Assessment in the Age of AI

“…purposeful teaching begins by defining where students need to go and [only] then using genAI, when appropriate, to help them get there. Never the other way around.” — Ludwig & Zakrajsek, The Science of Learning Meets AI

Introduction

Good assignment design has always started with the same question: what should students be able to do by the end of this? Answering this question clearly and then carefully designing an assignment that considers how AI might supplement or interfere with a student achieving that objective is key for effective assessment in the age of AI.

The four steps below provide a guide for this process:

1. Defining a strong learning objective
2. Protecting the skills AI can’t be allowed to replace
3. Stress-testing your work
4. Communicating your expectations clearly.

These steps can help you build (or refine) assignments with confidence, whether AI is part of the picture or not.

Note: throughout this article, we provide sample AI prompts that you can use in your assignment planning. You might consider using one of WashU’s supported AI tools to try out these prompts. If you do, please remember the following: 🔏 Maintain privacy and data security! Using WashU enterprise models can help, but it is still important to be careful what you put into an AI model. Never input sensitive student information, unpublished research, or proprietary institutional data into AI models. 🎓 You are the expert! AI outputs can be generic, biased, and uninspired. Be sure to rely on your own subject-matter expertise for final decisions. 🔎 Verify the outputs! AI can generate some really good sounding, completely fake information, including citations and references. Be sure to always verify AI-produced content. Additionally, one prompt, even if it is amazing, may not be enough. Iteration is the process of continuing a conversation with AI by asking for changes, providing more details, and etc. If you get a less-than-ideal output the first go-around, providing more context, asking more questions, and being more specific may help.

Step 1: Identify your learning objective 

Creating an AI-aware assignment starts in the same place as every good assignment always has: with a clear learning objective (sometimes called learning goals). In the AI era, in addition to identifying what should students know or be able to do and what evidence would demonstrate this mastery, it is also important to consider how AI might support or undermine student learning. 

In order to address this question, break your learning objective down into the individual tasks required to get there, then sort those tasks into two piles: the ones that are essential to the learning objective itself and the ones that are more supportive of the journey. You will likely find tasks that are core skills and those that are more supportive of those core skills. For those that are core, it is important to build resilience in to protect student learning. For more supportive tasks, AI use might be less of a concern. Breaking a learning objective into individual tasks allows you to think strategically about how you want students to engage with AI in your course – whether that be totally, a little bit, or none at all. 

Try this prompt: “I want to write a SMART learning objective for [course/topic] where students should be able to [skill or task]. Before writing it, ask me any clarifying questions you need about the course level, discipline, timeline, or how mastery will be assessed — then suggest a SMART objective with a measurable action verb from Bloom's Taxonomy.”

Step 2: Protect the foundational skills 

Unfortunately, AI can complete many of the tasks we assign as proof of foundational skills… both quickly and, often, decently. However, just because AI can do something doesn’t mean students shouldn’t still learn to do it: many foundational skills remain essential to their careers and lives as engaged citizens who can critically think. So how do we protect that learning even as AI gets better at doing the work for them? 

The strategies below are useful whether your goal is to exclude, minimize, or fully integrate AI. While these strategies do make outsourcing learning to AI harder (though not impossible), they also offer opportunities for increasing intrinsic motivation to an assignment regardless of whether AI use is allowed or not. 

• Purpose and meaning. Help students see why the underlying skill matters, beyond the assignment itself. This can be done by situating the learning within a student’s own context or offering opportunities for authentic exploration (e.g.: engaging them in real-world problem solving). Personal/local context can add AI resilience to an assignment where AI use detracts from learning, or AI use can be incorporated to facilitate the contextualization of the assignment.  

• Human-in-the-loop. Design tasks so that AI cannot effectively complete the assignment alone. Including human judgment and/or interaction in the process not only mimics real-world use of AI, but also trains students to use it with a critical eye. The Peer and AI Review and Reflection (PAIRR) Framework is a great example of how AI can be used in combination with peer review to provide students with additional feedback. In-class/in-person opportunities for students to collaborate, discuss, share, and exercise critique are excellent ways to apply human-in-the-loop design, whether used as part of an AI-integrated assignment or as a way to dissuade AI use on an assignment altogether. 

• Process over product. Grade and structure feedback around how students got to an answer, not just the end product. This might include grading components of a product as opposed to the final product in totality, for example assigning points to planning/progress/revision logs, peer review, planning artifacts (e.g.: concept maps, literature reviews), and etc. ProcessFeedback is a useful (and free!) tool that enables visualization of the writing or coding process, opportunities for self-reflection and feedback, abundant user support, and tools for scaling use for large-enrolling courses (Adhikari, 2023). Focusing on process rather than product highlights critical parts of the learning process (e.g.: brainstorming, connecting ideas, etc.) which may even get at the learning objectives we are truly intending to engage via the assignment. It also offers abundant opportunity for students to be reflective about what and how they are learning by focusing on the journey as opposed to the destination. More information an examples can be found in this resource: Focusing Assessment on Process and Product. 

