Introduction
Protecting teachers’ mental health during digital change requires attention to work organization. WHO recommends organizational interventions, training, and support for workers with mental health conditions (World Health Organization, 2022; World Health Organization & International Labor Organization, 2022). Teacher-facing AI raises the question of whether efficiency leaves time for recovery and relationships or adds demands. Occupational well-being has affective, cognitive, professional, social, and psychosomatic dimensions (Van Horn et al., 2004); favorable technology ratings cannot establish improvement.
Job Demands-Resources (JD-R) theory distinguishes effort-intensive demands from resources supporting goals, learning, and engagement (Bakker & Demerouti, 2017). AI-assisted drafting may coexist with checking and learning demands (Chuang et al., 2025). This lens connects implementation to workload and resources relevant to workplace mental health.
Self-determination theory (SDT) addresses autonomy, competence, and relatedness: volition, effectiveness, and meaningful connection (Ryan & Deci, 2000). Their satisfaction is associated with wellness at work (Deci et al., 2017). JD-R and SDT support a post-hoc interpretation; neither theory nor its mental-health pathways were tested.
In English as a foreign language (EFL) teaching, AI can assist with texts, questions, and differentiated exercises, while educators retain responsibility for linguistic accuracy, cultural appropriateness, and assessment validity. Prior research identifies quality concerns, technological change, training gaps, and unclear policies as demands (Khlaif et al., 2026; Kohnke et al., 2024; Zaiarna et al., 2024), alongside expectations of workload relief (Alharbi, 2025). The question for mental health promotion is how universities organize these responsibilities and support teachers during change.
Twee provides a case of remote university teaching during the war in Ukraine. This context makes access and continuity of teaching relevant without implying trauma or distress among participants.
Purpose
This analysis identifies perceived demands and resources surrounding teacher-facing AI and interprets their implications for teachers’ mental health promotion. Empirical questions concern Twee use, usefulness, and constraints; selected teacher-student differences; and teachers’ accounts of workload, learning, and their human role.
Methodology
Design and setting
An exploratory mixed-methods design combined a cross-sectional online survey with semi-structured teacher interviews, integrating survey patterns and interview accounts (Fetters et al., 2013). Data collection began on 18 January 2026 and continued in February at H. S. Skovoroda Kharkiv National Pedagogical University (HNPU) and National Technical University “Kharkiv Polytechnic Institute” (NTU “KhPI”), Ukraine. Teaching at both universities was remote. Students were in the later years of their programs; conflict exposure and trauma were not measured.
Participants and recruitment
Recruitment used university email lists, departmental announcements, and official messenger groups. The 425 invitees were unique contacts across both institutions, without duplicate invitations; 150 provided complete responses (35.3%). The sample included 56 EFL teachers (35 from HNPU and 21 from NTU “KhPI”) and 94 students/student-tutors (55 from HNPU and 39 from NTU “KhPI”). Of these students, 22 were student-tutors, defined as senior students assisting in language laboratories or peer mentoring, and 72 were other senior students. Twelve teachers were purposively selected from survey respondents for differing teaching experience and AI familiarity and interviewed online, including teachers temporarily abroad.
Eligibility required age 18 or older and employment as an EFL teacher or enrolment as a senior student/student-tutor (third- or fourth-year bachelor’s or master’s level) at a participating university, with internet access. Incomplete survey submissions were removed; no other post hoc exclusions were applied.
Women constituted 82.1% of teachers and 68.1% of students. Teachers’ means age was 39.5 years (SD = 8.5), with 12.4 years of teaching experience (SD = 7.8). The mean age of 20.2 years (SD = 1.4) applies to the entire student/student-tutor cohort (n = 94).
Survey instrument and procedure
The survey was administered in Ukrainian via Google Forms. The authors developed the items from the study objectives; three EFL colleagues reviewed them for face validity and clarity before distribution. Seven indicators covered AI awareness, Twee experience, AI-use frequency, preparation-time savings, task variety/differentiation, the main constraint, and future use/recommendation. Respondents selected one main constraint, not every difficulty experienced. These distinct constructs were not combined into a psychometric scale. Table 1 reports their distributions.
No standardized live demonstration or practical training session was provided. Non-users rated Twee on the basis of a brief description embedded in the survey, outlining functions available in early 2026: generating graded reading texts, creating comprehension questions from videos, and designing grammar exercises. Sixty-three respondents (42.0%) had never used or were unfamiliar with Twee. Non-users’ and students’ ratings are interpreted as anticipated usefulness rather than experience-based evaluations of teacher preparation; some students nevertheless had prior platform experience. Student responses provide a contrasting stakeholder perspective, not a direct measure of teachers’ workload.
