In the space of two years, generative artificial intelligence has moved from the research laboratory to the center of the global marketing conversation. The public release of ChatGPT in late 2022, followed by rapid successive waves of image, video, and audio generation tools, has transformed the way brands produce the content their customers see. A luxury brand's Instagram campaign, a fast-food chain's television advertisement, a bank's customer service chatbot response — all of these can now be produced, in part or entirely, by algorithms that no consumer will ever meet.
This transformation raises a question that is at once ancient and unprecedented. The question is ancient because it is the question of trust: when a consumer looks at a brand's message, on what basis do they decide whether to believe it? The question is unprecedented because, for the first time in the history of commercial communication, the message may have been created without any human author at all. Regulators around the world have begun to answer this question with a single tool — mandatory disclosure. The European Union's AI Act, adopted in 2024, requires that AI-generated content be clearly labelled. The rationale is straighWorward: consumers, when informed, will be able to make better choices about which brands to trust.
But is this rationale correct? An emerging body of Western research, published mostly in the last two years, has begun to question it. Studies conducted in the United States and Europe (Schilke & Reimann, 2025; Buder, Hesel & Dietrich, 2024; Kirkby, Baumgarth & Henseler, 2023) suggest that AI disclosure, far from strengthening trust, may erode it. The finding has been so consistent that it has been labelled the "transparency dilemma" — the tension between the ethical imperative of transparency and the commercial cost it appears to impose. This new literature poses a serious question for regulators, for brands, and for consumer researchers.
It also poses a serious question for the present author. The entire body of AI disclosure research has been conducted in Western contexts — in the United States, in the United Kingdom, in Germany, in the Netherlands. There is virtually no equivalent research on the Moroccan consumer, and by extension on the North African consumer more broadly. Yet Morocco is undergoing a rapid digital transformation of its own: 81 per cent of the population is active on social media (Sunergia, 2025), the domestic e-commerce market exceeded 22 billion Moroccan Dirhams in 2023 with more than 30 per cent annual growth (Morocco World News, 2025), and the Conseil Économique, Social et Environmental has published its own national framework for AI governance (CESE, 2024). Moroccan consumers encounter AI-generated marketing content daily, but their responses to it are documented nowhere in the international literature.
This absence is the central motivation of the present thesis. The research asks whether the transparency dilemma documented in Western contexts also manifests among Moroccan consumers, and if not, why not. To answer, an experimental study was designed and administered online to a sample of 468 Moroccan respondents. Each respondent was randomly exposed to one of three versions of a real luxury fragrance campaign (Emporio Armani "Power Of You"): a version disclosed as AI-generated, an undisclosed AI-generated version, and a real human-created version presented as such. Responses were measured on three central constructs — perceived brand authenticity, brand trust, and purchase intention — along with individual-difference moderators drawn from the Persuasion Knowledge Model, the algorithm aversion literature, and the Stimulus – Organism – Response framework.
The findings, developed in detail in Chapters 4 and 5, reframe the Western debate in an important way. On the average Moroccan respondent, disclosure has almost no measurable effect: five of the six formal hypotheses fail to reach conventional significance. But this null aggregate result conceals two substantive findings. First, the manipulation check reveals that Moroccan consumers apply a generalized suspicion of AI to nearly all marketing content they view, whether the content is actually AI-generated or not. This "ceiling of suspicion" reduces the marginal informational value of a disclosure label — because the audience is already suspicious. Second, the interaction test shows that algorithm aversion significantly moderates the disclosure effect (β = 0.41, p = .008), but not in the direction Western research would predict: it is the AI-friendly consumers who penalise disclosure, while the AI-sceptical consumers actually reward the transparency signal. The Moroccan data thus invert the implicit audience assumption of the disclosure literature.
These findings carry consequences beyond the individual thesis. Theoretically, they identify a boundary condition on the Western disclosure literature — the transparency dilemma requires a certain level of baseline consumer trust in AI to manifest, a level that appears to be absent in the Moroccan context. Managerially, they suggest that Moroccan brands wishing to build trust should invest less in disclosure labels and more in interpersonal social proof — real customer testimonials, family recommendations, and user-generated content that scored dramatically higher than brand-level communication in the descriptive analyses. Politically, they support the multi-lever approach recommended by the CESE (2024) framework: disclosure obligations should be combined with AI literacy programmes, technical verification standards, and industry self-regulation, rather than treated as a stand-alone solution.
The remainder of this thesis is organized in five chapters. Chapter 1 establishes the general framework of the study: the evolution of generative AI in marketing, the research problem, the central research question and its five sub-questions, the operational research objectives, the theoretical constructs to be developed, the regulatory and ethical landscape, and a review of the existing literature. Chapter 2 develops the theoretical foundations of the study in depth, articulating the Persuasion Knowledge Model, algorithm aversion theory, and the Stimulus – Organism – Response framework, and formally states the six research hypotheses. Chapter 3 presents the methodological design of the study — the epistemological posture, the experimental protocol, the sampling procedure, the measurement instruments, the statistical analysis plan, and the ethical considerations. Chapter 4 reports the results of the empirical study, from sample description and reliability analysis through the six formal hypothesis tests. Chapter 5 discusses the findings, situates them in relation to the Western literature, articulates the theoretical contributions, derives managerial and policy implications, acknowledges the limitations of the study, and proposes directions for future research.
