AI Resistant Assessment

Three Approaches to AI-Resistant Assessment

If artificial intelligence can produce the answer, what exactly are we assessing?

This question has become increasingly difficult to avoid. Generative artificial intelligence can do many things—it can compose seemingly flawless prose,  solve mathematical problems, summarize complex texts, generate scientific explanations, and produce remarkably sophisticated responses to questions that once required considerable independent effort. For educators, this presents an important challenge: a completed assignment can no longer be assumed to provide reliable evidence of what a student knows, understands, or can do independently.

One response has been to pursue increasingly sophisticated methods of AI detection. Yet research continues to raise concerns about the reliability and fairness of these technologies, particularly when detection results become the basis for academic-integrity decisions. Perhaps a better response (and one which we will consider in this month's edition involves reconsindering the conditions under which learning becomes visible).

This brings us to AI-resistant assessment.

What is AI-resistant assessment?

Broadly, this refers to any assessment designed to make students' own understanding, reasoning, and intellectual development observable, while reducing opportunities to substitute AI-generated work for independent learning.

Here are three approaches that teachers across primary and secondary classrooms can begin implementing immediately. Feel free to try them out! If you do (and if you're willing to share your findings) I would love to continue the conversation.

1. Process-based assessment

TLDR: Evaluate the thinking behind the product

Traditional assessment often privileges the final product: the completed essay, solved equation, research report, or presentation. Although these products can demonstrate learning, they reveal comparatively little about the intellectual decisions that produced them.

Process-based assessment introduces opportunities to observe learning as it develops. Instead of evaluating a finished assignment in isolation, teachers collect evidence of planning, experimentation, revision, feedback, and reflection.

Consider a research paper. Rather than assessing only the final submission, a teacher might also examine the student's initial question, preliminary evidence, evolving thesis, and reflections on revision. These intermediate artifacts provide a richer account of the student's intellectual development.

Importantly, this approach also emphasizes metacognition: the ability to examine and regulate one's own thinking. Research summarized by the Education Endowment Foundation identifies explicit instruction in planning, monitoring, and evaluating learning as an effective educational practice across primary and secondary settings.

In primary classrooms: Students completing a science investigation might draw their initial predictions, record observations, and explain how their understanding changed. In mathematics, students can preserve an early solution alongside a corrected solution, explaining what they discovered about their original reasoning. Younger learners might use photographs, drawings, or brief teacher-guided reflections.

In secondary classrooms: Teachers can introduce writing portfolios, annotated drafts, research journals, laboratory notebooks, or problem-solving logs. A history student, for instance, might document how encountering a contradictory primary source caused them to revise an interpretation. An English student might identify the most consequential revision made to an argument and justify it.

Try this tomorrow: Add two questions to an existing assignment: What did you initially think, and what caused your thinking to change? Ask students to respond using evidence from their actual work.

The purpose is not to generate more paperwork. It is to make the development of understanding part of the evidence being assessed. Process documentation is not, by itself, proof of independent authorship, but it becomes considerably more meaningful when paired with work teachers have directly observed.

2. Use dialogic assessment strategies

TLDR: Make student thinking audible

One of the oldest educational practices may also be among the most useful in the age of generative AI: asking students to explain their thinking.

Dialogic assessment uses conversation, questioning, and intellectual exchange to evaluate understanding. Rather than relying exclusively on a written response, the teacher invites students to articulate how they arrived at an interpretation, defend a conclusion, or reconsider an assumption.

This approach draws upon a longstanding tradition of oral examination and Socratic inquiry. A student may submit a compelling argument about a historical event. But can that student explain why one source was more persuasive than another? Can they identify the limitations of their conclusion? Can they respond thoughtfully when presented with contradictory evidence? Might you consider asking them to do this (either in speech or writing)?

These questions reveal dimensions of learning that a polished final product may conceal.

In primary classrooms: Consider incorporating a brief "teach me how" conversation during mathematics, reading, and science. After completing a problem, students explain their reasoning using manipulatives, drawings, or verbal descriptions. Following a story, they might explain why a character made a particular decision and identify evidence supporting their interpretation.

In secondary classrooms: Introduce two- to three-minute oral defenses following major assignments. Ask students to explain a central claim, justify methodological choices, or respond to an unexpected question. In mathematics, students might explain why a solution method works; in science, they might defend an experimental conclusion; in English, they might reconsider an interpretation in light of a different passage.

Try this tomorrow: Following an assignment, ask students three questions:

• What is your strongest idea?

• What evidence supports it?

• What might cause you to reconsider it?

The great part is that these conversations do not need to become formal presentations. They can occur during ordinary classroom circulation, small-group conferences, or brief individual check-ins.

3. Try situated performance assessment

TLDR: Emphasize the transfer of knowledge through adaptation

A third approach involves designing assessments in which students demonstrate learning through an observable performance, often in response to a specific context, unfamiliar problem, or immediate classroom experience.

Situated performance assessment emphasizes the application and transfer of knowledge. Instead of asking students merely to reproduce information, teachers ask them to use what they know in circumstances that require judgment, adaptation, or explanation.

This approach can include hands-on investigations, close reading of an unfamiliar passage, interpretation of newly presented data, live demonstrations, or short written responses completed during class.

The defining feature is that the teacher directly observes a student applying knowledge and can connect the performance to a clearly identified learning objective.

Try this tomorrow: Take an existing homework question and create a five- to ten-minute classroom variation that asks students to apply the same knowledge to a new example. Observe their reasoning, collect the response, and compare it with their prior work.

A few closing thoughts

As most educators know (or realize quickly!), none of these approaches is entirely immune to artificial intelligence. I don't think, however, that the objective is to make academic dishonesty impossible. Rather, I want to encourage you to think about how we can create more trustworthy, varied, and educationally meaningful evidence of learning already occurring.

Moreover, these three approaches need not compete. A single assignment might incorporate a documented process, a short conversation, and an independent application task. Together, they provide a more comprehensive picture of what a student understands than any one finished product could offer.

I leave you with a closing question to consider as you continue to design meaningful assessment: What evidence will convince me that meaningful learning has occurred?

And to that end, Who has done the consequential thinking?

These questions have been around much longer than artificial intelligence itself. It strikes me as particularly timely (and urgent) that we continue asking them every day.

Further Reading

• Education Endowment Foundation: Metacognition and Self-Regulated Learning — Research on making thinking, reflection, and learning strategies explicit.

• UNESCO: What's Worth Measuring? The Future of Assessment in the AI Age — A discussion of higher-order thinking, creativity, and ethical reasoning in assessment.

• Sun, Liao, and Ma (2026): Trusting AI to Detect AI? — Research examining AI detection tools and the case for process-oriented assessment.

Coming Next

Future editions of The AI Chronicles will continue examining assessment from several angles, including:

authentic assessment and oral defense; designing assignments around process rather than product; AI literacy as an assessable skill; what AI detection can and cannot tell us; the role of handwritten and in-class work; and what students themselves think meaningful assessment should look like.

Missed our discussion on NPR last month? Catch it below!