1. Introduction and problem

Faculties of education have installed a low-cost curricular gesture: an artificial-intelligence module, often housed in the educational-technology course, in which student teachers learn to “talk to” a generative model. The workshop product — a prompt, a generated lesson plan, a rubric, a worksheet — is exhibited as evidence of professional preparation. The enunciative leap is large. It moves from fluency with an interface to the claim that the future kindergarten and early-grades teacher is ready to teach three- to eight-year-olds in an environment saturated with opaque, unstable and social systems (Mishra et al., 2023). That leap is not authorised by the evidence of initial teacher education.

The thesis of this article is restrictive. An AI module does not prepare the early-childhood teacher if it only teaches prompts. The prompt is an act of technological knowledge: knowing how to formulate a request, iterate, copy and paste. Kindergarten teaching is another kind of knowing: noticing what a four-year-old does with a material, deciding whether an activity is developmentally pertinent, holding a group, improvising when the plan collapses, and — now — judging whether a model-generated text is didactically admissible or an adult artefact disguised as a children’s lesson. Confusing the two operations is a category error. It is the same error Trust, Whalen and Mouza (2023) describe when they recall that teacher-education programmes have siloed technology in a single course for decades, and that AI, precisely because it cuts across every subject, does not admit that isolation.

The problem is sharper in early childhood for an age-and-craft reason that generic “ChatGPT for teachers” workshops do not address. Miao and Holmes (2023) recommend a human-centred approach, ethical and pedagogical validation of tools, and an age threshold for independent conversations with generative platforms. Six expert professors in early childhood from China and the United States locate ChatGPT, at best, as an on-call facilitator for the educator and, with reservations, as a conversational agent for the child; its desirable trajectory is human intelligence augmentation, not replacement of the teacher’s action (Luo et al., 2024). None of that is taught when the module is exhausted in prompt engineering. Nor is it taught when the success indicator is technological acceptance — intention to use — rather than judgement on a circle-time, centre or emergent-literacy plan.

This article does not recycle UNESCO’s AI competency framework for in-service teachers (Miao & Cukurova, 2024): that instrument describes profiles of those who already teach. The question here is prior: what happens in initial teacher education when a kindergarten teacher is said to be “prepared” by an AI module. The contributions are three: to reconstruct the state of the art by separating TK, TPK and judgement (Celik, 2023; Mishra et al., 2023; Sperling et al., 2024); to examine three families of preservice cases — acceptance, literacy courses, and lesson planning; and to offer four tests for when a teacher-education design may claim that it “prepares.”

2. State of the art: from acceptance to judgement, from module to curriculum

Four strata that the literature often mixes should be kept apart. The first is technology acceptance: whether the future teacher declares that they will use AI. The second is professional knowledge for integrating AI into teaching, now formulated as TPACK and Intelligent-TPACK. The third is the curricular place of that knowledge in initial teacher education: a siloed course or a thread through the programme and practicum. The fourth is the specific craft of early childhood, where the interlocutor is not an adolescent writing essays but a young child whose independent conversation with a generative model is not authorised by UNESCO guidance (Miao & Holmes, 2023).

In the acceptance stratum, Zhang et al. (2023) find that perceived usefulness and ease of use predict intention to use educational AI applications. Runge, Hebibi and Lazarides (2025) extend the model with AI-related courses and a self-report of AI-TPACK: courses associate with both; AI-TPACK does not significantly predict intention to use AI in future teaching. Accepting, self-rating knowledge and declaring intention are not equivalent to knowing how to teach a group of five-year-olds.

In the professional-knowledge stratum, Celik (2023) models Intelligent-TPACK with 428 teachers in Turkey who used a platform with dashboards and tutors. Although the sample is in-service, the result transfers as a knowledge structure, not as a cohort equivalence: technological knowledge (TK) allows better assessment of AI decisions, but “only TK is not sufficient”; TK is meaningful when combined with pedagogical knowledge, reflected in technological pedagogical knowledge (TPK); ethical assessment influences TPACK as much as TPK does (Celik, 2023). Teaching how to interact with the tool — the territory of the prompt — is not teaching how to decide. Mishra, Warr and Islam (2023) add that generative systems are protean, opaque and unstable, and also generative and social. Contextual knowledge (XK) must widen beyond the immediate classroom. A module that never touches that opacity forms a consulting habit, not TPACK.

