may require another source of evidence when an agent can act inside the same digital environment.

Why Students Use AI to “Cheat”

Faculty sometimes frame unauthorized AI use as a moral failing. Students are lazy. Students do not care. Students are trying to get away with something.

Student motives vary. Abbas, Jam, and Khan (2024) found that workload and time pressure were associated with reported ChatGPT use, although their study did not establish that pressure caused misconduct. Johnston et al. (2024) found that lower academic-writing confidence was associated with greater AI use or consideration. These findings do not explain every case. They show why expediency, confidence, and academic support belong in the conversation alongside integrity.

Anna Mills (2025) frames this through intrinsic motivation. When students understand writing as a way to sharpen their own thinking (and as preparation for working effectively with AI tools), the task stops looking like busywork. But that reframing only works when the assignment itself communicates its purpose clearly. Many assignments do not.

Some students turn to AI because they are struggling. They may want another explanation, help getting started, or feedback before submitting. For a student who lacks confidence in academic writing or has difficulty accessing other support, an AI chatbot can feel easier to approach than office hours or a tutoring center. Faculty still need evidence that the student can perform the assessed capability.

Students also encounter ambiguous rules. In a survey of 733 undergraduates, Stone (2025) found permitted, prohibited, and policy-ambiguous uses in the same population; perceived peer behavior was associated with prohibited use. Liza Long (2025), a composition instructor, describes setting clear AI guidelines, requiring chat transcripts, and still finding students who submit fully AI-generated responses. Individual course policies help, but students still have to interpret different expectations across instructors and assignments.

None of these reasons excuse academic dishonesty. But understanding them helps us design courses that reduce the incentive to cheat.

The Vocational vs. Transactional Divide

One way to interpret tension around AI use is to examine the meanings faculty and students attach to course work.

Faculty may approach teaching as a vocation. We invest significant time and care into our courses. We may see our relationship with students as a kind of social contract, one where we provide expertise and feedback and students invest effort and engage authentically. When a student submits AI-generated work, it can feel like a personal betrayal.

Students may approach some courses transactionally. They are paying tuition, managing other responsibilities, and trying to earn credentials. They expect fair treatment and reasonable workloads. When an assignment feels disconnected from their goals or excessively burdensome, efficiency can become the priority. As Tawnya Means puts it in an interview with Lance Eaton, students using AI are often “not necessarily trying to cheat. They’re just being efficient.”

The vocational and transactional categories are a lens, not two fixed groups. The same student can care deeply about one course and treat another as a requirement to finish. Faculty can value learning while also designing for efficiency. AI makes the mismatch visible when an assignment asks for effort that students cannot connect to its purpose.

How Traditional Course Design Sets Students Up

Some traditional course design choices inadvertently encourage the very behaviors we want to prevent. These design features made sense before AI. They create vulnerabilities now.

The most common pattern involves visibility. When a major paper is due at the end of the semester with no intermediate checkpoints, faculty see one finished product and little evidence of how the student arrived there. If AI produced substantial parts of that product, the submission may no longer show the thinking the assignment was designed to develop. Leon Furze (2025) describes this as a “purpose problem.” A course that grades only the product can lose sight of the learning process behind it.

A second pattern involves AI exploitability. Current systems can often approximate standard outputs such as a five-paragraph essay on a common topic or answers to a problem set with known solutions. Auto-graded assessments with unlimited attempts create a similar evidence problem: a student can use AI and automated feedback to move toward correct answers while leaving the learning process invisible. Stefan Bauschard (2025) argues that educators should stop trying to outpace the technology and redesign assessments around what students can demonstrate with tools and information.

A third pattern is governance. When each instructor defines AI rules independently, students face a patchwork of expectations. A student who is allowed to use AI freely in one course and prohibited from using it in another may have difficulty remembering or interpreting the boundary for a particular assignment.

Recognizing these patterns is the first step toward redesign. Process requirements (drafts, reflections, check-ins) still need clear criteria and a genuine purpose. Jason Gulya (2025) identifies common mistakes instructors make when shifting from product to process, and the later chapters in this training will address those complexities directly.

Systemic Drivers

Rank used his Boise State classroom questionnaire to identify three systemic factors that may contribute to AI misuse.

Systemic Drivers in Rank’s Local Teaching Case
Factor	Description
Policy fragmentation	Different rules across instructors can make assignment-level expectations difficult to interpret or remember.
Assessment loopholes	Unlimited attempts, auto-graded systems, and asynchronous submissions may provide weak evidence of who completed the assessed work.
Time and perceived value	Deadline pressure and low perceived value may lead some students to prioritize speed over mastery.

Rank summarized his interpretation this way: “This is not a cheating problem. It is a governance problem in a world where powerful assistance is ambient and accessible.”

