If AI Can Write Essays, What Should Universities Assess?

As AI makes polished academic writing widely accessible, universities face a difficult question: how can they distinguish a convincing assignment from genuine understanding? The answer could reshape assessment, academic integrity, and the skills students need for their future careers.

A university professor receives two essays on the same subject. Both present strong arguments, cite relevant research, and communicate their conclusions in clear academic English. One student spent several days reading, comparing evidence, and developing an independent position. The other used generative AI to complete much of the intellectual work, then edited the result before submitting it. Although the essays look equally impressive, they may represent very different levels of learning.

This example captures a growing problem in higher education. Generative AI can now help students research topics, organise information, draft arguments, analyse data, and refine academic writing. As these capabilities become more accessible, the finished assignment offers less certainty about the knowledge and reasoning behind it.

Universities must therefore rethink what they measure. AI in university assessment should not be treated only as an academic integrity problem. It is also a question of how institutions establish that students have achieved the intended learning outcomes. A well-designed assessment needs to evaluate the quality of academic work while gathering reliable evidence of the student’s understanding, judgment, and ability to apply knowledge.

For students, this shift carries an equally important message. When almost everyone can access tools that produce sophisticated academic work, the lasting advantage comes from knowing how to question an answer, verify its evidence, explain its reasoning, and use that knowledge independently.

Why AI Is Changing University Assessment

Universities have traditionally relied on essays, research papers, presentations, and examinations to evaluate academic progress. These methods remain useful because they can encourage students to investigate problems, organise ideas, and communicate conclusions. However, generative AI has changed the relationship between the work a student submits and the abilities that work is supposed to demonstrate.

The scale of student adoption makes this challenge difficult to ignore. The Higher Education Policy Institute’s 2026 survey of 1,054 UK undergraduates found that 94% had used generative AI to help with assessed work. Yet only 48% believed their teaching staff were helping them develop the AI skills they would need for future careers.

These figures do not suggest that 94% of students are cheating or submitting AI-written assignments. AI assistance can include legitimate activities such as explaining concepts, improving language, or generating study questions. Nevertheless, the findings show that AI has become part of ordinary academic practice while institutions continue to develop consistent approaches to its use.

The central concern is whether existing assessment methods can still measure the skills they were designed to evaluate. For example, a literature review might demonstrate that a student can produce a coherent summary of research. If AI performs most of the searching, interpretation, and synthesis, however, the final document may provide limited evidence of the student’s independent research ability.

This is where assessment design becomes more important than the technology itself. Universities need to decide which skills students should demonstrate independently, where AI assistance is appropriate, and how educators can distinguish between the two.

The Difference Between a Good Assignment and Genuine Learning

A polished assignment and a well-educated student are not necessarily the same thing. Although the distinction sounds obvious, it becomes particularly important when technology can generate convincing academic work within minutes.

The OECD’s Digital Education Outlook 2026 examines this challenge through the relationship between AI-assisted performance and learning outcomes. Its findings indicate that general-purpose AI can improve the quality of completed tasks without producing equivalent gains in the knowledge students retain or demonstrate independently. The report also identifies opportunities for AI to support learning when educators use it with clear educational goals.

Consider a student preparing an economics assignment. An AI tool may produce a persuasive explanation of inflation, monetary policy, and interest rates. The student might understand the final document well enough to edit its language and structure. However, if the professor introduces an unfamiliar economic scenario, that student may struggle to explain how the same principles apply.

The assignment may be excellent, but the student’s understanding remains uncertain.

Educational researchers describe this challenge through the concept of assessment validity. In simple terms, validity concerns whether an assessment provides suitable evidence for the conclusions educators draw from it. If a university wants to measure independent reasoning, the assessment must provide evidence that the student can reason independently.

This does not mean universities should abandon essays. Academic writing still develops valuable skills, including argumentation, research, and communication. Instead, institutions need to recognise that the finished document may represent only one part of the evidence required to evaluate learning.

What Should Universities Assess When Students Can Use AI?

The future of AI in university assessment should recognise two complementary abilities. Students need independent subject knowledge, and they also need the skills to use AI responsibly in academic and professional settings.

A graduate who understands a subject but cannot evaluate AI-generated information may struggle in an AI-enabled workplace. Conversely, someone who can produce impressive results with AI but cannot explain the underlying concepts may lack the foundation needed to recognise mistakes or make sound decisions.