Sperling, Stenberg, McGrath, Åkerfeldt, Heintz and Stenliden (2024) review 34 papers (2000–2023) on AI literacy relevant to teacher education, using episteme, techne and phronesis. The topic is globally emergent and “almost absent” from the field of teacher education. Ethics are treated as technical configurations; practical knowledge translates into adopting resources or AI EdTech. Phronesis — situated professional judgement — remains outside. A prompt module operates in a narrow techne and leaves judgement untouched. Ng, Leung, Su, Ng and Chu (2023) adapt DigCompEdu and P21 to teachers’ AI digital competencies: a framework, not evidence of implementation in early-childhood teacher education. Kasneci et al. (2023) stress that teachers need training and that models do not replace the teacher. Celik, Dindar, Muukkonen and Järvelä (2022) already documented opportunities for planning, implementation and assessment, and limits of reliability and applicability; technically and pedagogically capable systems in diverse settings are “yet to be achieved.” Chiu, Xia, Zhou, Chai and Cheng (2023) identify thirteen roles, seven outcomes and ten challenges, and insist on professional development. No synthesis claims that a prompt workshop closes the challenge.

In the curricular stratum, Trust et al. (2023) recommend examining AI organically across the whole initial-teacher-education plan, not in a single course. Van den Berg and du Plessis (2023) show that ChatGPT can deliver materials and “level” access to plans, and they demand caution: the models “do not replace teachers.” The OECD Digital Education Outlook 2023 locates teachers’ digital competences in the system’s digital ecosystem, not in a workshop annex (OECD, 2023). UNESCO’s Recommendation on the ethics of AI anchors human rights and human oversight (UNESCO, 2021). The 2023 Guidance adds pedagogical validation and the age threshold (Miao & Holmes, 2023): a child aged three to six is not an autonomous ChatGPT user; the early-childhood teacher cannot be formed as if they were. Luo et al. (2024) synthesise interviews with six expert professors in early childhood: conversational agent for the child and on-call facilitator for the educator; 3A2S challenges (accessibility, affordability, accountability, sustainability, social justice); desirable trajectory, intelligence augmentation, not replacement. Inference, marked as such: if the value lies in the teacher’s agency, initial teacher education cannot be reduced to the agency of the prompt.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate a homogeneous effect size across TAM surveys, mixed-methods courses and planning analyses, but to articulate a curricular-policy argument with verified sources. Inclusion criteria: (a) 2021–2026; (b) initial teacher education (preservice, teacher education) or, if the sample is in-service, conceptual and flagged use (Celik, 2023; Celik et al., 2022); (c) AI, generative models or TPACK applied to AI; (d) relevance to early or basic education, or to the teacher-education plan that prepares those teachers; (e) peer-reviewed journal, DOI, or UNESCO or OECD report; (f) verifiable DOI or publisher page. Axes already used in this series (play and PopBots, parental mediation, inclusion, adaptive tutors, gaps in learning AI, student AI literacy, classroom privacy, multilingualism, teacher wellbeing, integrity and detectors, pedagogical documentation, low connectivity) were excluded, as was, as a governing frame, Miao and Cukurova’s (2024) in-service competency document.

The search was executed on 24 August 2026 on DOI pages, SpringerOpen, Elsevier, Emerald, Wiley, SAGE, MDPI, CITE Journal, UNESDOC, OECD iLibrary and repositories (Oulu, DiVA, East China Normal). Each source was checked against at least one of those pages. The corpus was organised into acceptance, module or course, and lesson planning. Saturation retained Celik (2023), Sperling et al. (2024), Mishra et al. (2023), Trust et al. (2023), van den Berg and du Plessis (2023), Luo et al. (2024), Miao and Holmes (2023), UNESCO (2021), Ng et al. (2023), Kasneci et al. (2023), Celik et al. (2022), Chiu et al. (2023) and OECD (2023).

The analysis distinguished three enunciative statuses. Empirical finding: what was observed or measured in the preservice sample (or, if in-service, so declared). Conceptual or normative frame: what a knowledge framework, a guidance or a recommendation prescribes. Pedagogical inference: the translation this article proposes for the design of kindergarten and early-grades teacher education, marked as such. The limits are those of any narrative review: no full PRISMA protocol, language bias toward English and German, and a scarcity of Latin American preservice-and-AI trials in early-childhood teacher education (section 9).