His conclusion directs attention toward course and institutional design alongside individual student decisions.

The policy-fragmentation concern also appears beyond Rank’s classes. Lance Eaton’s Syllabi Policies for Generative AI Repository contains a large collection of faculty AI policy statements with substantial variation in how instructors define acceptable use. Eaton et al. (2025) published a follow-up in EDUCAUSE Review examining how institutions create AI policies. Individual faculty still need clear course policies, but institution-level coordination can reduce unnecessary variation in terminology and expectations.

Agentic AI adds delegated action to these vulnerabilities. When an agent can navigate the same online environment as a student, unlimited-attempt quizzes and asynchronous submissions may provide even less evidence about who performed the work.

AI Assignment Audit

An AI assignment audit is a focused review of one assignment from the perspective of a student who has access to current AI tools. The purpose is diagnostic. You are looking for places where AI can produce a passing submission while leaving the student’s learning invisible.

This practice is showing up across faculty-facing AI guidance. NC State DELTA recommends asking AI to “tell on itself” by testing an assignment prompt and then asking which requirements were easiest, hardest, or least appropriate for the tool. WCET/WICHE’s AI Literacies in Practice includes an Assignment Authenticity Audit that asks whether learning is visible, situated, and owned, then adds questions about student agency, access, attribution, and one manageable revision. The language varies, but the basic move is the same: inspect the assignment before revising it.

Steve Covello’s three-step model of assignment analysis offers one concrete example. Covello examined a summative assignment, identified places where AI could support secondary parts of the work, and kept the primary assessed capability in view. That same distinction can help you decide whether an assignment needs stronger evidence of learning, a defined role for AI, or both.

Start with these questions.

What learning or disciplinary capability is this assignment supposed to demonstrate? Could a passing product be produced without demonstrating it?
Could a student produce an acceptable submission by prompting an AI chatbot? How easily?
Does the assignment require students to demonstrate process, or only produce a final product?
Is the assignment connected to something specific to your course, your students, or your context? Or could it apply to any course on this topic?
Do students see the purpose of this assignment? Do they understand how it connects to their learning and goals?
What would a student who wanted to learn gain from doing this assignment authentically?
Does the assignment assume access to a paid AI tool, a particular account, reliable connectivity, or advanced technical fluency? If AI use is optional, is there a workable route for students who do not use it?

These diagnostic questions connect to established assessment scholarship. Leon Furze (2025) argues that validity should be the starting point for assessment redesign. University of Queensland guidance hosted by TEQSA likewise begins with what the assessment is intended to measure and asks institutions to account for equitable access. If an assignment does not clearly measure what it claims to measure, AI makes that gap visible. The frameworks for addressing these gaps are the focus of the chapters ahead.

Assignments where these questions raise concerns are candidates for redesign. The audit does not ask you to fix everything at once. It gives you a clearer starting point for the rest of Part 2.

Faculty Spotlight: Marcus Hickens, Business

Marcus teaches several business courses at a mid-sized state college. In this chapter, we meet him in his Principles of Management course. When he sat down with these diagnostic questions and his syllabus, one assignment stood out immediately. His take-home case analysis asks students to read a Harvard Business Review article, summarize the key management principles, and recommend a course of action for the company described.

Marcus walked through the questions. Could a student produce an acceptable submission with AI? Easily. The prompt asks for a summary and a recommendation, precisely the kind of task current AI tools handle well. Does the assignment require students to demonstrate process? It does not. Students submit a final document with no drafts, no annotations, no evidence of how they engaged with the reading.

What struck Marcus was that nothing in the instructions would change if a student generated the entire response with AI. The prompt did not require students to connect the article to class discussions, draw on personal work experience, or defend a position in a follow-up conversation. It asked for a conventional analysis that current chatbots can often approximate.

The assignment still had value. Marcus appreciated that it required students to apply management concepts to a real scenario. But the design made authentic engagement optional. A student who wanted to learn would benefit from doing it. A student who wanted to save time could skip the learning entirely and still earn a good grade.

Practical Takeaway: Complete an AI Assignment Audit

Choose one major assignment from a course you teach. Copy the assignment instructions exactly as your students would see them.

Open an AI chatbot such as ChatGPT, Claude, Gemini, or another tool you have access to.
Paste your assignment instructions and prompt the AI with something like “Complete this assignment as written.”
Read the AI’s output carefully. Evaluate it against your rubric or grading criteria.
Ask the chatbot which requirements were easiest, which were hardest or least appropriate, and what features of the assignment made them so. Treat this response as another perspective to evaluate, not an authoritative diagnosis.

Now reflect on what you saw.

How close did the AI get to meeting your expectations? Would the submission earn a passing grade?
What did the current tool complete convincingly under this test? Where did it struggle?
What does this reveal about what your assignment actually measures?
Would a revised version assume that every student has the same tool access, connectivity, or technical fluency?