• Metacognitive friction. Build in moments where students have to reflect on, explain, or justify their own thinking or choices. Revision logs and reflective memos would be examples, but there are many others, as well. You might have students argue against their own conclusion, identify and explain parts of their product in which they are least confident, require justification for decisions, predict an outcome, or self-grade using checklists or rubrics. Notably, metacognitive friction is different than administrative friction: added tasks must add meaning (not just extra time!) to the learning experience. Importantly, some students may internalize friction in non-productive ways (e.g.: “I am bad at math” as opposed to “this is a challenging problem”). Thus, it is important to communicate that growth happens through challenge and self-reflection – that they are engaging in productive struggle, not just struggling. 

• Scaffolding and Transparency. Currently, literature suggests that unguided student use of AI can compromise learning, while carefully structured use of AI can supplement it (Bastani, 2025; Kestin, 2025). Providing scaffolded AI use alongside human-required tasks can help make sure AI is being used to benefit, rather than detract from, their learning. When AI use is scaffolded and thus varies step-by-step on an assignment, it is all the more critical to provide detailed instructions. Be upfront with students about where and how AI use is expected, allowed, or off-limits, and why. We will explore more about transparency in Step 4. 

Note: If you plan to incorporate AI into any step of the learning process, be sure that you provide resources for students to learn how to use the tools effectively. The AI Literacy for the WashU Scholar Canvas module provides general information about how AI works, but providing additional, tool-specific resources to students is also important. 

Try this prompt: “Here's my assignment: [insert assignment description or upload assignment]. Before suggesting anything, ask me any clarifying questions you need about my learning objective, course level, or where AI use is currently allowed. Then suggest specific ways to build in [purpose and meaning, human-in-the-loop checkpoints, process over product, metacognitive friction, and scaffolded AI use], with concrete examples for each.”

Step 3: Check your work 

Whether your goal is to carefully guide or restrict the use of AI, the following resources can help you analyze your assignment for both opportunities and vulnerabilities: 

• This AI Vulnerabilities + Opportunities Checklist can help you to reflect on where AI might quietly undercut the skill you’re trying to build or where it might genuinely support learning instead. Note: this resource does not require AI for use. 

• Try running an AI Assignment Completion Audit to walk through how a student could realistically complete the assignment effectively. Based on results, brainstorm with AI to help you build more AI resilience into the assignment. Note: this resource requires the use of AI. 

• Check your assignment against an AI Vulnerability Rubric to identify where your assignment might need guardrails. Note: this resource does not require AI for use. 

Step 4: Communicate transparently 

Even the most thoughtfully designed assignment won’t work if students don’t understand what’s expected of them. The following resources can help you communicate clearly: 

• The AI Assessment Scale (AIAS) offers a simple, shared language for specifying how AI use is appropriate for a given assignment (even if that means no use at all!). However, it may be helpful to adjust the AIAS levels to fit the needs of your class. If you don’t need to communicate 5 levels of AI use, customize this framework to one that makes sense to your course. Additionally, consider revising the language so that it fits your voice/makes the most sense to you and your students. 

• The TILT (Transparency in Learning and Teaching) Framework prioritizes communication of an assignment’s purpose, task, and criteria for success up front which can help students understand not just what to do, but why it matters and how they’ll be evaluated. We have built out the TILT assignment template to include information specific to AI based on the AIAS – we call this the TILT-AI template. The TILT-AI template helps communicate AI use policies that include full use of AI, absolutely no use of AI, and everything in between.
• The Provost’s Navigating Artificial Intelligence page as well as the CTL’s Language for Course Policies on Artificial Intelligence resource provide guidance and examples of Syllabus language.
  

If you think about it, none of these four steps are exclusively about AI. They’re about the same thing good assignment design has always been about: knowing exactly what you want students to learn, and making sure your assignment gets them there. AI just raises the stakes for doing that well. A clear learning objective (Step 1) tells you what matters. Protecting foundational skills (Step 2) tells you how to keep that learning intact. Checking your work (Step 3) can help you determine whether your assignment holds up under real conditions. And communicating clearly (Step 4) makes sure students know exactly what’s expected of them, so your intentions and your assignment actually match. Since the technology and impact of AI is constantly evolving, these specifics will too. However, what won’t change is the value of an assignment built with this much intention behind it!

Author note: Anna Cunningham worked collaboratively with Claude Sonnet 5 to write this resource. A content outline and first draft was created by Anna and Claude was used to copyedit and provide guidance on the summary and conclusion. Final copyediting was conducted by Rick Moore. However, all those m-dashes, exclamation points, and emojis? Those were all Anna ☺️

Additional Resources

Assessment in the Age of AI

Show Your Work: Assessment in the Age of AI (UCF) discusses ways to build assessments that facilitate AI co-creation, learning without AI, and AI friction.