4. Case 1. Accepting is not preparing: 452 future teachers and the ceiling of TAM

Zhang, Schießl, Plößl, Hofmann and Gläser-Zikuda (2023) published in International Journal of Educational Technology in Higher Education (vol. 20, art. 49) a multigroup analysis of AI acceptance among preservice teachers. At one German university, a complete dataset of 452 participants was obtained from 712 potential respondents (63.48%), of whom 325 were women (71.90%). Programmes included primary school education (n = 260), lower-school education (n = 44) and other teacher-education tracks. Data were collected in winter semester 2021/2022, before ChatGPT’s public mass-consumer eclosion, which matters so that the study is not read as a generative-prompt trial. The model was TAM3. Eight of nine hypotheses were supported. Perceived usefulness (β = 0.501, p < 0.001) and perceived ease of use (β = 0.297, p < 0.001) were the primary predictors of intention to use educational AI applications; usefulness weighed more. Latent mean differences by gender were significant for AI anxiety (z = −3.217, p < 0.01) and perceived enjoyment (z = 2.556, p < 0.05). Gender moderated the paths from anxiety to ease of use (p = 0.018) and from ease of use to usefulness (p = 0.002). The authors stress the need to address gender-specific aspects in teacher education given the female majority of cohorts (Zhang et al., 2023).

Status of the evidence. Empirical finding: in that one-university sample, with that instrument and at that moment, intention to use AI is explained mainly by how useful and how easy the tool is perceived to be. It is not a finding of didactic competence. The study does not observe practice with children, does not evaluate lesson plans, does not measure TPACK and does not disaggregate the SEM by programme beyond sample composition. The 260 primary-education students are the subset closest to this article’s object. Pedagogical inference, marked as such: a dean’s office that cites this kind of survey to claim that “our students accept AI, therefore they are prepared for kindergarten” makes the leap the thesis forbids. Acceptance is a possible precursor of use; it is not professional preparation. Ease of use — the phenomenological territory of a well-written prompt — is the weaker of the two main predictors, and still does not equal TPK.

Conceptual saturation comes from Intelligent-TPACK. Celik (2023) shows, in 428 Turkish teachers aged 29–38 (258 women), that TK predicts ethical assessment but not TPACK; TPK is the bridge. Transferring that result to teacher education is inference, not sample equivalence: they already taught and used a state platform. The value is the structure of knowing. A module that trains TK and celebrates intention to use leaves TPK and the ethics of the decision unformed.

5. Case 2. The AI course raises confidence and does not authorise “already prepared”

Hur (2024) published in Information and Learning Sciences (vol. 126, nos. 1/2, pp. 56–74) a mixed-methods study of AI-literacy lessons integrated into a technology-integration course for preservice teachers, with the Concerns-Based Adoption Model as the frame for concerns. Participants initially lacked AI knowledge and awareness. Targeted lessons increased awareness and confidence in teaching AI. At the same time, after the lessons, concerns persisted: fear of teacher displacement and adverse effects of generative AI on students’ critical learning-skill development (Hur, 2024). The study presents itself as one of the scarce empirical investigations of AI literacy in preservice; its own diagnosis of scarcity coincides with Sperling et al. (2024).

Status of the evidence. Empirical finding: a module nested in the technology course can move self-reported awareness and confidence, and does not dissolve craft concerns. It is not a finding that participants know how to teach in a kindergarten. The course is precisely the silo Trust et al. (2023) criticise: AI is taught in the technology subject, not in language methods, professional practice or child development. Hur does not claim that post-module confidence equals preparation; this article does not either. Pedagogical inference: a rise in confidence is a workshop result, not a criterion of professional licensure. A more confident kindergarten teacher equally unable to judge a generated plan is not better prepared; they are more willing to deploy an unexamined artefact.