Create a short audit note with four parts.

Assignment: Name the assignment and its current purpose.
AI result: Describe what the AI could complete successfully.
Evidence gap: Identify the learning you still could not see from the final product.
First revision idea: Name one change that might make student thinking more visible.

If you want to continue with the same assignment, Rethinking Bloom’s Taxonomy for the AI Era can help you audit the learning outcome behind it. You can also choose a different assignment.

Looking Ahead

The next chapter, Rethinking Bloom’s Taxonomy for the AI Era, asks what learning outcomes need to do in this environment. It revisits Bloom’s taxonomy, introduces the UnBlooms Framework, and helps you revise an outcome so the assignment makes student thinking visible.

From there, Part 2 moves into the course map, assignment-level AI expectations, authentic and UDL-aligned redesign, process evidence, and accountability.

References

Abbas, M., Jam, F. A., & Khan, T. I. (2024). Is it harmful or helpful? Examining the causes and consequences of generative AI usage among university students. International Journal of Educational Technology in Higher Education, 21, Article 10. DOI record

Bauschard, S. (2025, May 8). Stop chasing the algorithm: Re-architect classrooms for an AI-powered future. Education Disrupted. Source link

Brandon, E., Eaton, L., Gavin, D., & Papini, A. (2025, May). In the room where it happens: Generative AI policy creation in higher education. EDUCAUSE Review. Source link

Covello, S. (2025). AI-infusion through a three-step model of assignment analysis. Teaching Repository of AI-Infused Learning. University of Central Florida Libraries. Source link

Eaton, L. (2025, August 21). What’s in your statement? Syllabi policies for generative AI repository. AI + Education = Simplified. Source link

Eaton, L. (2025, November 13). They’re not necessarily trying to cheat; they’re just being efficient (interview with Tawnya Means). AI + Education = Simplified. Source link

Fernandes, D., Villa, S., Nicholls, S., Haavisto, O., Buschek, D., Schmidt, A., Kosch, T., Shen, C., & Welsch, R. (2026). AI makes you smarter but none the wiser: The disconnect between performance and metacognition. Computers in Human Behavior, 175, Article 108779. DOI record

Furze, L. (2025, August 9). AI in the writing process: A problem of purpose. Source link

Furze, L. (2025, August 18). Five principles for rethinking assessment with Gen AI. Source link

Gunder, A., & Herron, J. (2026). AI literacies in practice: A comprehensive playbook for higher education. WCET/WICHE. Source link

Gulya, J. (2025, October 17). Problems with “process over product” (Part 1). The AI Edventure. Source link

Johnston, H., Wells, R. F., Shanks, E. M., Boey, T., & Parsons, B. N. (2024). Student perspectives on the use of generative artificial intelligence technologies in higher education. International Journal for Educational Integrity, 20, Article 2. DOI record

Long, L. (2025, May 16). ChatGPT 101: The student ethics edition. Artisanal Intelligence. Source link

Mills, A. (2025, August 23). Strategies for reducing the likelihood of AI misuse. Anna Mills’ Substack. Source link

Mills, A. (2025, October 19). The time to reckon with AI agents in digital learning spaces is now. Anna Mills’ Substack. Source link

Parker, K. B. (2026, January 13). Ask AI to tell on itself. NC State DELTA News. Source link

Rank, K. (2025, November 17). Why do students cheat? A case study from BSU [Presentation section]. AI + Academic Integrity, Idaho AI Catalyst webinar.

Sperber, L., MacArthur, M., Minnillo, S., Stillman, N., & Whithaus, C. (2025). Peer and AI review + reflection (PAIRR): A human-centered approach to formative assessment. Computers and Composition, 76, Article 102921. DOI record

Stone, B. W. (2025). Generative AI in higher education: Uncertain students, ambiguous use cases, and mercenary perspectives. Teaching of Psychology, 52(3), 347-356. DOI record

University of Queensland. (2026, March 19). Principles, criteria and standards for assessment with generative AI use. Tertiary Education Quality and Standards Agency. Source link

Warner, J., & Watkins, M. (2025, November 7). Agentic AI invading the LMS and other things we should know. Inside Higher Ed. Source link

Watkins, M. (2025, December 21). If AI can’t stop a student from cheating, how can it ever be safe? Rhetorica. Source link

Watkins, M. (2026, February 27). Einstein and the rise of nuisance tech. Rhetorica. Source link

License

Icon for the Creative Commons Attribution-NonCommercial 4.0 International License

A Guide to Teaching and Learning with Artificial Intelligence Copyright © by Jason Blomquist; Liza Long; and Joel Gladd is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, except where otherwise noted.