Trust Matters More Than Hype for ChatGPT Adoption in Classrooms

The First Thing That Matters Is Not What You Think

Two Chinese university students look at the same ChatGPT interface. One sees a tool that will save hours of grunt work. The other sees a black box that might spit out plausible nonsense and get them accused of cheating. What separates them is not technical skill, age, or even how much they already know about AI. It is trust.

That finding, from a 2024 study by Muhammad Farrukh Shahzad, Shuo Xu, and Iqra Javed at the School of Economics and Management, Beijing University of Posts and Telecommunications, upends the usual story about why people adopt new technology. For years, the dominant model in tech adoption research has been the Technology Acceptance Model (TAM), which says two things matter: how easy a tool seems to use, and how useful it appears. Make something easy and useful, the logic goes, and people will pick it up.

Shahzad and his colleagues found something messier. They surveyed 320 Chinese university students about ChatGPT and ran their responses through a statistical method called partial least squares structural equation modeling (PLS-SEM). The results showed that trust does not just nudge adoption along. It fundamentally changes how students perceive the tool in the first place. When students trusted ChatGPT, they found it easier to use, more useful, and more intelligent. When they did not, no amount of slick marketing or peer pressure could compensate.

In other words, trust is not the cherry on top. It is the soil the whole plant grows in.

How Trust Rewires Perception

The study asked students about five things: their awareness of ChatGPT, how easy they found it to use, how useful they thought it was, how intelligent they believed the AI to be, and how much they trusted it. Then the researchers mapped how those perceptions connected to the students' actual intention to adopt ChatGPT for academic work.

The results were striking. Trust moderated every single path from awareness to perception. Students who trusted ChatGPT rated it as significantly easier to use than those who did not, even when both groups had the same level of familiarity with the tool. The same pattern held for perceived usefulness and perceived intelligence. Trust did not just make students more willing to use ChatGPT. It made them see the tool differently.

This is not just a statistical curiosity. It has real consequences for how universities design AI literacy programs. A workshop on how to write good prompts is not going to help the student who does not trust the tool in the first place. That student is not thinking about prompt engineering. They are thinking about whether the AI will fabricate a citation and land them in front of an academic integrity board.

The study also found that perceived ease of use, usefulness, and intelligence all acted as mediators between awareness and adoption intention. That means knowing about ChatGPT is not enough. Students have to go through a chain of judgments. They have to believe the tool is easy, useful, and smart before they will adopt it. And trust is the gatekeeper for all three judgments.

The Three Perceptions That Matter

Perceived Ease of Use

The standard TAM logic says that if a tool is easy to use, people will use it. But Shahzad and colleagues found that trust determines whether students experience the tool as easy in the first place. A student who trusts ChatGPT will overlook a confusing interface and assume the friction is temporary. A student who does not trust it will interpret the same friction as evidence that the tool is broken.

This is a subtle but powerful finding. It means that universities cannot just focus on making AI tools user friendly. They also have to build the trust that allows students to experience them as user friendly.

Perceived Usefulness

The study found that trust also shaped whether students saw ChatGPT as useful. This makes intuitive sense. A tool you do not trust is not useful because you cannot rely on its outputs. You have to fact check everything, which defeats the purpose of using it in the first place. Students who trusted ChatGPT reported that it saved them time and helped them understand difficult concepts. Students who did not trust it reported that it created more work by forcing them to verify everything.

Perceived Intelligence

This was the most interesting mediator. The researchers added perceived intelligence to the standard TAM model, and it turned out to be a significant factor. Students who saw ChatGPT as intelligent were more likely to adopt it. But again, trust was the gatekeeper. Students who trusted the AI were more likely to perceive it as intelligent. Those who did not trust it saw the same outputs as dumb or random.

This echoes a finding from human psychology: we trust people we perceive as competent, and we perceive people as competent if we trust them. It is a virtuous or vicious cycle, depending on where you start.