Universities should therefore assess several connected capabilities rather than relying exclusively on the quality of a final submission.

1. Subject Understanding and Independent Knowledge

Students should be able to explain the central concepts behind their work, connect related ideas, and demonstrate knowledge without depending entirely on AI-generated explanations. This is especially important in disciplines where professional decisions require a strong understanding of foundational principles.

For example, a medical student may use AI to organise study notes or compare clinical concepts. However, the university must still establish whether that student understands the relevant scientific principles. A convincing AI-assisted case report cannot replace evidence of clinical reasoning and subject knowledge.

Universities can evaluate understanding through supervised tasks, practical demonstrations, targeted discussions, and examinations that align with the learning objectives. The purpose is not to prohibit technology throughout the course, but to preserve reliable opportunities for students to demonstrate what they know.

2. Critical Thinking and Decision-Making

Generative AI can produce arguments, compare alternatives, and recommend solutions. As a result, the ability to generate options is becoming less distinctive than the ability to evaluate them.

A business student might ask AI to suggest three market-entry strategies. The meaningful academic challenge is deciding which strategy best fits the available evidence, identifying its limitations, and explaining the risks associated with that choice.

An effective assessment could ask students to justify their recommendations under changing circumstances. This would reveal whether they understand the trade-offs behind a decision rather than simply presenting an attractive solution.

Such tasks also reflect professional environments, where people increasingly use technology to support decisions but remain responsible for the consequences.

3. Research Quality and Evidence Verification

AI can accelerate research, but it can also produce inaccurate statements, incomplete interpretations, and unreliable references. Students therefore need the ability to verify information instead of assuming that a confident explanation is correct.

For researchers, this distinction is especially important. A credible literature review requires more than collecting summaries of published studies. It involves examining research methods, comparing findings, identifying limitations, and explaining where evidence remains uncertain.

Universities could assess these abilities by asking students to investigate a set of AI-generated claims. Students would need to locate the original research, evaluate whether the claims are supported, and explain any corrections.

This approach turns a common AI weakness into an opportunity to assess a valuable academic skill: the ability to distinguish persuasive information from trustworthy evidence.

4. Intellectual Ownership and Academic Integrity

AI has made authorship more complicated because academic work can involve several forms of assistance. A student may develop an original argument and use AI to improve its clarity. Another may allow AI to choose the argument, interpret the evidence, and draft the conclusion.

Although both students have used the same category of technology, their intellectual contributions differ substantially.

Universities need clear policies that distinguish permitted assistance from work students must complete independently. Where AI use is authorised, students should understand what they must disclose and which decisions remain their responsibility.

Assessment can also examine intellectual ownership through research notes, selected drafts, short explanations, and discussions about important decisions. However, these methods should not be treated as perfect proof of authorship, since AI can also generate plausible reflections.

The stronger approach combines transparent expectations with appropriate evidence of student understanding. Academic integrity then becomes a matter of responsible participation in learning rather than an endless attempt to identify which sentences came from a machine.

5. Applying Knowledge to New Problems

One of the clearest demonstrations of understanding is the ability to apply knowledge when circumstances change. A student may successfully explain a theory in an essay but struggle to use it in an unfamiliar situation.

Imagine an environmental science student recommending a renewable energy policy for a wealthy country. The professor then introduces a new case involving limited infrastructure, different economic conditions, and competing public priorities. The student must reconsider the original recommendation and explain which assumptions no longer apply.

This type of assessment examines how students transfer knowledge beyond the specific task they prepared. Depending on the discipline, universities can use case studies, simulations, laboratory activities, practical projects, or supervised problem-solving exercises.

The objective is to evaluate whether students can use their education in situations where an existing answer is insufficient.

How Universities Are Responding to Generative AI

Several institutions have already begun reconsidering their assessment systems. Their approaches offer practical examples, although no single model will suit every university or academic discipline.

University of Melbourne: Separating Learning Activities From Secure Assessment

The University of Melbourne has introduced a multi-year programme focused on assuring student learning. Its approach includes a principle that at least 50% of a subject’s marks should come from secure assessment, with stated exceptions through secure programme-level arrangements.

Secure assessment aims to provide stronger confidence that students completed the work under the relevant rules. It does not necessarily require every task to prohibit AI.

The university also recognises the value of open assessments, where students can develop skills, receive feedback, and use permitted tools. This distinction allows institutions to support AI-enabled learning while preserving reliable ways to verify student achievement.