Runge et al. (2025) provide the most precise quantitative saturation for this thesis. In Education Sciences (vol. 15, art. 167) they model paths with 143 preservice teachers at German universities (majority from Brandenburg, 42%, and Berlin, 13%; 56% bachelor’s and 42% master’s; summer-semester 2024 data). Participation in AI-related courses associated positively with AI-TPACK (β = 0.449, p < 0.001) and perceived usefulness (β = 0.235, p < 0.001), not with ease of use (β = 0.025, p = 0.808). AI-TPACK associated with usefulness (β = 0.260, p = 0.032) and ease of use (β = 0.417, p < 0.001). Perceived usefulness strongly predicted intention to use AI in future teaching (β = 0.722, p < 0.001) and, with smaller magnitude, actual AI use for profession-related tasks in teacher training (β = 0.266, p = 0.004). Ease of use predicted use in training (β = 0.362, p < 0.001), not future teaching intention (β = −0.061, p = 0.518). The path from AI-TPACK to intention to use in future teaching was not significant (β = −0.045, p = 0.690); nor was the path from AI-TPACK to use in training (β = 0.165, p = 0.169) (Runge et al., 2025).

Status of the evidence. Empirical finding: in that sample, taking AI courses associates with a higher self-report of AI-TPACK and with perceiving AI as useful; perceiving it as useful predicts wanting to use it when teaching; AI-TPACK, by contrast, does not predict that intention. Measured “actual use” is not use with children in kindergarten: it is use for tasks of the teaching degree. It is not a finding of transfer to the three-to-six classroom. It is not a finding of didactic quality of the output. Pedagogical inference, marked as such: the most uncomfortable datum for module policy is that non-significant coefficient. A programme may certify that its students “have AI-TPACK” by scale and still not have moved pedagogical intention to use nor, a fortiori, the capacity to use it well. The course produces perceived knowledge and usefulness; it does not produce, in this model, the criterion deans usually announce.

Read with Hur (2024), the pattern is coherent. The module changes subjective states and leaves craft concerns and practice with children intact. Sperling et al. (2024) anticipate this: without phronesis, the course forms interface techne.

6. Case 3. One prompt does not make a kindergarten lesson: 59 and 27 future teachers facing the generated plan

Kalenda, Rath, Abugasea Heidt and Wright (2025) published in Journal of Educational Technology Systems (vol. 53, no. 3, pp. 219–241) a pretest–posttest study with 59 undergraduate and graduate students enrolled in STEM, TESOL and social-studies methods courses for grades PreK–12. The procedure was a guided analysis of ChatGPT’s ability to write lesson plans, with a survey and a reflection on strengths and weaknesses. After the analysis, perceptions that ChatGPT can write a complete plan in the content area decreased statistically significantly, t(58) = 2.039, p = 0.046. Perceptions that it can write a plan detailed enough for a substitute teacher fell more strongly, t(58) = 5.959, p < 0.001. Participants stated that they plan to use ChatGPT for planning when they practise, and indicated that revision of the output is necessary (Kalenda et al., 2025).

Status of the evidence. Empirical finding: in that PreK–12 methods sample, a single prompt and a raw plan do not withstand the candidate’s own scrutiny once they are asked to analyse the content. The drop in the perception of “ready for a substitute” is the datum closest to kindergarten: a plan that could not be left with an adult substitute is not a classroom plan. It is not a finding that subsequent revision produces a quality plan; the study measures perceptions, not the quality of revised plans against a child-development rubric. It is not a randomised trial. The PreK–12 range mixes kindergarten with secondary; the value for this article is that the guided-analysis procedure — not the isolated prompt — is what moves judgement, and it moves it downward.

Sun and Huang (2025) published in Journal of Computer Assisted Learning (vol. 41, no. 4, e70098) a mixed study with 27 preservice teachers in a STEM unit of a graduate educational-technology course at East China Normal University. They analysed interactions with epistemic network analysis (ENA) and, qualitatively, final plans and generative-AI logs. GenAI functioned mainly as a direct information source for operational tasks; its role as cognitive scaffolding was limited. It significantly supported initial planning phases, with higher-order thinking (analysing, creating), and its contribution diminished in later iterative phases. The authors underline the need for structured scaffolds and human facilitation to complement GenAI and sustain deeper cognitive engagement (Sun & Huang, 2025).

Status of the evidence. Empirical finding: in that graduate sample, the model does not sustain pedagogical work when it becomes iterative — precisely when a kindergarten plan should be adjusted to a concrete group, a centre, a child who does not enter the activity, a developmental goal the model has not observed. It is not an early-childhood finding: the unit is STEM in an educational-technology course. Transfer to kindergarten is a risk inference, not an equivalence: if cognitive scaffolding already weakens in the iteration of a graduate STEM plan, an early-childhood candidate who receives a “ready” circle-time or emergent-literacy plan and has no human teacher educator in the loop is more exposed, not less.

