What happens when students are expected to use generative AI as part of an assessment, rather than being warned away from it?
Kardi Somerfield and Billy Little explored this question by redesigning a first-year research assignment. Their aim was not simply to permit AI use. They wanted students to learn how to use it thoughtfully within a demanding, professionally relevant task.
The early results suggest that this approach helped students overcome some familiar barriers to academic work. Submission rates improved and fewer students needed extensions. The weakest marks also rose, while the strongest students continued to distinguish themselves through the quality of their thinking.
This was an initial experiment rather than a finished model. Even so, it offers useful insights into what AI literacy can look like when it is embedded within assessment design.
When research becomes a word-count exercise
The original assignment asked students to complete a 2,500-word piece of desk research. It was positioned early in the term because the findings fed directly into a later creative pitch.
The timing gave the assignment an important role, but it also created pressure. Students needed to undertake substantial research before they had fully settled into the module. For some, the task became less about curiosity and more about reaching the required word count.
This risked working against the purpose of the assignment. The research was supposed to help students become interested in a problem before developing their pitch. Instead, the scale of the task could make the process feel like an obstacle to be completed.
Kardi and Billy saw an opportunity to use generative AI to change the nature of that experience. AI could reduce the friction involved in beginning an unfamiliar research task and help students move more quickly from an initial question towards something they wanted to explore.
The intention was not to make the assignment easier. It was to redirect students’ effort from producing enough words towards making sense of what they discovered.
More than permission to use AI
Allowing students to use AI does not automatically create an AI-literate assessment. If expectations remain vague, students may be unsure what responsible use looks like or how their use of the technology will affect their grade.
In the redesigned assignment, AI was made an explicit part of the assessment. The rubric included a dedicated criterion addressing transparent and responsible use, alongside research integrity. Students therefore knew that their decisions about AI mattered and that they would need to account for them.
The research task was also complex enough to make AI genuinely useful. Students needed to examine several organisations, understand the relationships between them and consider the strategic implications. This required synthesis rather than a series of simple searches.
Students were then asked to reflect critically on the process. They needed to explain how AI had supported their research and identify where its output required verification. They also had to consider its limitations.
This shifted the emphasis away from whether AI had been used and towards how well it had been used. The tool was part of the research process, but the student remained responsible for judging the quality and relevance of the material.
Designing for professional practice
The assessment was also intended to reflect changes within the creative industries. Channel 4 confirmed that its own teams now use AI when researching a potential client or responding to a new brief. Large agencies are also developing their own agentic AI tools to support this kind of work.
This industry perspective was important. The redesigned assignment was not lowering academic expectations to accommodate a new technology. It was helping students develop the judgement they would need in professional settings where AI-assisted research is already becoming routine.
Authenticity, however, involved more than reproducing workplace practices. Students still needed to demonstrate that they could interrogate the material and shape it into an informed response. The technology could accelerate the research process, but it could not replace the decisions that gave the work its direction.
Getting past the blank page
One of the clearest effects was that students found it easier to begin.
The blank page can be a significant barrier, particularly when an assignment arrives early in a course and asks students to navigate an unfamiliar field. In this case, students referred directly to AI’s role in helping them get started.
Submission rates were high and only one student requested an extension. This was a marked improvement on the pattern associated with the earlier version of the task.
AI also appeared to support comprehension. Stronger students used it to build an initial understanding of complicated relationships before developing their own analysis. It helped them reach the point at which more meaningful thinking could begin.
As the teaching team observed, these students were able to turn the initial output into something genuinely their own through careful editing. They could also explain why they had used AI and where they had challenged it.
This suggests that using AI for sense-making is not necessarily the same as bypassing learning. When the assessment requires students to evaluate what the tool produces, AI may provide a scaffold into complex material rather than a substitute for engaging with it.
A different relationship with the word count
The redesigned task also changed how some students experienced the word limit.
One student who usually struggled to write enough found that they had too much material and needed to make decisions about what to remove. Students who tended to overwrite reported that they were editing more effectively.
This is a subtle but significant change. Reaching the word count was no longer the main intellectual challenge. Students had to select what mattered and decide how it contributed to their response.
AI had made it easier to generate a starting body of material, but that created a new demand. Students needed to exercise judgement. The quality of the work depended on what they retained, how they connected it and whether they could make it their own.
Did AI make students less critical?
A common concern about AI-integrated assessment is that students will become less critical because the technology is doing too much of the work.
The grade profile from this first experiment does not support a simple version of that argument. The lowest mark increased from 35 to 48. At the top of the range, the highest mark was 82 compared with 85 previously.
In other words, the floor rose while the ceiling remained broadly stable.
This needs to be interpreted cautiously. One assessment cannot establish that AI caused the change. The pattern is nevertheless encouraging. Students who might previously have struggled to complete the task appeared to benefit from the additional scaffold, while the strongest students were still differentiated by the sophistication of their thinking.
If AI had simply completed the intellectual work for everyone, the marks might have become compressed at the top. That did not happen.
The assignments were graded anonymously. Once the students’ identities were revealed, their performance was broadly consistent with what the teaching team knew of their previous work. The assessment still appeared able to distinguish between different levels of understanding.
Teaching students to work with AI
The case study suggests that the most useful question is not whether students should use generative AI, but what an assessment asks them to do with it.
Simply adding an AI statement to an existing brief is unlikely to develop AI literacy. Students need a task in which the technology has a legitimate purpose. They also need to know that evaluating its contribution is part of the learning.
Making AI use visible within the rubric helped establish that expectation. Requiring critical reflection made students’ decision-making available for assessment. The professional context gave the activity a reason beyond compliance with university guidance.
The most promising finding is that AI may help more students gain entry to a demanding task without removing the opportunity for stronger students to excel. It can reduce the difficulty of getting started while leaving the harder work of interpretation with the learner.
That is the distinction at the heart of this case study. AI did not remove the need for critical thinking. The assessment was designed to make critical thinking about AI part of the work.