The broader implication is important: universities do not need to choose between unrestricted AI use and complete prohibition. They can design different assessments for different educational purposes.

King’s College London: Redesigning Assessment Around Authentic Learning

King’s College London has published guidance describing approaches to assessment in the age of AI. These include making existing assessments more resilient, incorporating AI into learning activities, and developing authentic assessments that reflect meaningful academic or professional tasks.

Authentic assessment asks students to apply knowledge in situations that resemble the problems they may encounter beyond the classroom. For example, an engineering student might evaluate competing designs under realistic constraints, while a public policy student might develop and defend a recommendation for a particular community.

These approaches do not eliminate the possibility of AI assistance. Instead, they place greater emphasis on the student’s decisions, explanations, and application of knowledge.

What the Research Suggests

A 2026 systematic review of 72 studies identified three prominent directions in higher education assessment: greater attention to the learning process, authentic tasks, and opportunities for students to explain their reasoning through discussion.

These approaches are promising because they gather evidence beyond the final document. However, the review also indicates that much of the existing research remains exploratory. Universities still need stronger evidence about long-term outcomes, fairness, workload, and effectiveness across different disciplines.

The responsible conclusion is therefore not that one assessment method has solved the AI challenge. Rather, universities are developing combinations of methods that may provide a more reliable picture of student competence.

A Practical Framework for AI-Era University Assessment

Universities can begin by separating three questions that traditional assignments sometimes combine: How good is the submitted work? What intellectual decisions did the student make? Can the student apply the underlying knowledge independently?

We propose a three-layer framework for thinking about these questions. It is an editorial model intended to support assessment design, not a scientifically validated grading system.

Layer One: Evaluate the Quality of the Work

The first layer examines whether the submission meets the academic standards of the discipline. Educators assess the argument, research quality, accuracy, communication, and suitability of the proposed solution.

This remains essential because universities should continue teaching students to produce useful, rigorous work. However, when AI assistance is permitted, the quality of the output should not automatically be interpreted as evidence that every underlying skill was demonstrated independently.

Layer Two: Examine the Student’s Intellectual Decisions

The second layer considers how the student approached the problem. Educators might ask why particular sources were selected, which alternatives were rejected, how conflicting evidence was resolved, and where AI influenced the process.

A short decision log or annotated bibliography can make these choices easier to examine. For major projects, selected drafts or a brief discussion may provide additional context.

The purpose is to understand the student’s reasoning without turning academic work into excessive documentation. Process evidence should support assessment, not become an administrative burden that distracts students from learning.

Layer Three: Test Knowledge Transfer

The third layer evaluates whether students can use their understanding in a new situation. An educator might introduce different assumptions, provide additional evidence, or ask students to respond to an unfamiliar case.

For example, a student could prepare an AI-assisted report on the use of automated recruitment systems. The professor might then introduce a scenario involving new legal restrictions or evidence of discriminatory outcomes.

A student who understands the original analysis should be able to reconsider the recommendation and explain the changes. This provides stronger evidence of flexible reasoning than simply repeating the submitted report.

Together, these three layers help universities distinguish the quality of a document from the competence demonstrated by the person who submitted it. Institutions can distribute them across a course rather than applying every layer to every assignment.

Why AI Detection Alone Cannot Solve Academic Integrity

AI detection tools have become part of the academic integrity debate because institutions want reliable ways to identify unauthorised assistance. However, identifying patterns associated with AI-generated text is different from establishing that a student violated a particular rule.

Research has found substantial variation in detector performance, particularly when tools encounter edited, mixed-authorship, or unfamiliar text. False positives and false negatives also create risks when institutions use detector scores in high-stakes decisions.

A fair investigation therefore requires more than an automated percentage. Universities need clear policies, appropriate supporting evidence, human review, and opportunities for students to explain their work.

More importantly, even an accurate detector would not solve the entire educational problem. It might help investigate certain forms of unauthorised AI use, but it would not establish whether permitted AI assistance improved a student’s understanding.

Academic integrity and learning assessment are connected challenges, yet they require different solutions.

The Practical Challenges Universities Cannot Ignore

Assessment reform introduces difficult questions about workload, accessibility, and fairness. A professor responsible for several hundred students may not have the time to conduct individual oral examinations. Similarly, extensive process documentation can create additional marking responsibilities without necessarily improving the quality of learning.

