Predicting Technology Adoption
One of the most well-known frameworks for technology adoption research is the UTAUT2 model, which seeks to understand the psychological and environmental factors influencing people’s decision to use a new tool or system. UTAUT2 considers not only whether we find a technology useful or easy to use, but also what our environment thinks of it, whether we receive support for its use, how enjoyable we find it, and how much it becomes part of our daily routine.
The original model – and its enhanced versions – have repeatedly proven capable of reliably predicting the likelihood of someone actually using a given technology. According to some research, UTAUT2 can explain up to 70–80% of the variance in technology use intention, making it particularly strong in areas involving the introduction of new, non-routine technologies – such as artificial intelligence.
AI Adoption in Higher Education
The emergence of artificial intelligence in higher education raises new questions: it’s not just about whether the technology aids learning or teaching, but also how it changes roles, responsibilities, and the nature of the learning process. An AI assistant is not simply a new educational tool – for the student, it can be a tutor, a partner, or merely an aid, depending on the context in which it is used and the degree of autonomy attributed to it.
Below, we review five recent studies that apply different versions of the UTAUT model to examine how students and instructors relate to AI solutions, particularly systems based on generative language models.
Comprehensive UTAUT Review
This study is a classic UTAUT-based systematic literature review that analyzes technology adoption research conducted in higher education contexts, published between 2010 and 2022. Its uniqueness lies in the fact that it does not examine a single technology, but rather compares the application of the UTAUT model among students and instructors, and for various technology types (e.g., LMS, mobile learning, AR/VR, artificial intelligence).
Key Findings
- The model generally performs better for students: the prediction of intentional use and actual use is more accurate than for instructors.
- Performance expectancy is the strongest predictor in all groups, but the effect of social influence is much more pronounced among students.
- For AI-based technologies (in early examples, e.g., intelligent tutors), the role of facilitating conditions is particularly crucial – meaning the extent to which the necessary institutional and technical conditions for tool use are available.
- The measurement of actual use remains a weak point: most research examines intention, not what happens afterward.
Why is this Relevant Now?
This analysis covers a period when AI had not yet massively appeared in higher education, and ‘technology use’ predominantly referred to LMSs, mobile applications, or AR tools. The results therefore do not reflect the shift brought about by generative AI, but they do help to understand the underlying attitudes and patterns that characterized the various stakeholders before this transformation.
Takeaway
The authors argue that the UTAUT model remains a valuable framework for examining educational technology adoption – but they emphasize that the model’s expansion and adaptation are necessary in situations where use is not merely a matter of intention, but increasingly a structural reality. This is particularly true for AI, which not only needs to be ‘accepted’ – but also used, interpreted, controlled, and evaluated.\
AI Adoption Based on UTAUT2
This systematic literature review analyzes how higher education stakeholders relate to AI technologies, based on 50 studies published between 2018 and 2023. Using the UTAUT2 model, they identified factors facilitating and hindering adoption among students, instructors, and administrative staff.
The study’s significance lies not only in its scope: it chronologically concludes the period before the generative AI explosion. It summarizes studies from the pre-ChatGPT era, a time when AI had not yet provoked the same sense of existential threat from teachers and institutions as it does today – yet adoption was already measurable and often positive.
Main Findings
- The most reliable predictors were performance expectancy and the enjoyment associated with AI use (hedonic motivation) – these appeared across all stakeholders.
- For students, social influence and ease of use of the tool are also important. For instructors, supportive conditions (facilitating conditions) and practical benefits are more significant.
- A clear difference emerged between STEM and non-STEM fields: in the former, self-confidence and professional relevance strengthened adoption, while in the latter, skepticism and lack of knowledge were more common.
- Technological background, age, and institutional context (e.g., differences between public vs. private universities) also emerged as moderating factors.
Conclusion
The main takeaway from the study is that the UTAUT2 model can not only interpret but also predict the likelihood of various higher education groups actually using AI technologies. Performance expectancy and the experience factor (hedonic motivation) are the strongest predictors: where these are high, the likelihood of use also increases.
Furthermore, it becomes clear that students, instructors, and institutions accept or reject AI for different reasons and under different conditions. This differentiation suggests that general strategies are not sufficient for successful AI integration. They will truly use it if the tool aligns with the specific group’s needs, motivations, and prior experiences.
Future research should measure the likelihood of use not only through intention but also through actual behavior – especially in the new contexts of generative AI, where use may no longer be just a choice, but eventually a necessity or expectation.