Van den Berg and du Plessis (2023) saturate: the model delivers materials and does not replace the teacher. Trust et al. (2023) require modelling critical evaluation of the output. Luo et al. (2024) locate the value in human agency. Inference: the prompt opens a working file, it does not close a competence. A module that assesses the prompt and not judgement on the plan — age-appropriateness, sequencing, orality, observation — does not form the kindergarten teacher. It forms an interface operator who, after looking at the output, no longer believes the plan would suffice for a substitute (Kalenda et al., 2025).

7. Inferential frame: four tests for claiming that initial teacher education “prepares”

The frame that follows is this article’s pedagogical inference, anchored in the cases and in the verified instruments. It is not a new international standard. It distinguishes four tests. If a teacher-education design does not pass them, it cannot declare that an AI module prepares the early-childhood teacher.

7.1. Test of professional judgement, not of interface fluency. Sperling et al. (2024) show that phronesis is almost absent from the AI-literacy literature for teacher education. Celik (2023) shows that TK does not suffice and that TPK and ethics are the bridges. Kalenda et al. (2025) show that guided analysis of the output lowers the illusion of completeness. Sun and Huang (2025) show that the model does not scaffold iterative phases. Inference: the evidence of preparation is not a prompt portfolio nor an ease-of-use score (Zhang et al., 2023). It is the capacity, observed in practicum or in dense simulations, to accept, reject or rewrite a generated artefact with didactic reasons. The module’s success criterion, if the module exists, is judgement, not generation speed.

7.2. Test of curricular integration, not of a technological silo. Trust et al. (2023) reject the single course. Hur (2024) illustrates the typical design: AI lessons inside technology integration. Runge et al. (2025) measure use in degree tasks, not in the children’s classroom. OECD (2023) locates teachers’ digital competences in the system’s ecosystem, not in an annex. Inference: AI must appear in language methods, child development, professional practicum and work with the field supervisor. An optional credit in “AI for teachers” does not meet the test, even if it raises self-reported AI-TPACK. The place of the knowledge is the whole programme, because the generative model touches every teaching decision (Mishra et al., 2023).

7.3. Test of fit to early childhood education, not of a generic K–12 lesson. Miao and Holmes (2023) require pedagogical validation and an age threshold for independent use of generative platforms. Luo et al. (2024) distinguish the educator-facilitator role from the child-conversational-agent role, and privilege augmentation of human agency. A plan generated for “grade 2” is not, by existing, a four-year-old classroom plan. Inference: early-childhood teacher education must train detection of age-inappropriateness — adult language, sedentary activities, cognitive objectives from another stage, absence of observation, of body, of orality. That detection is early-childhood PCK mediated by technology, not a prompt trick. The generic “teaching with AI” module fails this test by construction.

7.4. Test of irreplaceable human mediation in the training itself. Sun and Huang (2025) require human facilitation and scaffolds. Van den Berg and du Plessis (2023) and Kasneci et al. (2023) deny replacement of the teacher. Celik et al. (2022) conclude that technically and pedagogically capable systems in diverse settings are yet to be achieved. UNESCO (2021) requires human oversight. Inference: the teacher educator — the methods professor, the practicum tutor — is the condition of preparation, not a cost to be optimised with licences. A design in which the candidate “learns AI” by talking only to the model, without human contradiction, reproduces in training the error one wants to avoid in kindergarten: leaving the pedagogical decision to a system with no human in the loop (Miao & Holmes, 2023).

The frame admits models in planning with guided analysis and age-appropriateness rubrics (Kalenda et al., 2025); AI in methods and practicum, not only in technology; assessment of TPK and of decision ethics, not only of TK (Celik, 2023); and educator-of-educators formation. It rejects certifying preparation by attendance at a prompt module; using intention to use or self-reported AI-TPACK as a proxy for kindergarten competence (Zhang et al., 2023; Runge et al., 2025); authorising independent conversation of early-childhood children with generative platforms (Miao & Holmes, 2023); and substituting the practicum tutor with a feedback chatbot (Luo et al., 2024).

