Assessment in the AI Era

When grading demands increase, many faculty face a familiar question: Are my assessments still capturing what students have actually learned?

That question isn’t new, but AI has made it harder to ignore. When students can generate a plausible case analysis or reflection in seconds, assessments designed for a pre-AI classroom may no longer reliably distinguish between demonstrated learning and generated output. Likewise, during grading season, long-standing inefficiencies in feedback practices also become more visible.

This teaching guide addresses both pressures. It considers whether current assessments align with the environment students now navigate and offers practical strategies for grading.

What's on this page:

  1. The assessment challenge — what’s changed and what it means
  2. Making Thinking Visible: Assessment Through a ReCAP Lens
  3. Using AI to support assessment and grading
  4. Try it now: CBS tools and resources

New to ReCAP?

The activities and tools on this page are built around ReCAP, the Samberg Institute’s framework for structuring active learning across four phases: Readiness, Check for Understanding, Apply & Practice, and Reflection. Everything here stands on its own, and the framework will become familiar as you read. If you want the fuller picture, the ReCAP teaching guide has a short video introduction, downloadable handouts, and the Gemini Gem referenced in Section 4.

 

1.  The Assessment Challenge

What hasn't changed

Students still attain mastery when the assessment design provides meaningful tasks. The evidence base is robust: frequent low-stakes assessment improves long-term retention (Roediger & Karpicke, 2006), and assessments that ask students to apply and transfer knowledge, rather than reproduce it, result in more durable learning outcomes (Wiggins & McTighe, 2005). However, generative AI now complicates our ability to measure that learning.

Formative and summative assessment, and why the distinction matters more now

Two essential functions underpin the assessment landscape: measuring established outcomes and illuminating the path toward them. Summative assessments, such as final exams, end-of-term papers, and capstone deliverables, assess what students have learned. Formative assessments, such as low-stakes quizzes, exit tickets, brief reflections, and problem sets, support and make the learning visible as it’s happening. While both remain essential, an AI-saturated environment places a new premium on formative assessment.

This shift is the “gift” AI has inadvertently given us: it has made the case for formative assessment impossible to ignore. When a student can generate a plausible final paper with minimal cognitive effort, the summative deliverable alone provides less reliable evidence of learning. But in-class performance—discussion, real-time problem-solving, and think-pair-share—cannot be outsourced. The thinking has to happen in the room. The Samberg teaching guide, Bridging the AI Engagement Gap, covers these strategies in depth; its section on AI-resistant activity design serves as a useful companion to the framework below.

Formative assessment also enables just-in-time feedback, arriving at a moment when students can still apply it. A student who receives substantive feedback mid-semester, while there’s still time to act on it, learns more than a student who receives the same feedback on a terminal deliverable (Hattie & Timperley, 2007). The question isn’t just whether you’re assessing; it's whether the assessment is happening at a point where it can actually change something.

The importance of timely, actionable feedback

Feedback is only as useful as the window of opportunity it creates: responses that arrive quickly and identify a concrete next step produce significantly better outcomes than comprehensive end-of-term annotation. This necessitates a strategic approach to grading: use rubrics to keep feedback targeted, prioritizing fast, one-sentence "just-in-time" responses for formative work while reserving honest, forward-looking evaluations for summative deliverables.

A note on rubrics

Rubrics are the connective tissue between assessment design and feedback. A well-constructed rubric makes your expectations legible to students before they submit, speeds your grading, and keeps your feedback consistent across a large stack. The most effective rubrics are narrow: three to four criteria, described in language that maps directly to the assignment, with enough specificity that students know what “meeting the standard” actually looks like. A “single-point rubric”—one column describing proficiency, with space to note where work falls short or exceeds it—is an easy way to focus your feedback where it actually matters, rather than calibrating across four levels of performance, you note only what fell short or exceeded the standard, which tends to produce more precise and actionable comments (Fluckiger, 2010).

What has fundamentally changed

The core challenge AI introduces to assessment isn’t just an increase in academic dishonesty; it is the decoupling of thinking from output. For the first time, an assessment designed to surface a student’s thought process can instead surface a generated product. The two are no longer reliably the same thing.

Research on post-AI assessment design suggests that the most durable response is not detection or restriction but intentional redesign (Boud & Bearman, 2023); effective redesign centers on tasks where student thinking is not just visible but genuinely irreplaceable.