Universities must also consider students who communicate differently, study in a second language, or require disability accommodations. Oral assessment may provide useful evidence of understanding, but it should not introduce unnecessary barriers when spoken fluency is unrelated to the skill being measured.

Unequal access to AI technology creates another concern. Students may have different subscriptions, digital skills, and levels of institutional support. If universities expect students to demonstrate AI literacy, they should provide clear guidance and reasonable opportunities to develop those skills.

A sustainable assessment strategy must therefore balance educational quality with practical implementation. Short discussions, selected supervised tasks, consistent rubrics, and programme-level assessment planning may help institutions gather stronger evidence without redesigning every assignment.

The goal should be a fairer and more meaningful evaluation of learning, rather than a more complicated process that simply increases pressure on students and educators.

What Students Should Focus on in an AI-Enabled University

As universities reconsider assessment, students should also reconsider how they use AI. The most valuable study habits are those that improve understanding and judgment rather than merely reducing the time required to finish an assignment.

Build Understanding Before Improving the Writing

Students often approach AI when they feel overwhelmed by a difficult topic or an approaching deadline. Although the temptation to request a complete answer is understandable, a more productive approach begins with identifying what they do not understand.

For example, a student struggling with a research paper could first ask AI to explain the central theory, compare different interpretations, and clarify unfamiliar terminology. After reviewing those explanations, the student should attempt to describe the argument independently.

This approach makes AI a source of assistance during learning instead of allowing it to replace the learning process.

Practise Defending Your Arguments

Before submitting an assignment, students should examine whether they can explain the reasoning behind their conclusions. They can begin by identifying the strongest evidence supporting their argument, considering alternative explanations, and recognising important limitations.

A useful exercise is to imagine that a professor disagrees with the conclusion. What evidence would justify the student’s position, and what new information might cause them to reconsider it?

Students who practise these questions develop the ability to defend their thinking in seminars, presentations, research discussions, and professional settings.

Verify Research Instead of Trusting Fluent Answers

AI-generated summaries can make complex research easier to navigate, but students should still examine the original sources. This includes checking whether a cited study exists, whether the findings support the stated claim, and whether important limitations have been omitted.

For postgraduate students and researchers, source verification is particularly important because errors can influence research questions, methodologies, and subsequent conclusions.

The ability to evaluate evidence is likely to remain valuable even as AI becomes better at finding and organising information.

Combine AI-Assisted Study With Independent Practice

AI can support revision through explanations, flashcards, practice questions, and summaries. However, students also need opportunities to retrieve information and solve problems without assistance.

After studying a difficult concept, a student might use AI to generate practice questions and then attempt the answers independently. Reviewing mistakes afterward can help identify gaps in understanding.

This creates a healthier relationship with the technology. Students benefit from its convenience while continuing to develop the knowledge and confidence required for independent work.

Develop Responsible AI Literacy

Knowing how to write effective prompts is useful, but AI literacy involves more than obtaining better outputs. Students should understand the limitations of the tools they use, recognise potential errors, protect sensitive information, and follow their institution’s academic integrity policies.

They should also learn to decide when AI assistance is appropriate. Some tasks benefit from collaborative exploration, while others require independent practice because the learning process itself is the objective.

The strongest students will not necessarily be those who use AI most frequently. They will be those who understand how to use it purposefully and responsibly.

Why This Matters for International Students and Researchers

The discussion becomes more complex when students work across different languages and educational backgrounds. A student may have strong subject knowledge but struggle to express it in formal academic English. Another may write fluently while demonstrating weaker reasoning.

Universities should distinguish language proficiency from subject understanding when designing assessments. Where communication skills are an explicit learning outcome, they deserve appropriate evaluation. However, when the objective is to assess scientific knowledge or analytical reasoning, unnecessary language barriers can make the assessment less accurate.

AI-assisted language support may help international students express their own ideas more clearly, provided their institution permits that assistance. The student must still understand the argument, verify the evidence, and take responsibility for the submission.

For researchers, the challenge is slightly different. AI can help organise extensive literature and identify possible connections between studies, but research quality depends on careful interpretation. Understanding methodological limitations, recognising conflicting findings, and identifying unanswered questions remain central responsibilities.

In both cases, educational AI offers meaningful opportunities when it supports the student’s existing intellectual abilities rather than concealing gaps in understanding.

