How to Design an Assignment That Still Works When Students Have ChatGPT

College instructors across higher education recognize the frustration of evaluating contemporary student submissions. Essays arrive with polished grammar, coherent transitions, and clean bibliographic formatting. Yet after reading three paragraphs, the instructor realizes they cannot determine whether the student spent twenty hours researching the topic or sixty seconds prompting a foundation model.

For generations, written submissions functioned as a dependable proxy for student understanding. Writing an extensive paper demanded cognitive labor: searching archives, evaluating evidence, and refining arguments. Producing text served as a filter for comprehension.

Generative artificial intelligence permanently dissolved that proxy. Large language models reduce the marginal cost of fluent prose to zero. A student can paste an assignment prompt into a browser window and receive an essay complete with topic sentences and synthesized claims.

When software generates plausible papers on command, text stops serving as evidence of understanding. Banning digital tools fails in practice. Instructors need assignments structured around direct verification, where students explain decisions, apply concepts to unfamiliar problems, and verbally defend their conclusions.

Why Defensive Countermeasures Failed

When large language models entered college classrooms, institutions initially reached for defensive mechanisms. Over four semesters, these technical countermeasures collapsed under practical scrutiny.

First, universities deployed statistical AI detection software. Within eighteen months, prominent institutions including Vanderbilt, Northwestern, and UT Austin disabled these tools. Research by Liang and colleagues at Stanford University revealed severe false-positive rates, disproportionately penalizing international students and non-native English speakers. Accusing an innocent student of dishonesty based on an uninterpretable probability score damages trust and creates administrative appeals. For details, see our guide on AI detection vs. verification of learning.

Second, instructors attempted prompt engineering. Faculty asked students to write personal memoirs, only to discover that modern models generate moving personal reflections on command. Mandating recent news citations failed because frontier models browse the live internet in real time.

Surveillance software and defensive prompt tweaks fail because they target the wrong problem. Restoring educational integrity requires changing what faculty actually assess.

The Pedagogical Anchor: Assess the Student, Not the Submission

Solving this challenge requires an intentional shift in course design. In their 2024 report for Australia's TEQSA, researchers Margaret Bearman and Phillip Dawson argue that universities must prioritize assessment validity over punitive surveillance. Assessment must evaluate what students can articulate, justify, and execute.

At Rocketproof, this pedagogical orientation centers on a direct guideline: assess the student, not the submission alone.

The written research paper, technical blueprint, or software repository remains a valuable milestone during learning. That document serves as an exhibit of student preparation, while spoken demonstration provides the evaluative verdict. Authentic assessment evaluates whether the student understands the disciplinary principles embedded within their exhibit.

When students understand that grades depend on explaining reasoning in a spoken conversation, preparation changes. Students cannot rely on outsourced text because they must account for assertions and trade-offs in their own voice. To understand how institutions scale this approach, see our guide to scalable authentic assessment.

Five Principles for AI-Resilient Assignment Design

Instructors can design coursework that remains rigorous when students have access to generative models by applying five structural principles.

1. Decouple Preparation from Evaluated Demonstration

Monitoring how students draft outlines creates unnecessary friction. Faculty separate research preparation from evaluated demonstration.

Instructors permit digital tools during preliminary research. Then, faculty evaluate mastery through an interactive conversation where students justify analytical choices. If a student uses AI during brainstorming, they must still master the underlying logic. A student unable to explain why their paper applied a specific theoretical model reveals an immediate comprehension gap.

2. Probe the Decision Seams

Language models generate prose presenting smooth consensus while glossing over disciplinary trade-offs.

Resilient assignments evaluate the decision seams of a project. Instructors formulate questions targeting professional disagreements:

• Which alternative methodology did you reject, and what evidence prompted that choice?
• If your baseline data changed by twenty percent, how would your operational recommendations shift?
• Where does your conclusion contradict opposing perspectives in assigned course readings?

