Research Article
Teachers’ attitudes and behavioural intentions towards AI integration in Chinese higher education: a qualitative interview study
Synthesis: Fourteen university teachers across four Chinese higher education institutions describe generative AI less as a rival than as an assistant whose limits they are still negotiating. Interviewed for approximately 40–60 min each, the 6 female and 8 male participants produced five interrelated themes, and the through-line is a shift in what "threat" means: their strongest concerns were not about their own replacement but about what AI use does to students' independent thinking, reasoning, originality, and authentic understanding. Acceptance was conditional and bounded — shaped by pedagogical purpose, disciplinary context, student readiness, and the wish to keep teacher control — and the support teachers asked for went past general encouragement, toward AI literacy, discipline-sensitive tools, and infrastructure.
Key Findings
- AI was framed as support rather than substitution. Across the 14 interviews, participants rarely described AI as something that could replace a teacher. They positioned it as an auxiliary assistant: "only playing an auxiliary role" (Participant 002), "human-led, with AI as support" (Participant 004), and, most directly, "I use AI; AI does not use me" (Participant 001). Even the most enthusiastic participants insisted AI remain secondary to human judgment and teacher control.
- Usefulness was judged discipline by discipline. Teachers weighed AI against the knowledge demands and aims of their own field. Design-oriented teachers associated it with idea generation and creative expansion, while an engineering teacher argued AI-generated content should be treated as "information" rather than "knowledge" until professionally checked (Participant 013). Teachers in ideological and political education asked for content aligned with disciplinary and value-oriented requirements.
- Emotional responses split into three broad patterns, with declining satisfaction in a few cases. A substantial group described enthusiasm, curiosity, and a sense of achievement (Participants 009 and 012). Others combined acceptance with caution: one described "anxiety" and "a very obvious feeling of losing control" when AI-generated interpretations pulled her from her intent (Participant 012). A smaller but important group reported neutrality, treating AI as "just a tool" (Participant 010), while in a few cases initial appreciation turned to dissatisfaction as outputs became formulaic (Participant 011).
- Students, not teachers, sat at the center of concern. The most consistent tension concerned students' misuse of AI. Teachers worried that students moved too quickly from a task to an answer, "handing over the process of thinking to AI" (Participant 003) or using it "just to see the answer and complete the task, but they do not think about it" (Participant 009). The worry was strongest where foundational understanding or step-by-step reasoning mattered, since the result could survive while the logical chain disappeared.
- Weak discrimination and polished surface work compounded the worry. Participants doubted students could judge output quality: one warned they might not "identify whether it is right or wrong" (Participant 004). Another described presentations that looked "very well done" but, when delivered, "it was not very good" (Participant 011) — AI improving the appearance of performance without strengthening understanding. One teacher called the result "a false sense of achievement, or false prosperity" (Participant 008).
- Adoption was conditional, need-based, and bounded. Teachers used AI when it served a clear purpose rather than as habit, and insisted human thinking come first: "I do not stop them from using AI, but I first ask them to think by themselves" (Participant 009). Suitability varied by course stage and perceived student capability, and several framed the boundary through professional identity — "I control the class; I would not let AI control it" (Participant 006). Broadly positive attitudes therefore did not produce high-frequency classroom use.
- Support had to go beyond generic encouragement. Participants wanted training grounded in their own teaching tasks and disciplines, and several distinguished frequent AI use from genuine AI literacy, noting many teachers could not yet formulate effective prompts. They asked for specialized or "domesticated" models (Participant 013) and reliable platforms and hardware, warning that without platform development many proposed applications would remain "castles in the air" (Participant 014).
How the study was done
Fourteen academic staff with ongoing or recent teaching responsibilities were recruited from four higher education institutions in China through the research team's professional and institutional networks, selected for variation in discipline, rank, and teaching experience. The final sample included 6 female and 8 male participants from mental health education, ideological and political education, education, digital media art, programming, engineering, and financial management, with teaching experience from approximately 2 weeks to 30 years (Table 1). Semi-structured interviews of approximately 40–60 min were conducted in person or online and transcribed verbatim.
Analysis used codebook-informed thematic analysis combining deductive and inductive coding in ATLAS.ti, with three researchers involved in coding and analytic discussion and an independent researcher reviewing a subset as a credibility check. The authors state this was not a statistical reliability calculation but a means of improving conceptual clarity. Deductive coding drew on the three-component attitude model, TAM, and the theory of planned behavior; inductive coding captured patterns beyond those lenses. The 14 interviews generated five interrelated themes, mapped onto the study's four research questions.