What the Future of Assessment Means for AI Education Technology

The debate about university assessment should influence how educational AI companies define the value of their products. Faster writing, convenient summaries, and automated research assistance can improve productivity, but productivity alone does not establish educational benefit.

A tool may help a student finish an assignment while contributing little to the knowledge they retain. Another tool may encourage the student to explore a difficult concept, test their understanding, and improve an argument through repeated reflection.

Both experiences involve AI, yet their educational implications differ.

This distinction creates an important responsibility for the edtech industry. Developers should consider whether their products encourage students to engage with information, examine uncertainty, verify sources, and practise independent thinking.

For example, flashcards become more useful when students actively retrieve information instead of repeatedly reading answers. Mind maps can help students explore relationships between concepts, while conversational explanations can support understanding when learners ask follow-up questions.

However, the presence of these features does not automatically prove that a product improves learning outcomes. Educational value depends on product design, the quality of information, and how students use the tools.

The next stage of AI education technology should therefore focus on supporting the learning process alongside academic productivity.

Conch AI’s Perspective: Better Tools Should Help Build Better Thinkers

At Conch AI, we believe academic technology should help students develop their knowledge while making studying and writing more manageable. Our perspective is that AI assistance works best when students remain actively involved in understanding, questioning, and communicating their ideas.

Conch AI brings together academic writing support and study tools that help learners engage with information in different ways. Smart chat can help explain difficult material, while flashcards and mind maps support revision and the organisation of ideas. Audio-based study formats offer another way to explore learning material, and multilingual capabilities can help students approach complex subjects in a familiar language.

These tools can support different stages of studying, but students remain responsible for their research, reasoning, and academic submissions. They should verify sources, practise independently, and follow the AI-use rules established by their institutions.

As university assessment evolves, our position remains straightforward: educational AI should help students become more capable of understanding and expressing their own ideas, not simply more efficient at producing finished documents.

The Future of University Assessment Is About Demonstrating Competence

Generative AI has changed how easily students can produce sophisticated academic work, but it has not changed the fundamental purpose of higher education. Universities still need to develop knowledgeable graduates who can evaluate evidence, make informed decisions, communicate clearly, and apply what they have learned.

What must evolve is the evidence institutions use to determine whether those abilities have developed.

Essays and research papers will continue to play an important role, although universities may increasingly combine them with demonstrations of reasoning, supervised activities, practical applications, and transparent accounts of AI assistance.

For students, the implication is equally clear. Academic success should involve more than submitting a polished assignment. It should reflect knowledge they can explain, arguments they can defend, evidence they can evaluate, and skills they can apply when circumstances change.

The future of AI in university assessment is therefore not simply about deciding whether students should use AI. It is about creating an educational system that recognises both independent competence and responsible AI-enabled work.

When producing an answer becomes easier, understanding why that answer is right becomes even more valuable.

Study Smarter. Write Better. Submit With Confidence.

Conch AI supports students and researchers with academic writing assistance, smart chat, flashcards, mind maps, multilingual learning, and audio-based study tools. Explore different ways to understand complex subjects, organise research, and communicate your ideas with greater clarity.

Use AI as a study partner while keeping your knowledge, judgment, and academic responsibility at the centre of your work.

References

Higher Education Policy Institute. (2026). Student Generative Artificial Intelligence Survey 2026. https://www.hepi.ac.uk/reports/student-generative-ai-survey-2026/
OECD. (2026). OECD Digital Education Outlook 2026. https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026_062a7394-en.html
Higher education assessments in the age of GenAI: A systematic review. (2026). International Journal of Educational Technology in Higher Education. https://link.springer.com/article/10.1186/s41239-026-00627-4
University of Melbourne. Assuring Learning at Melbourne. https://about.unimelb.edu.au/education-melbourne/assuring-learning-at-melbourne
King’s College London. Approaches to assessment in the age of AI. https://www.kcl.ac.uk/about/strategy/learning-and-teaching/ai-guidance/approaches-to-assessment
EDUCAUSE. (2026). The Impact of AI on Learning Assessment. https://library.educause.edu/resources/2026/6/2026-educause-the-impact-of-ai-on-learning-assessment-report
OECD. (2026). Policies Supporting Responsible and Systematic GenAI Adoption in Higher Education. https://www.oecd.org/en/publications/policies-supporting-responsible-and-systematic-genai-adoption-in-higher-education_c4e5621f-en.html

AI in University Assessment: If AI Can Write the Essay, What Should Universities Assess?