 

2.  Making Thinking Visible: Assessment Through a ReCAP Lens

Assessment doesn’t have to mean a single high-stakes deliverable at the end of a unit. Each phase of ReCAP is an opportunity to make student thinking visible in ways that are lighter to grade, harder to shortcut, and more informative about what students actually understand.

ReCAP Phase

What it looks like (human-led)

Assessment applicationAI can help you build this

Re - Readiness

Pre-class activators, concept primers, curiosity-gap questions.

Low-stakes prediction tasks: Ask for position statements before class. Personal experience and "gut" predictions are difficult to outsource.Generate a pre-class prompt [topic] that asks students to take a position before reading.

C – Check for Understanding

Real-time polls,  think-pair-share.Active Comprehension: Use instant, class-specific checks during the session. These require real-time processing and immediate engagement that AI cannot prep in advance.Write 6 comprehension questions on [case/topic] at varying difficulty levels to use during the lecture.
A – Apply & PracticeScenario-based problems, role-play, decision simulations.Authentic Assessments: Use real-world constraints (e.g., $40K budget, actual company data). Specificity exposes generic AI output as insufficient.Create a constrained scenario in which students apply [framework] under specific [budget/time] constraints.
P – Prompt ReflectionMetacognitive prompts, synthesis questions, and exit tickets.Process Portfolios & Exit Reflections: Ask students to narrate how their thinking evolved during the session. This captures the growth, which AI cannot replicate.Write reflection prompts connecting [today’s topic] to the student's specific professional experience.

A note on AI-generated assessment tasks: as with activities, every draft needs your review for accuracy, alignment with your specific learning goals, and appropriateness for your students. The prompts in the table above are starting points; it is your course knowledge that makes them useful.

3.  Using AI to Support Assessment and Grading

While AI complicates assessment design, it can also make the grading process more manageable if you remain deliberate about where it adds value. 

Where AI can help:

  • Drafting rubric language: AI can quickly generate descriptions across proficiency levels. You provide the standards; let the AI provide the initial phrasing.
  • Summarizing patterns: Upload a set of anonymized submissions to identify recurring conceptual gaps. This helps you draft a single, high-impact class debrief rather than repeating the same feedback 60 times.
  • Drafting formative comments: Use AI to draft clear explanations for common technical errors. You edit for tone and accuracy before sharing.

Where human judgment is essential:

  • Evaluating argument quality: AI identifies structure, but it cannot judge if a strategic recommendation is sound within your specific course framework.
  • Personalized feedback: Students value the "human signal." For students making significant strides or facing unique struggles, your direct voice is irreplaceable.
  • High-stakes evaluation: Only provide feedback on final deliverables that you can support as your own professional judgment.

4.  Try It Now

CBS tools to use

Gemini Gem (CBS ReCAP Gem): TYour fastest starting point. The ReCAP Gem is pre-configured for CBS faculty — input your course topic, session goal, and student level, and it generates draft activities and assessment prompts for each ReCAP phase in under a minute. Treat it as a springboard, not a finished product.
 

All Day TA: CBS’s AI-powered course assistant, currently in pilot. Useful for supporting student questions outside class time, freeing your class sessions for higher-order work. Reach out to the Samberg team to explore use cases for your course.

Other AI tools faculty are using

ChatGPT / Claude:  Good for generating rubric language, scenario variations, and assessment prompt drafts. Works best with a specific prompt that includes your course level, topic, and what you want students to demonstrate — not just what they should know.

NotebookLM (Google):  Particularly useful for generating formative questions from your own course documents. Upload a case or article and ask it to generate questions tied to that specific text — it stays grounded in what you gave it.

 

 

5. Bibliography

Sources cited in this teaching guide, listed alphabetically. APA 7th edition.

Boud, D., & Bearman, M. (2023). Rethinking assessment for a digital world. Assessment & Evaluation in Higher Education, 48(4), 1–12. https://doi.org/10.1080/02602938.2023.2165524


Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81–112. https://doi.org/10.3102/003465430298487

Fluckiger, J. (2010). Single point rubric: A tool for responsible student self-assessment. Teacher Education Faculty Publications, 5. https://doi.org/10.1080/87567555.2010.507023


Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x


Wiggins, G., & McTighe, J. (2005). Understanding by design (2nd ed.). Association for Supervision and Curriculum Development.