8. Discussion

Three tensions organise the discussion. The first is between acceptance and preparation. It is a finding that perceived usefulness predicts intention to use, with 260 primary-education students in the German sample (Zhang et al., 2023), and that usefulness, not AI-TPACK, predicts intention to use AI when teaching (Runge et al., 2025). It is a frame that TK does not suffice (Celik, 2023) and that phronesis is missing from teacher education (Sperling et al., 2024). It is not a finding that whoever accepts or self-rates knowledge knows how to teach in a kindergarten. Module policy measures what TAM knows how to measure and declares what only situated judgement would authorise.

The second is between the course that raises confidence and the craft that does not move. Hur (2024) documents more awareness and confidence, and persistent concerns. Runge et al. (2025) document AI-TPACK without prediction of pedagogical intention. Trust et al. (2023) had warned of the silo. A technology course does what it can do. It does not, by itself, form the kindergarten teacher. Declaring that it does is not an empirical result.

The third is between the generated plan and the early-childhood lesson. Kalenda et al. (2025) lower faith in the raw plan; Sun and Huang (2025), scaffolding in iteration; van den Berg and du Plessis (2023), the equivalence between access to plans and teaching. Luo et al. (2024) and Miao and Holmes (2023) locate the age limit and human agency. The prompt produces plausible text; kindergarten demands a decision about a body, a developing language and a group. Mishra et al. (2023) describe the model as social and generative. Treating it as a word processor in two weeks underestimates the technology and the craft. Models can save a draft (Kasneci et al., 2023); that saving, without judgement or practicum, is not preparation. OECD (2023) locates the problem in the ecosystem. Cutting the answer to a prompt workshop is not answering.

9. Limits

This review is narrative. It does not apply PRISMA or estimate pooled effects. The preservice cases have biased geographies (Germany, United States, China) and unequal n (452, 143, 59, 27; Hur, 2024, is not cited with n because the verified publisher abstract does not report it univocally). Zhang et al. (2023) measure acceptance before mass-consumer ChatGPT. Runge et al. (2025) measure use in training, not in kindergarten. Kalenda et al. (2025) cover PreK–12 and perceptions, not development rubrics. Sun and Huang (2025) work graduate STEM. Celik (2023) is in-service, used as a frame. No Latin American kindergarten-teacher-education and AI trials were located with the same degree of DOI; that absence is a gap. UNESCO and OECD are prescriptive. Section 7 inferences are design hypotheses, not implementation evidence.

10. Conclusions

An artificial-intelligence module does not prepare the early-childhood teacher if it only teaches prompts. The preservice evidence verified here does not authorise that policy declaration. Intention to use is predicted by perceived usefulness in 452 future teachers, 260 in primary education (Zhang et al., 2023). A literacy course raises awareness and confidence and leaves craft concerns standing (Hur, 2024). In 143 teacher-education students, courses associate with self-reported AI-TPACK, and that construct does not predict intention to use AI in future teaching; measured use is that of degree tasks (Runge et al., 2025). In 59 PreK–12 candidates, guided analysis of generated plans reduces belief in their completeness and, more strongly, in their usefulness for a substitute (Kalenda et al., 2025). In 27 future teachers, GenAI informs initial phases and does not scaffold iterative ones (Sun & Huang, 2025). Intelligent-TPACK, generative TPACK and the 34-paper review converge on the diagnosis: isolated technological knowledge does not suffice and professional judgement is not at the centre of teacher education (Celik, 2023; Mishra et al., 2023; Sperling et al., 2024).

The applicable guidance is not ambiguous for what matters in kindergarten. UNESCO requires ethics, human oversight, pedagogical validation and an age threshold for independent use of generative platforms (UNESCO, 2021; Miao & Holmes, 2023). Early-childhood experts locate ChatGPT as augmentation, not replacement (Luo et al., 2024). Where the sources do not measure didactic competence with three- to six-year-olds, this article does not affirm it. Preparing the kindergarten teacher is forming, in the programme of study and in practicum, the judgement that accepts, rejects or rewrites a generated artefact. Teaching prompts is another operation, and it is not enough.

Editorial Laboratory of NEXTECH.IA / Ingeniero Mitre.

