Generative artificial intelligence (GenAI) is entering schools at speed. In policy and professional discourse, GenAI is often framed as an innovation that can improve teaching quality, support differentiated instruction, and reduce teachers’ routine workload (Kasneci et al., 2023; Selwyn, 2019; UNESCO, 2023). Yet emerging practice points to a tension that existing research has not fully explained: GenAI does not simply “add capacity” to teaching. Instead, it is taken up inside evaluation-dense school systems where teachers’ work is continuously sorted, compared, and rewarded across multiple tracks - teaching performance, research/lesson-study outputs, administrative tasks, and competitive awards - often under performative and audit-like expectations (Ball, 2003; Power, 1997). In such contexts, GenAI’s effects are likely to be uneven. It offers immediate gains in work that is highly text-based, templateable, and output-oriented (e.g., drafting lesson plans, polishing narratives for competitions, packaging research/teaching outputs), while providing less direct leverage for relational, situated, and responsibility-heavy labour (e.g., classroom responsiveness, long-term pastoral care, ethical judgment in interaction) (Zawacki-Richter et al., 2019). As a result, “efficiency” does not automatically translate into “reduced burden.” It may instead reconfigure what is experienced as worthwhile work by shifting effort toward activities that are more easily optimised and made visible as evaluative evidence, thereby intensifying metric effects and reactivity (Espeland & Sauder, 2007; Williamson, 2017).

This study addresses this puzzle by theorising GenAI as an amplifier of evaluative visibility rather than a neutral instructional tool. It asks how evaluation-driven visibility regimes interact with GenAI’s selective optimisability to generate misalignment pressures: pressures arising when the work that is easiest to enhance and document diverges from the work teachers value as educationally meaningful but harder to render as evidence (Ball, 2003; Espeland & Sauder, 2007). The study centres teacher agency as a situated professional accomplishment rather than an individual trait, analysing agency as interpretive and strategic work under constraint - reallocating time and attention, deciding where and how to use GenAI, maintaining boundaries around “non-delegable” professional judgment and relational work, and managing moral and reputational risk as visibility expectations intensify (Biesta & Tedder, 2007; Priestley et al., 2015).

Three research questions guide the inquiry:
RQ1 What misalignment pressures arise when GenAI selectively increases the optimisability and visibility of particular teacher-work dimensions under multi-track evaluation systems?
RQ2 What distinct agency strategies do teachers develop in response?
RQ3 What professional consequences are associated with these strategies for sustainability (e.g., meaning, stress, perceived control, identity coherence)?

Empirically, China serves as a strategic case because evaluation density and competitive visibility mechanisms are pronounced, making the underlying mechanisms easier to observe (Ball, 2003; Power, 1997). The study does not treat China as exceptional; rather, it uses a “high-intensity” context to surface dynamics that resonate globally as many systems - including in Europe - expand performance indicators, accountability infrastructures, and digital governance tools (Espeland & Sauder, 2007; Williamson, 2017). Methodologically, the analysis combines (1) publicly available policy and institutional texts that specify evaluative evidence and priorities with (2) semi-structured interviews with teachers across school types and career stages. By integrating institutional-level evidence with practice narratives, the study will map visibility regimes, trace GenAI-driven payoff shifts across work domains, develop a typology of agency strategies (e.g., visibility-optimising, boundary-protecting, pedagogically re-centring, and hybrid forms), and link these strategies to differentiated experiences of professional sustainability. The expected contribution is an explanatory account of why similar GenAI tools can produce divergent effects across contexts, offering a transferable framework for European debates on teacher workload, accountability, and responsible AI integration (Selwyn, 2019; UNESCO, 2023).