Interviews
The six-question guide covered AI familiarity, anticipated benefits, curriculum integration, assessment, engagement, and the future teacher role. The lead researcher conducted individual Zoom interviews in Ukrainian, lasting 25–40 minutes. Interviews were audio-recorded with explicit consent and transcribed verbatim; participants were coded N1–N12. Selected quotations were translated into English and checked against the original Ukrainian audio by a bilingual researcher for conceptual accuracy. Prompts about future use elicited expectations about AI broadly, rather than verified platform capabilities.
Analysis
Counts and percentages describe the survey. Four exploratory role-group contrasts were recalculated from reported 2 × 2 counts using two-sided Fisher exact tests. Sample odds ratios (ORs) for teachers relative to students/student-tutors are accompanied by conditional exact 95% confidence intervals. Holm adjustment controls family-wise error across the four contrasts; unadjusted and adjusted p values are reported. Calculations were verified in Python 3.12.13 with SciPy 1.17.0. Aggregate counts did not permit adjustment for prior AI experience or institutional differences.
The first author conducted interpretive thematic analysis of 30 selected English-translated extracts (five per question), developing themes iteratively following Braun and Clarke (2006). JD-R/SDT concepts were applied post hoc to interpret workload and relational factors. Psychological constructs were not measured and the framework was not empirically tested. Analysis did not cover the full Ukrainian-language corpus; no claims are made of saturation, independent double-coding, or inter-rater reliability.
Ethical considerations
H. S. Skovoroda Kharkiv National Pedagogical University Faculty of Mathematics, Informatics and Natural Sciences Education Ethics Committee, Protocol No. 1, 16 January 2026. Procedures followed the Declaration of Helsinki and comparable institutional ethical standards.
Informed consent: Survey participants provided digital informed consent before accessing the questionnaire. Interview participants provided separate, explicit written consent for audio recording and publication of anonymized quotations. Consent records are stored on a password-protected institutional drive.
Data availability: The raw datasets are not publicly available because participants’ informed consent did not include permission for open data sharing and the data contain potentially identifiable qualitative extracts.
Declaration of Generative AI and AI-Assisted Technologies in the Writing Process
During manuscript preparation, the authors used ChatGPT (OpenAI) for restructuring, English-language editing and literature identification and formatting. The authors retain full responsibility for the analyses, references, interpretations, and final text.
Results
Survey profile and overall perceptions
High or moderate AI awareness was reported by 121 participants (80.7%). Eighty-seven (58.0%) had tried Twee, but only 12 (8.0%) used it regularly. Broader AI use was more established: 59 (39.3%) reported at least weekly use and 42 (28.0%) monthly use.
Overall, 77.3% rated Twee very or somewhat useful for preparation-time saving, and 82.7% for task variety/differentiation. Access was the leading constraint (27.3%), followed by quality assurance (22.0%), curriculum alignment and training/time (20.0% each), and privacy (10.7%). Most respondents (74.7%) intended to use or recommend it next semester (Table 1).
Table 1
Complete Survey Response Distributions (N = 150)
| Measure | Response Distribution, n (%) |
|---|---|
| Prior AI awareness |
|
| Experience with Twee |
|
| AI-tool use for learning/teaching |
|
| Usefulness for saving preparation time |
|
| Usefulness for task variety/differentiation |
|
| Main implementation constraint |
|
| Intention next semester |
|
Exploratory role-group comparisons
Teachers reported more prior Twee use and more often selected the highest usefulness categories for preparation-time saving and task variety. These differences remained significant after Holm adjustment. The access contrast was inconclusive (exact p = .058). Table 2 gives the counts, effect estimates, and confidence intervals; differences may reflect responsibilities and prior exposure as well as role.
Table 2
Exploratory Teacher and Student/Student-Tutor Comparisons
| Outcome | Teachers, n/N (%) | Students/Student-Tutors, n/N (%) | OR (95% Exact CI) | p | Adjust ed p |
|---|---|---|---|---|---|
| Any prior Twee use | 42/56 (75.0) | 45/94 (47.9) | 3.27 (1.50–7.32) | .001 | .005 |
| “Very useful” for time saving | 30/56 (53.6) | 28/94 (29.8) | 2.72 (1.29–5.72) | .005 | .016 |
| “Very useful” for task variety | 32/56 (57.1) | 34/94 (36.2) | 2.35 (1.13–4.89) | .017 | .034 |
| Access as main constraint | 10/56 (17.9) | 31/94 (33.0) | 0.44 (0.18–1.05) | .058 | .058 |
Interview themes
Theme 1: Anticipated workload relief could create room for individual support. N7 expected AI to “take over mundane and repetitive tasks” and associated released time with personalized learning and direct interaction. N12 called such technologies “partners in education,” emphasizing assistance with materials and strategies. These accounts locate value in what teachers might do with time saved; they do not demonstrate observed reductions in working time.