Researchers who conducted the work answer promptly. Students who submitted generated text struggle because summaries cannot convey procedural reasoning.

3. Integrate Structured Asynchronous Oral Defense

Verbal examination represents the oldest evaluation method in higher education. European universities relied on viva voce defenses for centuries because spoken dialogue reveals genuine intellectual competence.

Live interviews with one hundred students historically required thirty faculty hours. Structured oral defense platforms make verbal evaluation sustainable. Students submit work and complete an asynchronous defense answering targeted questions grounded in their submission. Faculty evaluate timestamped spoken responses in three minutes, avoiding long hours spent parsing questionable text. For details, see how oral assessment works in a class of 100 students.

4. Replace Static Case Prompts with Dynamic Simulations

Static case studies invite passive summarization. When faculty assign a traditional case memo, software outlines the dilemma and recommends strategy in seconds.

Resilient assignments place students inside dynamic conversational scenarios. In AI-powered simulations, students interact with calibrated personas. A nursing student interviews an anxious patient, or a business student negotiates with a skeptical buyer. Students react live under professional constraints, demonstrating practical judgment that static written text cannot measure.

5. Decouple Ungraded Rehearsal from Summative Grading

Unfamiliarity with oral defenses or simulations can induce student anxiety. If a new format carries immediate high-stakes grading, evaluative anxiety interferes with authentic performance.

Effective course design provides ungraded private practice. Instructors allow students to complete rehearsal attempts where they test audio settings, answer sample prompts, and adjust to conversational pacing without grade penalties. Practicing in a private sandbox builds confidence and aligns with Universal Design for Learning.

Four Concrete Disciplinary Blueprints

To see how these principles function across academic divisions, consider four assignment redesigns implemented across undergraduate and graduate curricula.

1. Public Policy and Economics (PUBP 410)

• Traditional Deliverable: Students submit a twenty-page policy memo on transit congestion pricing.
• Redesigned Blueprint: Students submit a four-page brief outlining toll schedules and revenue allocations, then complete a five-minute defense answering three targeted prompts:
  1. "Defend choosing a cordon toll over dynamic pricing against readings on transportation equity."
  2. "If bus ridership drops twelve percent after launch, which capital project absorbs the deficit?"
  3. "Explain the empirical assumptions behind your price elasticity coefficient in Table 3." The defense verifies whether the student commands transportation economics or pasted generic models into a template.

2. Nursing and Clinical Health Sciences (NURS 350)

• Traditional Deliverable: Students submit a static care plan for a diabetic patient presenting with foot ulcers.
• Redesigned Blueprint: Students enter an interactive nursing simulation. An AI persona portrays an elderly patient exhibiting frustration, dietary non-adherence, and vascular symptoms. The student conducts a focused history intake, orders diagnostic labs, and communicates findings using the SBAR protocol. The instructor evaluates clinical judgment across the recorded dialogue, confirming patient communication competence.

3. Computer Science and Systems Programming (CS 360)

• Traditional Deliverable: Students write an asynchronous proxy in C and submit code to an autograder.
• Redesigned Blueprint: Students commit code, then complete a code defense through our computer science solutions. The platform analyzes the repository and generates source-grounded technical prompts:
  1. "In your socket routine between lines 64 and 88, explain why you selected epoll over select primitives."
  2. "Walk through how your buffer allocation logic handles slow connections without memory leaks." The defense confirms that the student wrote and understood systems code, avoiding uninspected code generated by software.

4. Business Management and Organizational Strategy (MGMT 480)

• Traditional Deliverable: Students write an eight-page paper detailing negotiation strategies for renewing an industrial supply contract.
• Redesigned Blueprint: Students enter a business negotiation simulation, speaking with an AI persona representing a hostile manufacturing vendor. The vendor holds private operational constraints, raw material inflation data, and firm walk-away thresholds. The student probes underlying interests, defends price points, and navigates pushback in real time. The instructor evaluates principled negotiation, objective criteria, and strategic composure.