The China Context and Why It Matters

The study was conducted in China, which raises a natural question: does this apply elsewhere? China has a unique educational and technological environment. The government has invested heavily in AI education, and students are generally more familiar with AI tools than their peers in many Western countries. The sample of 320 students came from a single university, which limits generalizability.

But the underlying mechanism trust as a moderator of perception is likely universal. Shahzad and colleagues grounded their framework in established psychological theory. The Technology Acceptance Model has been validated across dozens of countries and contexts. Adding trust as a moderator is a logical extension, not a cultural outlier.

What would be genuinely useful is a cross cultural replication. Do students in the United States, where skepticism of institutions is higher, show an even stronger trust effect? Do students in Nordic countries, where trust in government and technology is generally high, show a weaker one? The study does not answer these questions, but it raises them in a productive way.

What the Research Does Not Prove

This is where journalism has to be honest with readers. The study shows correlation, not causation. The PLS-SEM method can test whether a proposed model fits the data, but it cannot prove that trust causes changes in perception. It is possible that students who find ChatGPT easy to use and useful simply develop trust as a consequence. The arrow might run the other way.

The researchers acknowledge this limitation. They call for longitudinal studies that track how trust and perception evolve over time. A student who starts out skeptical might become trusting after positive experiences. Or a trusting student might become skeptical after a bad interaction. The study captures a snapshot, not a movie.

Another limitation is the sample. All 320 students were from one Chinese university. They were likely younger, more tech savvy, and more homogeneous than the general population of higher education students worldwide. The findings might not replicate with older students, students in non technical fields, or students in countries with different cultural attitudes toward AI.

Finally, the study measured intention to adopt, not actual adoption. There is a well known gap between what people say they will do and what they actually do. A student might report high trust and high intention to use ChatGPT, then never open the app because they run out of time or get distracted by TikTok. Intention is a useful proxy, but it is not the real thing.

The Hard Question for Educators

The study has a clear implication for universities: if you want students to use AI tools effectively, you have to build trust first. But how do you do that?

Trust in technology is not the same as trust in people. You cannot have a conversation with ChatGPT and read its facial expressions. You cannot check its references or ask it for a second opinion. Trust in AI is built through reliability. The AI has to consistently produce accurate, useful outputs over time. But students cannot build that trust if they are afraid to use the tool in the first place.

This is the paradox. Universities are right to be cautious about AI in the classroom. They need to teach students about plagiarism, hallucination, and the limits of machine intelligence. But if the messaging is purely negative, it destroys trust and makes students less likely to use the tool safely. The baby goes out with the bathwater.

Shahzad and colleagues put it bluntly in their paper: "assessments must promote the safe use of ChatGPT" while "maintaining students' critical thinking skills and inventiveness." That is a delicate balance. Too much restriction, and students will either ignore the tool entirely or use it secretly without guidance. Too little restriction, and they will copy outputs without understanding them.

What This Actually Means

• ▸Trust is the bottleneck. If students do not trust an AI tool, no amount of training on how to use it will help. Universities should measure trust levels before rolling out AI literacy programs, and design interventions that address skepticism directly.

• ▸Perception is not reality. What students experience as ease of use or usefulness is partly determined by their trust level. A tool that one student finds intuitive might feel confusing to another student who distrusts it. Universities should not assume that a well designed interface is enough.

• ▸The three perceptions are a chain. Awareness leads to perceived ease of use, usefulness, and intelligence, which lead to adoption intention. But trust moderates every link in that chain. Building trust is not a one time event. It has to be maintained across every interaction.

• ▸Safe use requires trust, not fear. The instinct to protect students from AI by emphasizing risks can backfire. Students who are afraid of the tool will not learn to use it safely. They will either avoid it entirely or use it in secret without guidance. Clear guidelines and instructions, as the authors recommend, are better than blanket warnings.

• ▸This is a design problem, not a marketing problem. Universities often treat AI adoption as a matter of communication. Tell students about the tool, show them how it works, and they will use it. The study suggests that is wrong. AI adoption is a trust problem, and trust is built through reliability, transparency, and consistent positive experiences. That is harder than sending an email blast. It is also the only thing that works.