Theme 2: Adoption requires learning and negotiated use. N6 anticipated a learning curve for teachers and students. N2 wanted a platform that “matches my curriculum standards and educational goals.” N10 anticipated difficulty balancing technology with established methods. N8 raised access to devices and dependable internet, and N11 raised student privacy. These accounts identify implementation conditions without establishing anxiety, impaired competence, or clinical distress.
Theme 3: Guidance and connection define a continuing teacher role. N8 anticipated helping students use tools and interpret feedback through “guidance and mentoring.” N11 identified “empathy, motivation, and personal connection” as indispensable, alongside attention to students’ circumstances. These hopes coexisted with reservations: N7 anticipated distraction when technology takes priority over learning, and N11 called for balance in screen use.
Mixed-methods integration
Time-saving ratings align with N7’s anticipated workload relief, training constraints with N6’s learning demands, and access concerns with N8’s account. Mentoring gives possible time savings a relational purpose. Table 3 translates these observations into candidate work-design responses relevant to mental health promotion. Neither the survey nor the extracts establish associations between AI exposure and mental-health outcomes.
Table 3
Perceived Work Conditions and Implications for Mental Health Promotion
| Empirical Observation | Psychological Interpretation | Proposed Organizational Response | Future Assessment |
|---|---|---|---|
| Time-saving ratings; N7 on individual support | Possible relief from time pressure | Evaluate how saved time is used | Net task time; recovery; well-being |
| Curriculum fit; N2 and N10 on appropriate use | Discretion and autonomy | Preserve teacher choice and final review | Perceived autonomy |
| Training/time constraint; N6 on learning | Mastery and competence | Allocate practice time and feedback | Need satisfaction; AI self-efficacy |
| N8 on mentoring; N11 on human connection | Relatedness and work meaning | Protect human feedback and peer contact | Relatedness; meaningful work |
| Access constraint; N8 on infrastructure | Resources enabling participation | Provide access and equivalent alternatives | Access failures; support |
Discussion
The contribution concerns working conditions that warrant attention when universities integrate AI. Expectations of efficiency coexist with checking, learning, and access requirements. Teachers’ stronger usefulness ratings may reflect responsibilities and prior experience; they cannot indicate better mental health. A JD-R interpretation makes workload redistribution and resources the starting point for prevention-oriented inquiry (Bakker & Demerouti, 2017; Chuang et al., 2025).
Workload and recovery as mental health priorities
Usable drafts could reduce preparation pressure, but correction, after-hours learning, or higher output targets may absorb savings. Evaluation should examine total preparation/checking time, work extending into personal time, and opportunities to disengage. N7’s expectation of more student support raises the possibility that time saved becomes an unlimited caring obligation. Protecting recovery time is a proposed response, not an observed benefit of Twee, consistent with WHO’s organizational approach (World Health Organization, 2022).
Autonomy and competence during technological change
N2’s concern with curriculum fit makes professional discretion relevant. SDT defines autonomy through volition (Ryan & Deci, 2000). We propose allowing teachers to question, revise, or decline unsuitable output without penalties for low uptake. Autonomy satisfaction and psychological distress were not assessed.
N6’s anticipated learning curve highlights competence support. Practice, feedback, and paid learning time may make responsibilities manageable; compulsory learning outside working hours adds demands. Workplace SDT distinguishes autonomous from controlled motivation (Deci et al., 2017). Joint review of generated materials could support teachers’ judgment and normalize requests for help.
Social connection and professional identity
N8 and N11 emphasized mentoring, empathy, and personal connection. These accounts suggest a possible relational resource during change, although they express beliefs about teaching rather than measured protection against distress. AI-assisted preparation could leave room for dialogue, or accompany reduced contact if automated feedback displaces interaction. Evaluation should examine the quality of human contact alongside efficiency.
Professional development involves a coherent teaching identity. Kelchtermans (2009) connects professional self-understanding with vulnerability and reflection. Peer discussion may help teachers negotiate responsibilities, but these accounts do not demonstrate identity transitions or developmental stages.
Gast et al. (2022) explored perceived links between professional development and well-being among ten new university teachers. Peer groups could help teachers discuss AI-related difficulties and share checking work; their mental-health benefits in EFL teaching require evaluation.
Institutional prevention and psychosocial support
Access was the leading constraint in the mixed sample; the teacher–student contrast was inconclusive. In wartime remote teaching, electricity or internet interruptions could compound access barriers, but outages were not separately measured and their contribution remains untested. Devices and low-bandwidth alternatives are practical priorities; any additional support workload for teachers requires assessment.