Sustainable Faculty Workflows and LMS Integration

A common faculty concern is whether conversational verification increases instructor grading workloads. Structured authentic assessment reduces grading fatigue by organizing student evidence.

In Rocketproof, artificial intelligence never issues autonomous grades. The platform serves as an evidence aggregator, not a substitute for faculty judgment. When a student completes a defense or simulation, the system generates an indexed transcript, time-stamped audio playback, and preliminary rubric alignments grounded in verbatim student quotes.

Instructors open an evaluation console presenting the submission alongside the defense transcript and audio controls. Instead of spending twenty minutes reading questionable prose, faculty review key spoken exchanges at 1.5x speed, verify highlighted rubric alignments, and finalize scores.

Through native LMS integration using LTI 1.3 Advantage standards, finalized grades synchronize directly to Canvas, Blackboard, D2L Brightspace, or Moodle gradebooks.

Universal Design and Student Privacy

Authentic assessment must remain accessible and fair for all learners. Rocketproof supports Universal Design for Learning by providing flexible multi-modal response options, consistent with established standards for accessibility in higher education.

Students complete defenses through spoken audio or conversational text dialogue. For students with speech accommodations or temporary medical conditions, the text engine evaluates answers against identical rubric criteria without penalizing verbal pacing. Instructors can adjust timers, enable cognitive processing pauses, or permit assistive devices.

Furthermore, Rocketproof complies with FERPA regulations and maintains tenant isolation. Student recordings and submissions are never used to train commercial foundation models.

A Practical Four-Step Implementation Roadmap

Faculty can adapt existing assignments for the AI era using four concrete steps:

1. Trim the Written Deliverable: Reduce required page counts by forty to fifty percent. Direct students to focus on core claims, data exhibits, and analytical calculations.
2. Formulate Three Decision-Seam Defense Prompts: Craft questions requiring students to justify methodology, defend trade-offs, and predict outcomes under altered conditions.
3. Reallocate Rubric Weighting: Assign forty to fifty percent of the total project grade to the defense or simulation, establishing that spoken demonstration carries equal academic weight.
4. Deploy Asynchronous Verification with Practice: Provide students with an ungraded rehearsal mode before opening the summative submission, reducing performance anxiety.

Frequently Asked Questions

How can instructors manage oral defenses in classes with 100 or more students?

Asynchronous oral defenses eliminate individual scheduling appointments. Students record spoken responses on personal devices within assigned submission windows. Instructors review an indexed console displaying transcripts and audio segments, evaluating an asynchronous defense in three to four minutes per student.

How does oral defense accommodate students with speech anxiety or disabilities?

Universal design requires flexible response modes. Rocketproof supports conversational text chat for students with speech accommodations, allowing them to defend work in writing under identical analytical criteria. Instructors can also provide untimed preparation windows and private rehearsal attempts to reduce anxiety.

Should faculty ban ChatGPT in the course syllabus?

Banning generative software outright creates unenforceable rules and adversarial relationships. A sustainable policy defines acceptable AI assistance during preliminary brainstorming, while establishing that final grades depend on demonstrating independent comprehension during the defense.

Does oral assessment require intrusive video proctoring or webcam surveillance?

No. Authentic assessment avoids room scans, gaze tracking, and browser lockdowns. Assessment measures cognitive understanding through dialogue. Rocketproof captures spoken audio and submission exhibits, respecting student privacy while verifying individual authorship.

Moving Beyond Text Generation

Generative artificial intelligence does not threaten academic rigor in higher education. It challenges the assumption that generating a written document guarantees student learning.

When instructors redesign assignments around conversational defense, realistic simulation, and authentic verification, students re-engage with foundational concepts. Students stop searching for prompts that generate acceptable prose and start focusing on understanding the ideas they submit.

To learn how Rocketproof supports assignment redesign across departments, explore our authentic assessment platform or request a faculty pilot.