A Model Incorporating Complexity
Carlos Enrique George-Reyes and his colleagues, as researchers at Tecnológico de Monterrey, published a surprisingly thorough, methodologically extremely rigorous study in June 2025. Their goal was to create and validate a questionnaire that reliably measures the factors influencing students’ acceptance of artificial intelligence in higher education.
The model is built upon a customized version of UTAUT2, but it was supplemented with an additional variable: perceived complexity. The researchers started from the premise that if students perceive AI use as too complicated, they will not use it – regardless of whether they otherwise consider it useful.
The Model’s Key: Perceived Simplicity
The researchers used a UTAUT2-based model but modified it: they omitted the factors of social influence and hedonic motivation, as they believed these relate more to the technology’s post-adoption experience rather than the initial decision. Instead, they introduced a new, central variable: perceived complexity, which plays a mediating role in the likelihood of AI use. According to the model, the less complicated the technology’s use is perceived to be, the higher the likelihood of its adoption.
Based on structural equation modeling (SEM), one of the most important factors was habit: those who regularly used AI found it far less complicated – and thus were more likely to continue using it.
The Likelihood of Use is Structured Complexly
The model predicted AI adoption with an explanatory power of 57% (R² = 0.567), which is a strong value in social science research. The key, therefore, is not merely that AI is “smart “or “works well”, but rather:
- is it understandable, manageable, and experiential,
- does the institution support its use, and
- how much students feel it’s worth using.
Why is this Interesting Now?
This study took place in 2023–2024, meaning precisely at the moment when generative language models (like ChatGPT) suddenly and massively entered education – but before the actual use of these technologies raised regulatory, ethical, or pedagogical questions.
The authors do not address students’ potential for manipulation or the problem of plagiarism – instead, they assume a manageable, measurable, pedagogically integrable technology. This makes the study an excellent benchmark: this is what the likelihood of AI use looked like during a period when artificial intelligence did not yet need to be defended from itself.
LLM Use in Real-World Contexts
This study approaches the topic from a radically different angle: it does not measure AI acceptance, but examines actual use – and its consequences. In research conducted at the University of Groningen in the Netherlands, 150 university students were surveyed about how and for what purpose they use generative artificial intelligence in their studies.
The authors do not apply the UTAUT model and do not attempt to measure ‘acceptance intention’. Instead, they examine how ChatGPT and other LLM-based tools actually appear in students’ learning practices – with particular attention to motivation, strategic decisions, and the perception of ethical boundaries.
Main Findings
- A significant portion of students use generative AI not as a tool for knowledge acquisition, but for academic survival.
- Use is often instrumental and superficial, for example, for quick text generation, summaries, or producing assignments.
- The majority of students see no problem with its use – especially if it does not involve explicit copying.
- AI use has become a type of learning strategy that does not replace learning, but reorganizes it: students use AI to decide what to focus on, what to omit, and what to entrust to ‘the machine’.
No Model – just Practice
The authors deliberately reject engineering models (like UTAUT) and instead approach the phenomenon with qualitative tools. Their argument is that measuring acceptance is no longer sufficient when the actual use of technology is de facto present, and institutions often have no answer on how to respond.
Takeaway: AI is not an Option, but Infrastructure
The study’s implicit assertion is that generative AI – especially large language models – has emerged in higher education not as an option, but as a new learning ecosystem. Instead of the question of acceptance, today the stakes are managing, supporting, and restricting its use.
Strategy is the Question – not Acceptance
As the Dutch research clearly shows: a significant portion of students use generative AI not for knowledge acquisition, but as a tool for academic survival.
The majority see no problem with it – especially if it doesn’t involve explicit copying. Thus, artificial intelligence does not replace learning, but reorganizes it. Students use AI to decide what to learn in depth, what to omit, and what to ‘entrust to the machine’.
This practice is already here – and it’s happening in the present, not in the future. The only substantive question is no longer whether students accept AI – but whether the university provides a framework, direction, and pedagogical control for it.
How Long Can the University Remain Disengaged from Students’ AI Use?
Students are already using it. The question now is whether the university guides, interprets, and can embed this process to improve the quality of learning – or leaves them to their own devices.
GoSchool helps ensure that artificial intelligence is not an uncontrolled tool, but a guided, curriculum-based learning partner. It learns from content defined by the instructor, and students can ask questions, practice, and explore in natural language.
Instructor control. Student autonomy. AI support tied to the curriculum. Learn what a course-assigned AI tutor means in practice.