Process-Oriented Collaborative Essays: Empowering Undergraduate Students Through Assessment Redesign in the Age of Generative AI

Synthesis: Higher-education responses to generative AI have mostly been detection or prohibition, and both keep the finished essay as the unit of evidence. Alam and Jackson describe a third move: redesign the assessment so the process itself carries the evidence. In a first-year Business Analyst degree apprenticeship module at Queen Mary University of London, 23 students worked in 11 self-selecting teams on an iterated 1,500-word essay, with GenAI permitted inside stated boundaries and iteration and collaboration written into the grading criteria. The authors read the first run through Task-Technology Fit and Institutional Theory and present it as practice-based insight rather than measured outcomes, finding that GenAI worked as something for students to interrogate where the task rewarded judgment, while the assessment's risks — uneven participation, documentation load, performative evidence — stayed real.

Key Findings

1. 23 first-year Business Analyst degree apprenticeship students at Queen Mary University of London worked in 11 self-selecting teams, one of three members and the rest of two, on a 1,500-word essay restricted to three pre-set titles and requiring at least two documented stages of iterative development.
2. The assessment criteria made collaboration and iteration explicit contributors to the final grade, so evidence of process was structurally required rather than merely invited — the design claim that permitting GenAI is not the same as designing for its thoughtful use.
3. GenAI was permitted but not mandatory. The brief barred submitting generated text as one's own, discouraged rewriting functions in grammar tools, and warned that AI-formatted references frequently do not exist, which would undermine credibility and might count as misconduct.
4. The paper is explicitly practice-based and exploratory: the authors describe a program-wide pedagogical redesign rather than a controlled intervention, and present emerging insights instead of generalizable findings, with no measured learning outcomes and no comparison condition.
5. Where students used GenAI, some interrogated its output — contesting counterarguments, revising, rejecting — which the authors read as dialogic assistance that stimulates intellectual work rather than offloading it.
6. Under Task-Technology Fit, the tool's fit is strong for producing counterarguments, alternative structures, and language fixes, and weaker where an assessment rewards the student's own judgment, appraisal of evidence, professional experience, or negotiation with peers.
7. The 11 teams collaborated in visibly different ways — sharing drafts, exchanging peer feedback, revising iteratively, and in some cases explaining decisions over Microsoft Teams — and the authors conclude that collaboration quality needs Scaffolding in class before students can deploy it autonomously.

Why the final essay stopped being sufficient evidence

The paper opens by shifting the question. Acceptance is settled: a 2026 Higher Education Policy Institute survey found 94 percent of students surveyed had used GenAI in some way for assessed work. The live problem is therefore not whether students adopt the tools but what an assessment can still be evidence of when the tools can perform parts of the task — generating ideas, proposing structures, summarizing, refining arguments, drafting text.

The authors' target is narrower than the general complaint about detection. Their argument is that a final essay is a proxy for intellectual processes it does not record: how ideas were built, how Feedback was absorbed, how arguments changed. That gap can be worked around with supervising conditions or examinations, but the authors point to work suggesting an unassisted final product can remain an indicator of achievement while being complemented by evidence of how the student got there.

What the redesign actually required

The intervention is a structure, not a policy. Students wrote in teams, chose from three module-related titles, and had to document at least two stages of development. The marking criteria spelled out how evidence of iteration and collaboration would move the grade, which converted process from a suggestion into a requirement.

GenAI sat inside that structure with conditions: no submitted generated text, no rewriting functions, and no AI-formatted reference lists because fabricated references would damage credibility and might read as misconduct. Because the documentation was graded, the authors argue, students had reason to show their reasoning rather than their output — the assessment criteria demonstrate that a grade, not a reminder, is what makes process evidence load-bearing.

Two frameworks organize the reading. Task-Technology Fit asks how well the tool's capabilities match specific writing activities rather than the essay as one undifferentiated task, so a strong fit for generating counterarguments coexists with a weak fit for demonstrating judgment. Institutional Theory supplies the second half: what counts as legitimate evidence of learning is set by institutional expectations, authorship norms, and integrity rules as much as by pedagogy, so an assessment design has to be defensible to the institution and not only effective in the classroom.

What the first run suggested, and what it could not show

The reflections come from the first delivery, in early 2026. The observation the authors single out is that some students who used GenAI challenged what it produced — following an AI counterargument with a revision, an objection, or a rejection. They treat that as the difference between a tool that replaces thought and one that provokes it, and note the implication for design: the value does not come from permitting the tool but from requiring students to account for what they did with it.

Collaboration behaved the same way — present in every team, but uneven in quality. Sharing drafts and exchanging feedback appeared across the 11 teams, with some groups using Teams calls and chat to explain their reasoning. The authors are careful that group work is not by itself evidence of equitable learning, citing research on socially shared regulation in which stronger collaboration involved negotiated feedback and collective decision-making. Their conclusion is a design note rather than a result: students need explicit experience of different collaborative modes before they can use them independently.

What this means for practice

• Instructors. Make process evidence carry marks rather than goodwill. In this design the criteria, not the brief's tone, are what required students to document iteration, so a reader who wants this behavior should put it in the rubric — otherwise the documentation becomes optional work that only diligent students do.
• Instructors. Write explicit boundaries for GenAI and give the reason behind each one. The paper's brief is short and specific: no generated text submitted as your own, no rewriting functions, and no AI-written reference lists — the last aimed at fabricated citations rather than at cheating.
• Assessment designers. Split the task before judging the tool. Task-Technology Fit suggests asking which activities in the assessment the AI genuinely supports and which ones the assessment exists to observe; the essay is not one task, and permission granted at the level of the essay hides that difference.
• Faculty developers. Teach collaboration separately from assessing it. The authors' own reading is that students arrived with uneven collaborative repertoires and needed scaffolded practice in class before group process could serve as evidence.
• Administrators. Expect the design to be argued institutionally, not only pedagogically. The paper frames validity as a matter of what the institution will accept as legitimate evidence of authorship and learning.

Limitations

• The study is a practice-based exploration of one module in its first run, with 23 students on a single degree apprenticeship program. The authors state it offers emerging insights rather than generalizable findings, and it reports no measured learning outcomes, no comparison condition, and no analysis of whether the redesign changed what students learned.
• Process documentation carried its own cost. Some students described the required process evidence as lengthy and time-consuming, and some were initially unsure how to write a meaningful commentary on their own working process.
• Analyzing process evidence invites performativity. The authors warn that once collaboration, revision, and reflection are graded, students may produce evidence matching what they believe the marker expects, so reflections and collaboration records should not automatically be treated as proof of learning.
• GenAI use was optional and unevenly taken up, and participation differed across groups, so the reported benefits rest on the authors' reading of the submissions they saw rather than on a measured contrast between AI-using and non-using teams.

Citation

Alam, F., & Jackson, J. (2026). Process-Oriented Collaborative Essays: Empowering Undergraduate Students Through Assessment Redesign in the Age of Generative AI. The European Conference on Education 2026: Official Conference Proceedings, 989–1001. IAFOR.