AI Makes Student Marketing Projects Look Brilliant, How Can Teachers Judge Learning?

A study of digital marketing coursework explores what teachers can actually judge when appearances reveal so little.

A convincing marketing campaign can arrive with polished graphics, confident forecasts and a persuasive sales pitch, leaving teachers with little sense of the thinking behind it. Generative AI makes that gap harder to ignore because students can produce professional-looking work without explaining their decisions. A study of digital marketing coursework explores what teachers can actually judge when appearances reveal so little.

Researchers Antonia Stefani and Eleni Vamvatsikou examine this problem in "Quality of Authentic Assessment in Digital Marketing Education in the GenAI Era," published in the journal Information. Their research follows three versions of an undergraduate assignment at the University of Patras in Greece, covering 2024, 2025 and 2026. They analysed the assignment instructions and 65 submissions: 19 from the first year, 29 from the second and 17 from the third.

The central concern was whether student work made professional judgement visible, including the evidence behind a decision and the contribution made by AI. The researchers examined how changing the assignment exposed some weaknesses and created fresh questions about others. Their findings offer practical ideas for assessment, without claiming that more detailed instructions improved students' learning.

Real Websites Could Not Deliver Meaningful Campaign Results

The 2024 assignment asked students to promote an existing company's product, describe its intended customer, create marketing materials, research keywords and build a test website. Students were expected to connect analytics tools and evaluate the campaign using the resulting data. Limited audiences, budgets and operating time made that final requirement difficult to fulfil.

Fourteen of the 19 submissions included a website or preview link, and 13 mentioned an analytics or measurement tool. None established a defensible connection between a defined campaign period, real traffic, reproducible measurements and a justified marketing decision. Some students openly reported missing data, others described technical setup without analysis, and three claimed positive results without adequate evidence.

The tools themselves sometimes pulled students away from useful thinking, as a suggested business-buyer template produced unsuitable customer profiles for consumer products. Keyword research ranged from careful testing and revision to broad labels with little supporting detail, differences that the grading scheme did not make explicit. AI image generation already appeared in submissions before the assignment introduced formal guidance on disclosure.

Imaginary Products Made Strategy Possible and Evidence More Important

The 2025 redesign removed live implementation and asked students to invent a digital product or service with innovation and sustainability built into it. Their work included branding, a social post, a storytelling video, search optimisation and five justified key performance indicators, or KPIs. This approach gave students a professional planning role without requiring an audience or advertising budget they did not have.

Twenty-eight of the 29 submissions included numerical targets, making students' expectations easier to examine. Many targets lacked historical data, external comparisons or a calculation explaining why the number was reasonable, and examples supplied in the instructions repeatedly resurfaced in answers. A forecast could look precise without giving a teacher enough evidence to judge its credibility.

Sustainability was central or substantial in 26 submissions, with ideas clustering around circular consumption, environmental action and sustainable transport. Environmental naming patterns appeared in 15 projects, and innovation often involved combining familiar features such as rewards, tracking and personalisation without demonstrating a clear market gap. No submission contained a structured AI-use declaration, and incomplete materials or inaccessible links sometimes made the work difficult to assess.

Clearer Rules Exposed the Distance Between Disclosure and Responsibility

The 2026 assignment explicitly described the projects as hypothetical and required at least two external sources to support the proposed market need. It restored detailed customer profiles, introduced eight weighted marking criteria, required comparisons with industry benchmarks and asked students to collect their evidence in one PDF. Mandatory AI declarations made the tools' involvement visible across all 17 readable submissions.

Only three submissions provided comparatively comprehensive accounts connecting tools, purposes and the materials they affected. Several declarations described AI assistance with market research, keyword selection or performance targets, placing it inside the reasoning teachers were assessing. Admitting that AI helped did not establish whether students checked its output, understood its limitations or retained control over the decisions.

13 submissions cited at least two identifiable external sources, and 15 reported numerical keyword information, but the supporting records were often difficult to reproduce. Just one supplied source-linked benchmarks across its KPI set, a weakness partly connected to instructions that requested benchmark comparisons without explicitly requiring their sources or grading traceability separately. Eight proposed products incorporated AI, with little discussion of privacy, bias, errors, human oversight or computing's environmental impact.

Better Assessment Needs Reasons That Students Can Defend

The projects' themes changed with the assignments, moving from familiar products to green services and then a broader mix that included financial management, health and wellbeing. Environmental naming became less common in 2026, and functional or technological identities gained ground, without proving that the ideas had become more innovative. The researchers observed signs of better customer fit and more reasoned platform choices, but differences between assessors made some interpretations tentative.

The authors recommend grading the chain connecting evidence, creative choices, expected customer behaviour and numerical targets, with examples that teach a method rather than supply an answer. Their proposed responsible-AI framework goes beyond disclosure to include justified use, verification, human decision-making and attention to possible harms. Future assignments could require a documented fact-check, retrievable research records, a brief explanation of risks and an oral defence of selected decisions, supported by a simpler ten-indicator assessment framework.

These recommendations come from one course with voluntary participation, different yearly groups and an instructor who designed the assignments and conducted the main analysis. A second assessor checked 30 submissions and agreed strongly on several AI-related measures, but agreement was weaker on more interpretive judgements, and the study did not measure learning gains. Its practical contribution is a way to make student reasoning easier to question and assess, giving teachers something more dependable than the confidence of a presentation.