AI won’t get you a pass. Engagement will.

Categorised: Teaching tips | Tags: AI
Posted by Claire Fenton. Last updated: July 15, 2026

Over six years, through a consistent assessment administered to cohorts of 500 to 800 students, an unintended experiment took shape. I didn’t set out to create an AI study. I was designing a realistic assessment of workplace communication and judgement. AI simply arrived mid-experiment, like a colleague who joins a meeting halfway through and immediately dominates the narrative.

This is a reflection on that journey, beginning with introducing AI into the assessment as soon as it became widely available. By analysing the performance data, I considered how AI was influencing students’ approaches to learning, their engagement with the assessment, and the ways in which they were working. These insights informed changes I made to teaching and assessment.

The assessment: “Please act like an adult with an inbox”

Back in 2020, I introduced a work-simulated task assessment into a large Management Accounting module. My aim was to move away from the traditional academic essay and instead recreate something closer to early professional practice. Students were placed into the role of junior graduate accountants in a fictional organisation. They received a time-pressured email from a line manager and were asked to respond as they would in a real working environment. The idea was simple. Make it feel real enough that students must behave like professionals, not just students writing for a marker.

The assessment covers technical accounting knowledge, but also requires judgement, prioritisation, clarity and communication under pressure. The kind of decision-making that happens in employment when you can’t just bury your head in a textbook and cite other people’s opinions.

The core assessment follows a consistent structure that has remained largely unchanged for six years.

1. The scenario arrives

Students are introduced to a fictional organisation and their role as graduate accountants. Considerable care is taken to make the organisation feel real, with enough contextual detail to signal workplace culture, hierarchy, and expectations.

This matters more than it first appears. Students do not just interpret numbers. They interpret what is going on, what will be required, the culture, and embed it into their responses. The design intentionally mirrors the ambiguity of early professional roles where nothing is fully explained, and everything is slightly urgent.

2. The email lands

Students then receive an email from their line manager. This email contains multiple requests embedded within a single communication. There are financial appendices, operational details, and layered instructions that require interpretation and prioritisation. One element of the email always introduces a tension. It might be unrealistic targets, ethically questionable reporting expectations, or subtle pressure to present data in a favourable way.

In addition to completing routine tasks, students are asked to decide whether the task itself is appropriate. This is where professional scepticism becomes central. The stronger students pause to interpret intent rather than just follow the instructions.

3. Time constraints and submission

The assessment is completed within a strict 80-minute window and submitted through Turnitin (plagiarism-detecting software). There are no extensions and no flexibility. This is not about being harsh. It is an attempt to reflect reality: time pressure is part of the design. It forces prioritisation, sometimes at the price of perfection.

4. Output format: a professional email

Students respond in the form of a professional email. This includes subject line, greeting, structured content, and sign off. Students are expected to explain the numbers, adapt to the audience, and translate analysis into actionable recommendations. Word counts and citations are not relevant, as with most emails, but clarity, judgement, and usefulness are, and this is reflected in the marking rubric.

5. Rubric design: building a mark scheme for a changing world

One of the most unexpected aspects of this six-year journey is how little the marking rubric’s structure has changed. While technology, student behaviour, and institutional discussions around AI have evolved rapidly, the rubric itself has remained remarkably resilient.

It has two core components. The first is task-specific performance. This assesses how well students complete the three embedded tasks using appropriate academic accounting knowledge and reasoning. The second is a generic skills component, divided into completion with added value and professional communication.

Completion with added value

Before generative AI, weaker students often left sections incomplete or underdeveloped. There were clear gaps, missing explanations, or partial answers. AI has largely solved this issue. Completion rates have improved significantly. Students now almost always produce a response for every section. However, completion is not the same as adding value. I often ask students why a company would hire them instead of simply using AI. If their answer is indistinguishable from AI’s, they should ask themselves whether they would have been hired at all.

If their answer is indistinguishable from AI’s, they should ask themselves whether they would have been hired at all.

A new pattern has emerged in AI-assisted submissions. Work is complete, polished, and structurally sound, but often lacks judgement. It reads like a summary of a summary of a summary, each layer slightly more confident than the last, but further removed from original thinking. The recurring marker comment becomes: “This is well written, but what have you actually added other than larger words?”

As a result, this criterion has become more focused over time. Students are rewarded for evidence of revision and critical reflection. A colleague described this element as “a quiet but effective measure of engagement”. With the broader generational shifts in how students approach university learning, I found this a reassuring observation.

Professional communication

The assessment is deliberately based around business emails because email remains a dominant communication channel in UK workplaces and is widely used as a primary method of organisational communication (IoIC IC Index, 2024). It also reflects a clear educational need — I consistently received student emails that lack basic conventions of professional communication, including missing or incorrect greetings, the use of text speak, ALL CAPS, and messages so brief and abrupt that they felt more like a ransom note than professional correspondence.

Whatever the reasons, many students arrive at university with little experience of professional written communication. I therefore give students explicit guidance on writing effective business emails, and, where AI is used, on critically reviewing its output rather than accepting it at face value. They are encouraged to be concise, polite, and mindful that small errors can carry disproportionate weight. Misnaming a recipient, for example, can undermine a message before it is even read.

Whether emails are drafted independently or with the support of AI is largely irrelevant. What matters is the quality of the final communication. As a result, “professional communication” has remained a stable principle in the rubric. However, the types of errors have changed over time. For example, with the increasing use of AI, spelling and formatting errors have reduced whereas reviewing for “AI waffle” has become increasingly important.