References

  1. Celik, I. (2023). Towards Intelligent-TPACK: An empirical study on teachers’ professional knowledge to ethically integrate artificial intelligence (AI)-based tools into education. Computers in Human Behavior, 138, 107468. https://doi.org/10.1016/j.chb.2022.107468
  2. Celik, I., Dindar, M., Muukkonen, H., & Järvelä, S. (2022). The promises and challenges of artificial intelligence for teachers: A systematic review of research. TechTrends, 66(4), 616–630. https://doi.org/10.1007/s11528-022-00715-y
  3. Chiu, T. K. F., Xia, Q., Zhou, X., Chai, C. S., & Cheng, M. (2023). Systematic literature review on opportunities, challenges, and future research recommendations of artificial intelligence in education. Computers and Education: Artificial Intelligence, 4, 100118. https://doi.org/10.1016/j.caeai.2022.100118
  4. Hur, J. W. (2024). Fostering AI literacy: Overcoming concerns and nurturing confidence among preservice teachers. Information and Learning Sciences, 126(1/2), 56–74. https://doi.org/10.1108/ILS-11-2023-0170
  5. Kalenda, P. J., Rath, L., Abugasea Heidt, M., & Wright, A. (2025). Pre-service teacher perceptions of ChatGPT for lesson plan generation. Journal of Educational Technology Systems, 53(3), 219–241. https://doi.org/10.1177/00472395241301388
  6. Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., Krusche, S., Kutyniok, G., Michaeli, T., Nerdel, C., Pfeffer, J., Poquet, O., Sailer, M., Schmidt, A., Seidel, T., Stadler, M., Weller, J., Kuhn, J., & Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 103, 102274. https://doi.org/10.1016/j.lindif.2023.102274
  7. Luo, W., He, H., Liu, J., Berson, I. R., Berson, M. J., Zhou, Y., & Li, H. (2024). Aladdin’s genie or Pandora’s box for early childhood education? Experts chat on the roles, challenges, and developments of ChatGPT. Early Education and Development, 35(1), 96–113. https://doi.org/10.1080/10409289.2023.2214181
  8. Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  9. Mishra, P., Warr, M., & Islam, R. (2023). TPACK in the age of ChatGPT and Generative AI. Journal of Digital Learning in Teacher Education, 39(4), 235–251. https://doi.org/10.1080/21532974.2023.2247480
  10. Ng, D. T. K., Leung, J. K. L., Su, J., Ng, R. C. W., & Chu, S. K. W. (2023). Teachers’ AI digital competencies and twenty-first century skills in the post-pandemic world. Educational Technology Research and Development, 71(1), 137–161. https://doi.org/10.1007/s11423-023-10203-6
  11. OECD. (2023). OECD Digital Education Outlook 2023: Towards an effective digital education ecosystem. OECD Publishing. https://doi.org/10.1787/c74f03de-en
  12. Runge, I., Hebibi, F., & Lazarides, R. (2025). Acceptance of pre-service teachers towards artificial intelligence (AI): The role of AI-related teacher training courses and AI-TPACK within the Technology Acceptance Model. Education Sciences, 15(2), 167. https://doi.org/10.3390/educsci15020167
  13. Sperling, K., Stenberg, C.-J., McGrath, C., Åkerfeldt, A., Heintz, F., & Stenliden, L. (2024). In search of artificial intelligence (AI) literacy in teacher education: A scoping review. Computers and Education Open, 6, 100169. https://doi.org/10.1016/j.caeo.2024.100169
  14. Sun, Y., & Huang, Y. (2025). Exploring the role of generative AI in collaborative lesson planning for pre-service teachers. Journal of Computer Assisted Learning, 41(4), e70098. https://doi.org/10.1111/jcal.70098
  15. Trust, T., Whalen, J., & Mouza, C. (2023). Editorial: ChatGPT: Challenges, opportunities, and implications for teacher education. Contemporary Issues in Technology and Teacher Education, 23(1). https://citejournal.org/volume-23/issue-1-23/editorial/editorial-chatgpt-challenges-opportunities-and-implications-for-teacher-education
  16. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  17. van den Berg, G., & du Plessis, E. (2023). ChatGPT and generative AI: Possibilities for its contribution to lesson planning, critical thinking and openness in teacher education. Education Sciences, 13(10), 998. https://doi.org/10.3390/educsci13100998
  18. Zhang, C., Schießl, J., Plößl, L., Hofmann, F., & Gläser-Zikuda, M. (2023). Acceptance of artificial intelligence among pre-service teachers: A multigroup analysis. International Journal of Educational Technology in Higher Education, 20, 49. https://doi.org/10.1186/s41239-023-00420-7