Attitudes were treated as cognitive, affective, and behavioral components that remain interconnected, which is why teachers can recognize AI's practical value while resisting its educational implications.
Support without substitution, and the discipline that decides usefulness
Efficiency was the most consistent finding. Teachers used AI for lesson preparation, material retrieval, wording refinement, routine writing, and presentation design, describing it as "very convenient… greatly improving efficiency and saving time" (Participant 008). What distinguishes the account is the division of labor teachers described: AI could draft the frame of a presentation, but "the flesh and blood are still filled in by me" (Participant 001). One called AI "a secretary" that "cannot lead us" (Participant 014); another insisted "the core, the absolute core, is still your own thinking" (Participant 007).
That boundary work explains why usefulness did not generalize into uniform adoption; the authors read it as an extension of TAM rather than a refutation: usefulness still predicts intention, but it is assessed against pedagogical purpose rather than treated as inherent to the tool.
Emotionally, the picture was uneven. Positive affect was real — relief, inspiration, increased confidence — but traveled with caution, and in some accounts decayed: one teacher reduced her dependence after outputs began repeating "the same routines again and again" (Participant 011).
Why student learning, not teacher replacement, set the boundary
The paper's most useful contribution is displacing the replacement question: participants rarely framed AI as a direct threat to their roles, but as a threat to the learning processes they were responsible for protecting. The concerns recurred across interviews: students outsourcing reasoning (cognitive offloading), over-trusting plausible but flawed output, completing tasks without understanding them, and losing originality and authentic engagement. A further point cut the other way — that some students, particularly in higher vocational contexts, did not yet know how to use AI meaningfully for study at all. Student-related concerns therefore took at least two forms: excessive dependence and underdeveloped educational use.
These are perceptions, not measurements. But they explain the shape of the adoption decision: teachers wanted AI to arrive after students had attempted their own thinking, to be reserved for tasks where students could check the output, and to stay under teacher direction. That is a pedagogical sequencing principle, not a technology policy, and it explains why the same teacher could be enthusiastic about lesson planning with AI and restrictive about it in a foundational course.
The support findings follow the same logic. Teachers asked for professional development grounded in their own tasks, for AI and prompt literacy rather than mere exposure, for discipline-specific tools, and for infrastructure that made classroom use possible. The authors summarize this as a move "beyond technological access" toward a support ecology combining teacher literacy, student critical engagement, and discipline-responsive institutional conditions — closer to governance than to tool procurement.
What this means for practice
- Instructors. Sequence tasks so student thinking precedes AI use; participants objected not to use but to use that replaced the reasoning step.
- Faculty developers. Replace one-off general workshops with demonstration-based sessions tied to specific disciplines, and teach prompt formulation and output evaluation explicitly, since participants distinguished frequent use from genuine AI literacy.
- Administrators. Treat platforms, hardware, and discipline-sensitive models as prerequisites rather than extras; teachers described infrastructure gaps as the binding constraint, not motivation.
- Policymakers. Build in recognition and workload relief for the learning costs of adoption, and resist standardized solutions for teaching participants described as varied.
Limitations
- The study rests on a relatively small purposive sample of 14 teachers from four higher education institutions in China, recruited through professional and institutional networks, which may have favored teachers more willing to discuss AI integration.
- Findings are context-specific and should not be assumed to represent all Chinese higher education institutions or other systems; curricular influences, including those associated with ideological and political education, were not independently examined.
- Data come from self-reported interviews rather than classroom observation, so teachers' concerns about students are not evidence of changes in cognitive ability or learning outcomes.
- Only teachers were included, so students' experiences and institutional leaders' views are absent.
- A shared codebook supported coding consistency, but statistical inter-coder reliability was not calculated, and the independent researcher reviewed only a subset of the coded data and developing themes.
- One participant (P010) had no teaching assignment during the interview semester, and teaching experience ranged from approximately 2 weeks to 30 years, so accounts mix current practice with institutional and recollected perspectives.
Citation
Xiong, J., Zhang, W., Ouyang, S., Zhu, F., Zhang, S., & Jin, Q. (2026). Teachers’ attitudes and behavioural intentions towards AI integration in Chinese higher education: a qualitative interview study. Frontiers in Psychology, 17, 1968803. https://doi.org/10.3389/fpsyg.2026.1968803