Research design A qualitative explanatory case study using multi-source evidence to reconstruct mechanisms linking (a) evaluation-driven visibility regimes, (b) GenAI’s selective optimisability across work domains, and (c) teachers’ agency strategies and consequences. Data sources (1) Public/institutional texts (visibility regime corpus). Documents were purposively collected to represent the evaluative environment that structures “what counts” as teacher work. The corpus include: (a) national/regional policy and guidance on teacher evaluation, workload, and AI-in-education; (b) school- and district-level appraisal rubrics, performance score sheets, portfolio/“evidence” requirements, lesson observation protocols; (c) competition and award guidelines (e.g., demonstration lessons, teaching contests), including scoring criteria and submission templates; and (d) official exemplars of “high-quality” outputs (model lesson plans, competition narratives). Collection will follow a transparent protocol specifying source types, issuing bodies, and time window; items will be logged with metadata (date, level, genre, stated criteria). (2) Semi-structured interviews (agency and consequence corpus). Interviews: 35 teachers (iteratively adjusted to meaning saturation), sampled for maximum variation by school level, subject, career stage, school location/type, and perceived evaluation intensity. Interviews (60–90 minutes) were elicited concrete episodes of GenAI use/non-use across domains (teaching, research/lesson-study writing, admin, competitions), decision rationales, boundary judgments, perceived risks, and consequences (stress, meaning, control, identity). Optional supplementary interviews (5–10) with middle leaders/evaluators may be used to clarify how evidence is interpreted and rewarded. Analysis A three-stage deductive–inductive workflow: 1.Framework coding (deductive) using initial codes for visibility demands, optimisable tasks, misalignment pressures, and boundary practices. 2.Strategy construction (inductive) from action narratives to build a typology of agency repertoires and their internal logic. 3.Cross-case matrices linking strategy × conditions × consequences, identifying patterned associations and negative cases. Trustworthiness & ethics Triangulation (texts × interviews), negative-case analysis, peer debriefing, and an audit trail of coding decisions. Informed consent, anonymisation, and careful handling of reputational risk around GenAI use.

This study shows that GenAI’s early effects in schools are best understood as an amplification of evaluation-driven visibility, not a uniform enhancement of teaching. In the Chinese case, multi-track evaluation systems - where evidence requirements, competitive showcases, and performance metrics coexist - create a payoff structure in which some teacher tasks become disproportionately “optimisable” and documentable through GenAI. The analysis identifies misalignment pressures that arise when GenAI strengthens the efficiency and polish of output-oriented work (e.g., narrative packaging, template-based documents) faster than it improves relational, situated, and responsibility-laden labour that is central to educational quality but harder to turn into recognised evidence. Under these conditions, time savings are not necessarily released to classroom improvement; they are frequently reabsorbed into intensified visibility work and escalating expectations. The study further demonstrates that teachers respond through distinct agency strategies rather than simple adoption/non-adoption. These strategies include visibility-optimising use (targeting high-reward outputs), boundary-protecting use (restricting GenAI in domains tied to professional judgment and relationships), pedagogically re-centring use (redirecting GenAI toward instructional design and feedback in ways that preserve teacher authority), and hybrid combinations shaped by career stage and local governance. These strategies are associated with differentiated professional consequences: some patterns stabilise workload and preserve meaning by protecting non-delegable practices, while others increase strain through performative escalation and moral ambivalence. Beyond China, the findings offer a globally relevant mechanism: wherever teacher work is governed through expanding indicators and evidence demands, GenAI is likely to shift effort toward what is most optimisable and visible. The study therefore reframes “responsible AI integration” as an organisational and evaluative design problem - about what systems reward and recognise. This aligns with ECER’s concern for how digital technologies and standardised indicators reshape knowledge credibility and action in education systems.

Ball, S. J. (2003). The teacher’s soul and the terrors of performativity. Journal of Education Policy, 18(2), 215–228. Biesta, G., & Tedder, M. (2007). Agency and learning in the lifecourse: Towards an ecological perspective. Studies in the Education of Adults, 39(2), 132–149. Espeland, W. N., & Sauder, M. (2007). Rankings and reactivity: How public measures recreate social worlds. American Journal of Sociology, 113(1), 1–40. Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kasneci, G., Lefkir, Y., Maier, U., Mueller, T., Pfeifer, K., Severin, T., Shankar, S., Strzelecki, A., … Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Computers and Education: Artificial Intelligence, 4, 100114. Power, M. (1997). The audit society: Rituals of verification. Oxford University Press. Priestley, M., Biesta, G., & Robinson, S. (2015). Teacher agency: An ecological approach. Bloomsbury Academic. Selwyn, N. (2019). Should robots replace teachers? AI and the future of education. Polity Press. UNESCO. (2023). Guidance for generative AI in education and research. UNESCO. Williamson, B. (2017). Big data in education: The digital future of learning, policy and practice. SAGE. Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education - Where are the educators? International Journal of Educational Technology in Higher Education, 16, 39.