Awareness and Confidence, Not Ease of Use, Drive ChatGPT Adoption Among Nigerian Students

When ChatGPT burst onto the scene in late 2022, universities around the world scrambled to understand whether their students would embrace it, ignore it, or abuse it. Most of the early evidence came from wealthy countries and technologically saturated campuses. Now a new study from Nigeria offers a striking counterpoint: among final-year marketing undergraduates in a developing economy, the factors that predict whether students intend to adopt the AI chatbot are not how easy it is to use, but how much they know about it and how confident they feel wielding it. In an era when generative AI is rewriting the rules of work and learning, the finding suggests that cognitive readiness may matter far more than slick interfaces.

The research, published in Discover Education, was led by Godswill Agu of Abia State University together with colleagues in Nigeria, Poland, and Spain. The team surveyed 213 final-year undergraduate marketing students drawn from Nigerian higher education institutions, focusing deliberately on students in their final year for three reasons: they have experienced the full breadth of their curriculum, they face cognitively demanding tasks such as dissertations and strategic campaign design where the temptation and utility of generative AI are greatest, and they are on the verge of entering a labor market where AI literacy is increasingly tied to employability. Data were collected over two weeks in October 2024 through a structured online questionnaire, and the researchers excluded eleven respondents who reported no prior awareness of ChatGPT, since awareness was a prerequisite for meaningfully reporting adoption intentions.

At the heart of the study lies a classic psychological framework: the Technology Acceptance Model, or TAM, first formulated by Fred Davis in the 1980s. TAM holds that two beliefs govern whether people embrace a new technology: perceived usefulness, the sense that the tool will improve performance, and perceived ease of use, the belief that operating it will require minimal effort. These beliefs shape attitudes, and attitudes in turn shape behavioral intention, the strongest proximal predictor of actual use. The model has endured for nearly four decades, but its authors and successors have long argued that its explanatory power can be enhanced by adding context-specific variables. Generative AI, the Nigerian team reasoned, is precisely the kind of technology that demands such an extension.

Unlike the relatively stable information systems of the pre-internet era, ChatGPT is interactive, adaptive, and demands active engagement. Users must craft effective prompts, critically evaluate AI-generated output for errors and bias, and integrate that output into academic work. The researchers therefore folded two additional constructs into the model. The first was awareness, drawn from Everett Rogers’s diffusion of innovations theory, which treats awareness as the opening stage of the innovation decision process: without knowing what a technology is and what it can do, people cannot form sensible judgments about its value. The second was perceived self-efficacy, rooted in Albert Bandura’s Self-Efficacy Theory, which captures a person’s confidence in their own ability to perform the tasks a technology requires. Crucially, self-efficacy is conceptually distinct from ease of use: the former is a judgment about the person, the latter about the system.

The analysis employed partial least squares structural equation modelling, a statistical technique well suited to exploratory models in emerging technology contexts, run in SmartPLS with bootstrapping on 10,000 subsamples. The sample skewed young and female: 57.28 percent of participants were women, and 88.26 percent were under 25 years old. The measurement scales, adapted from previously validated instruments and each measured with four items on five-point Likert scales, passed standard tests of reliability and validity, including factor loadings above the required threshold, satisfactory Cronbach’s alpha and composite reliability values, and convergent validity confirmed by average variance extracted figures exceeding the accepted minimum. Checks for common method bias, including Harman’s single-factor test and a full collinearity assessment, came back clean.

The results were dramatic. The extended model explained 81.6 percent of the variance in students’ attitudes toward ChatGPT and 80.1 percent of the variance in their intention to adopt it, figures the authors note are consistent with or higher than those reported in comparable studies elsewhere. Awareness emerged as the foundational driver, significantly boosting perceived usefulness, perceived ease of use, perceived self-efficacy, attitude, and intention alike. Its strongest single effect was on perceived ease of use, followed closely by self-efficacy and usefulness. Perceived usefulness significantly predicted both attitude and intention, and self-efficacy did the same, confirming that students who believed in their own capability to use ChatGPT effectively were markedly more positive about the tool and more determined to integrate it into their studies.

The surprise came with perceived ease of use. Although its path coefficients were positive, they failed to reach statistical significance for either attitude or intention. In other words, how easy students found ChatGPT to operate had no reliable bearing on whether they planned to use it. The authors offer a compelling explanation: for a generation of digitally fluent students, ease of use has become a baseline expectation rather than a differentiator. Moreover, the real challenge posed by ChatGPT is not technical but cognitive. The interface is simple; the harder skills involve formulating prompts, judging the reliability of outputs, and weaving AI content into rigorous academic work. Students, the study suggests, perceive the challenge of ChatGPT as skill-based rather than usability-based, which is precisely why self-efficacy, not ease of use, emerged as the significant determinant of adoption.

The study’s strongest single pathway ran from attitude to intention, with a path coefficient of 0.895. The authors caution that this very high value should be interpreted carefully, but they demonstrate that discriminant validity between the attitude and intention constructs held, with a heterotrait-monotrait ratio well within accepted bounds, meaning the effect is not an artifact of construct overlap. Notably, the finding challenges Davis’s original assertion that attitude is a weak predictor of intention, suggesting that for generative AI among technology-oriented students, attitudes function as a robust engine of adoption. It also aligns with prior work showing that as users grow familiar with AI and perceive its benefits, their attitudes shift from negative or merely instrumental to genuinely positive.

For educators and policymakers, the practical implications are pointed. Simply giving students access to generative AI tools may accomplish little if they lack the cognitive preparation to use them well. The authors argue that institutions should embed ChatGPT in structured learning activities that demonstrate its relevance to concrete marketing tasks, such as campaign development, while offering guided practice, prompt-design training, and explicit evaluation criteria to build student confidence. Clear ethical guidelines could reduce uncertainty and misuse. The authors also highlight that marketing students, as future shapers of consumer perceptions, may carry their attitudes toward AI well beyond the classroom, making mentoring and training in ethical and strategic AI use particularly consequential for this discipline.

The study has limits, as the authors acknowledge. It examined only final-year marketing students in a single country, limiting generalizability to other fields, levels, and national contexts, and its exclusively quantitative design cannot capture the texture of students’ lived experiences with the technology. Future work could test alternative frameworks such as the Unified Theory of Acceptance and Use of Technology, explore moderators like prior AI experience, digital literacy, or task complexity, and include the perspectives of administrators and policymakers. Still, the core message lands with force: in the age of generative AI, the decisive question is no longer whether a tool is easy to use, but whether people understand it and believe in their own ability to master it.

Subject of Research: Determinants of ChatGPT adoption among Nigerian undergraduate marketing students using an extended Technology Acceptance Model

Article Title: Extending the technology acceptance model with awareness and self-efficacy in predicting ChatGPT adoption among Nigerian marketing students

Article References: Agu, G., Obinna, E.-U. C., Okereafor, G. E., Omotosho, T. D., Chiana, C. A., Onyeokoro, S. C., Njoku, P. O., Okpara, G. S., & Margaça, C. (2026). Extending the technology acceptance model with awareness and self-efficacy in predicting ChatGPT adoption among Nigerian marketing students. Discover Education, 5(1), Article 1006. https://doi.org/10.1007/s44217-026-02152-y

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02152-y

Keywords: ChatGPT, generative AI, technology acceptance model, self-efficacy, awareness, higher education, Nigeria, marketing students, PLS-SEM, AI adoption, perceived usefulness, educational technology