Consequently, I have recently introduced a simple quality-control check in the form of a hidden instruction in white text within the PDF, typically requesting a sentence on something deliberately irrelevant, such as a comment about rabbits. This is not intended as a trick or a source of amusement for markers. Instead, students are made aware of its existence during the mock assessment and instructed to remove it if they encounter it. It acts as a simple proxy for key professional habits: engaging fully with instructions and carefully reviewing work, especially when it is AI-generated.

The AI timeline: six years of gradual disruption

2020 to 2022: stability before the noise

For the first two years, performance was stable. Average marks sat between 62 and 65 per cent. The distribution followed a predictable bell curve. Students understood the task. Variation existed, but the overall pattern was consistent. At this point, the only real challenge was improving clarity of written communication.

November 2022: the audit question

ChatGPT had only just arrived when the question landed in my inbox ahead of the January 2023 assessment: do I allow students to use AI? Having previously worked as an EY auditor, I defaulted to a familiar lens: controls. Could we prevent it? Could we detect it? The uncomfortable answer to both was no.

This is an audit perspective, and it leads to a fairly blunt principle: if something cannot be reliably prevented or detected, assume it exists in the system. Thus, I said yes. In other words, we would redesign the rubric, if necessary, around the reality of AI, not the illusion of its absence. The question then became simpler, and slightly more unsettling: what happens next and what do we do about it?

January 2023: AI arrives… nothing happens

The marking team entered January 2023 braced for disruption but nothing much happened. Average marks stayed around 65, and the distribution barely shifted. ChatGPT had entered student awareness, but not yet student habit.

Most students ignored it entirely. A small number experimented, often clumsily. Weaker submissions sometimes pasted AI-generated text directly, usually identifiable by its generic tone, time lag and confident-but-generic statements.

January 2024: “Why does everything sound like the same confident, surface-level, slightly unethical robot?!”

By 2024, the situation changed dramatically. Average marks dropped to around 51 per cent. More importantly, the character of student work changed. Responses became more uniform. Language converged and structure became repetitive.

Students were less likely to identify ethical issues or prioritise correctly. The instinct to question appeared weaker. A particularly noticeable change was structural fragmentation. Instead of one coherent email, submissions often arrived as multiple disconnected responses, suggesting that AI outputs had been copied without integration or review. The most concerning shift was a decline in judgement.

January 2025: structural changes in teaching

We are making a step. It’s just a baby step, but it’s a step.
Miss Maudie in Harper Lee’s “To Kill a Mocking Bird”

In response, teaching practices evolved. Two workshops were introduced. One focused on identifying where AI performs poorly, particularly in ethical reasoning. The other was a mock assessment allowing students to choose whether or not to use AI, helping them develop critical awareness of it. Students were encouraged to test AI outputs rather than accept them. This led to a partial recovery in performance, with average marks rising to 56 per cent.

However, engagement began to decline. Attendance dropped. Participation weakened. A paradox emerged. More structured support did not necessarily lead to more engagement.

January 2026: “Integrating AI into pedagogy — but is anyone still listening?”

By 2026 I embedded AI as a standard tool within workshops rather than a special topic. I designed five learning themes within the workshops, including “recognising human superiority and empowering confidence”, “adapting communication to audience while revising content”, “applying theory to real-world contexts”, “identifying AI limitations”, and simply “practising in a safe environment”.

Disappointingly, marks rose only slightly to 58 per cent. However, engagement continued to decline. The pattern was now clear. Performance stabilised, but participation weakened.

Six years on: the AI story I share with students

My story begins with my time at EY, where the work was challenging, but engaging as it involved problem solving, client interaction, and the interpretation of messy real-world systems. Even tasks like stock counts, which might appear dull, became surprisingly interesting when you were physically observing what a client produces and the processes involved. The challenge, however, came after the investigative work was done, when findings had to be translated into structured, polished reports, often late at night, long after my brain had signed out for the day.

This is where AI becomes helpful, or for me at least. The message is not to stop thinking, but instead to use these tools to make the workload manageable without removing yourself from the intellectual process.

I emphasise that AI does not decide what matters. It does not define priorities. And when it is confidently wrong, it does not correct itself. Those responsibilities remain firmly with the humans. The moment that changes, it is no longer a tool in any meaningful sense, but something closer to a liability that simply happens to be well formatted.

Finally, I transparently discuss with students what I have observed over time: the increased support and tools available, but ever-decreasing engagement and slow return to performance norms. My message is simple: tools will keep evolving and behaviours will adapt, but engagement still does the heavy lifting. No model, however sophisticated, replaces showing up, thinking actively, and working with others. That, ultimately, is the challenge I intend to focus on in the coming years.

Claire Fenton

Claire Fenton is an Associate Professor at Royal Holloway, University of London, qualified as a Chartered Accountant (ICAS), and is a Senior Fellow of the Higher Education Academy (SFHEA). After training with EY, she discovered that her favourite part of the profession was helping others learn, inspiring a move into higher education that she has never looked back from.

A full-time working mum of two, Claire is passionate about making accounting engaging, accessible, and relevant to students’ future careers. Her teaching and scholarship focus on assessment and classroom innovation, AI in accounting education, and student engagement. Alongside her teaching, she contributes to programme development with a commitment to creating learning experiences that are academically rigorous, supportive, and most of all, enjoyable.

How to cite this article: Fenton, C. (2026) ‘AI won’t get you a pass. Engagement will.’, Accounting Cafe, [insert publication date]. Available at: PLACEHOLDER_ARTICLE_URL (Accessed: [insert date])