Quality procedures should address accuracy, task level, bias, accessibility, and assessment alignment. Shared task banks could distribute checking work, consistent with UNESCO’s emphasis on human oversight (UNESCO, 2023, 2024). Privacy arrangements should cover student data, accounts, and queries (Colonna, 2024); fewer selections as the main constraint do not imply lesser importance.
Mental health promotion links work design with cooperation between managers, teaching-support teams, and mental-health professionals. Manager training and confidential professional support can complement organizational prevention (World Health Organization, 2022). Technical training cannot substitute for care, and reluctance to use AI does not indicate a mental-health problem.
A pilot could involve teachers in workload decisions, allocate paid learning time, protect recovery, and provide technical and psychosocial assistance. Evaluation should assess net workload and barriers to seeking help. These proposals draw on the WHO/ILO framework; no preventive intervention was evaluated (World Health Organization & International Labor Organization, 2022).
Limitations of the Study
This exploratory study used a regional, non-probability sample, predominantly students/student-tutors. Self-reports and the cross-sectional design limit generalizability and causal inference. Aggregate counts prevent adjustment for exposure, age, and institutional differences. Non-users and students evaluated anticipated usefulness without a standardized demonstration; these ratings do not establish objective time savings. Mental health, psychological distress, well-being, burnout, technostress, and psychological needs were not measured. Implementation constraints cannot establish the prevalence or severity of psychosocial risks. Qualitative analysis covered 30 selected translated extracts and one analyst, not the full Ukrainian-language corpus. Anticipatory prompts and selection may favor optimistic accounts. The mental-health framing is post hoc: psychological pathways and proposed preventive actions are interpretive propositions, not demonstrated mechanisms or effective interventions. Platform functions and policies may change.
Practical Value
The findings can guide university managers, EFL departments, and staff-support services in planning AI implementation. Priorities include paid practice time, shared review of generated materials, reliable access and low-bandwidth alternatives, and teacher discretion over tool use. Any time saved should support recovery and student interaction without automatically increasing workload. Cooperation with staff-support services can provide confidential routes to psychosocial assistance. These measures address reported implementation concerns and may support equitable participation; their effects on workload and mental health require evaluation.
Conclusions and Future Work
Teachers’ accounts and survey perceptions identify anticipated workload relief alongside learning, checking, and access demands, with continuing value placed on empathy and mentoring. Their relevance to workplace mental health lies in the organization of AI-related work: who carries new responsibilities, how much discretion teachers retain, and whether time for recovery and human connection is protected. Universities can use these concerns to guide participatory work-design reviews and coordination of technical and psychosocial support. The study does not establish that AI improves mental health or prevents burnout. Its contribution is a set of empirically informed priorities for mental health promotion that require direct, longitudinal evaluation.
Future research
Longitudinal research should assess mental health directly, using validated well-being and psychological-distress measures alongside work demands, resources, and net task time. It should separate actual from anticipated use and account for baseline mental health, prior experience, and institutional clustering. Repeated interviews could examine recovery, relationships, and professional identity. Evaluations of organizational support should prespecify outcomes and examine unintended workload increases.
Notes
[4] Funding Statement
The authors received no financial or other support from any organization for the submitted work.
[5] Project development statement
This article was prepared as part of the development of a grant proposal entitled “WELL-LANG AI: Human-Centered Generative AI for Language Pedagogy and Educator Well-Being”, intended for submission to the Erasmus+ program in 2027. No Erasmus+ funding was received for the research reported in this article.
[6] Conflict of Interest
The authors declare that they have no financial or non-financial competing interests relevant to this article and no financial or proprietary interests in any materials discussed.
[7] Authors' Contributions
Liliya Batyuk: Conceptualization (lead); Methodology (lead); Formal analysis (lead); Data curation; Project administration (lead); Writing – original draft (lead); Writing – review & editing.
Volodymyr Borysov: Conceptualization (supporting); Methodology (supporting); Supervision (lead); Project administration (supporting); Writing – review & editing.
Iryna Hlazkova: Conceptualization (supporting); Methodology; Supervision (supporting); Validation; Writing – review & editing.
Yana Romantsova: Methodology (supporting); Investigation; Data curation (lead); Writing – original draft (supporting); Writing – review & editing.
Olena Karasova: Methodology (supporting); Writing – review & editing.
Oksana Dubrova: Investigation (supporting); Data curation (supporting); Writing – review & editing.
Olena Ronzhes: Conceptualization (supporting, with a focus on psychological interpretation and occupational well-being); Writing – review & editing (supporting).
[8] Accountability Statement:
All authors attest that they meet the internationally accepted criteria for authorship. All authors have critically reviewed and approved the final manuscript and agree to be accountable for all aspects of the work, ensuring that any questions related to the accuracy or integrity of any part of the manuscript are appropriately investigated and resolved.