Education & Universities

Assessing students when everyone uses AI: what works

How lecturers can assess students who use AI: oral follow-ups, drafts, AI-use logs, local cases and in-class work, and why detectors alone are not evidence.

  • By
  • Updated
  • 4 min read
  • Facts checked
Short answer

Assess the process as well as the product: drafts, an AI-use log, a short oral follow-up, local or class-specific cases and in-class work. Do not rely on an AI detector score alone, because detectors are unreliable. Tell students the rules before the task.

Part of the guide: AI for Teachers and Lecturers in Malaysia: Use, Assess and Set Rules

On this page
  1. How Can Educators Reliably Assess Learning When AI is Widely Accessible?
  2. What Role Do Process Evidence and Draft Submissions Play in Evaluation?
  3. How Can AI-Use Logs and Transparency Be Integrated into Assignments?
  4. Are Local Context and In-Class Activities More Reliable Indicators Than Remote Submissions?
  5. What Types of Assessment Carry Higher Risks of AI Misuse?
  6. How Can We Redesign Assignments to Demand Human Value?
  7. Where to go next
  8. Sources
  9. Frequently asked questions

If every student can produce a clean essay in a minute, the essay alone no longer shows what the student knows. This guide covers assessment designs that still show thinking, and a step-by-step redesign of one assignment. AI detectors are unreliable and can accuse the wrong student, so we do not recommend using a detector score alone as evidence.

How Can Educators Reliably Assess Learning When AI is Widely Accessible?

The proliferation of generative AI tools means that simply submitting a polished final product is no longer a sufficient measure of a student’s understanding or skill acquisition. The focus must shift from the final output to the process, the application of critical thinking, and the ability to refine and critique AI-generated content. Relying solely on AI detectors is insufficient because these tools frequently produce false positives and negatives, leading to either unnecessary accusations or missed opportunities to evaluate true learning.

The core strategy involves designing assessments that require human-centric skills—synthesis, contextual application, ethical reasoning, and iterative improvement—which current AI models struggle to execute authentically without significant human guidance.

What Role Do Process Evidence and Draft Submissions Play in Evaluation?

When the final submission might be heavily influenced by AI, the journey to that submission becomes the most valuable data point. Submitting multiple drafts, annotated outlines, and early iterations provides a clear timeline of cognitive engagement. These drafts allow instructors to observe where the student struggled, what prompts they used, and how they refined the initial AI suggestions.

Process evidence moves the evaluation from "What is the answer?" to "How did the student arrive at this answer, and what did they learn along the way?" This requires a change in submission expectations, moving away from single, polished deliverables.

How Can AI-Use Logs and Transparency Be Integrated into Assignments?

Requiring students to maintain an AI-use log is a direct way to foster transparency and metacognition. This log should not just state that AI was used, but how it was used and why. For example, a student might log: "Used ChatGPT to generate three potential market segmentation ideas for the local F&B sector. I then rejected two based on local cultural nuances and developed the third idea further."

This log forces the student to engage in a critical dialogue with the tool, transforming AI from a cheat into a collaborator whose output must be vetted and improved by the human user.

Are Local Context and In-Class Activities More Reliable Indicators Than Remote Submissions?

For Malaysian SMEs, practical application within the local business environment is crucial. Assessments that mandate real-world interaction—such as interviews with local stakeholders, site visits, or the development of a solution tailored specifically to Malaysian regulatory or cultural norms—are inherently more resistant to generic AI outputs.

In-class activities, especially those requiring spontaneous discussion, debate, or immediate problem-solving based on a shared, dynamic scenario, provide immediate, observable evidence of comprehension that AI cannot replicate in a controlled setting.

What Types of Assessment Carry Higher Risks of AI Misuse?

Not all assignments are equally susceptible to AI manipulation. Some tasks are inherently generative and easily automated, while others demand unique, lived experience or complex, multi-layered synthesis.

The following table categorizes common assessment types based on their inherent risk profile concerning AI misuse:

Assessment Type AI Risk Level Rationale Recommended Mitigation Strategy
Summarizing a public article High AI excels at condensing information accurately. Require critical analysis of the summary, not just the summary itself.
Generating boilerplate business plans High AI can produce structurally sound, generic plans quickly. Mandate hyper-local data incorporation and unique strategic pivots.
Solving standard mathematical problems Medium Basic calculation is easily automated. Require detailed step-by-step reasoning and justification for each step.
Presenting a novel, real-time case study solution Low Requires immediate, context-specific decision-making. Conduct live presentations followed by Q&A sessions.
Writing a standard literature review Medium-High AI can synthesize existing academic literature efficiently. Require the student to critique the selection of sources, not just the synthesis.

How Can We Redesign Assignments to Demand Human Value?

The shift requires a deliberate redesign process. Instead of asking, "Write an essay on X," the prompt must become, "Using the AI-generated draft on X as a starting point, identify three critical flaws specific to the Malaysian SME context, and rewrite the conclusion to address these flaws."

Here is a step-by-step guide to redesigning a standard argumentative essay assignment:

  1. Identify the Core Skill: Determine what skill you actually want to test (e.g., persuasive argumentation, data interpretation, ethical reasoning).
  2. Deconstruct the Prompt: Break the original prompt into its essential components (Topic, Goal, Deliverable).
  3. Introduce the AI Component: Explicitly allow or require the use of AI to generate a baseline response for the core component.
  4. Layer the Human Task: Add a mandatory, high-value task that AI cannot perform authentically. This task must require personalization, local knowledge, or critical self-reflection.
  5. Define the Submission Structure: Require multiple artifacts: (a) The AI-generated draft, (b) The AI-use log detailing prompts and critiques, and (c) The final, human-refined submission.

Where to go next

The institution-wide rules sit in AI policy for universities in Malaysia, and the classroom side in AI for teachers and lecturers. A one-day AI course for lecturers ends with a redesigned assignment. Faculties can book AI training for universities.

Sources

Links below checked November 2026.

Frequently asked questions

Do AI detectors work?

They are unreliable and can falsely accuse students. Treat a score as a prompt for a conversation, never as proof.

What is the quickest assignment fix?

Add a short oral follow-up and ask for drafts or an AI-use log alongside the final text.

Should students declare AI use?

Many Malaysian universities require it. State in each course outline what may be used and what must be declared.