Development and validation of a forward-looking metacognitive AI readiness scale for Jordanian teachers Authors: Haitham Salem Baibers, Mohammad Nayef Ayasrah Published: 2026 ABSTRACT With the rapid expansion of artificial intelligence (AI) in education, valid tools are needed to assess teachers’ readiness for anticipatory, adaptive, and reflective engagement with AI-supported instruction. This study developed and evaluated the Forward-Looking Metacognitive–AI Readiness Scale (FLMAIRS) among secondary school teachers in Jordan. An exploratory mixed-methods design was used. Items were generated through a literature review and semi-structured interviews with 22 teachers, followed by expert review, content and face validity assessment, and cognitive interviewing. The field-test version was administered online to 640 teachers and randomly split for exploratory factor analysis (EFA; n = 320) and confirmatory factor analysis (CFA; n = 320). EFA supported a five-factor, 45-item structure: AI Foresight Planning, Adaptive AI-Oriented Instructional Regulation, Algorithmic Anticipatory Reasoning, Human–AI Instructional Synergy, and Long-Term AI-Reflective Practice, explaining 58.29% of the variance. CFA supported acceptable first- and second-order model fit. Reliability was high across dimensions (Cronbach’s α = 0.897–0.931; McDonald’s ω = 0.900–0.932), and test–retest stability was good (ICC = 0.752–0.851). Convergent and discriminant validity were supported. FLMAIRS provides a context-sensitive instrument for assessing teachers’ forward-looking metacognitive readiness for AI-integrated education.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-59423-9. Introduction AI-based technologies from personalized learning systems to automated assessment tools have reshaped both learning interactions and the teacher’s role. These transformations are evident at the levels of educational policy and strategic redesign1, as well as in approaches to individual learning and self-regulation2. Teachers’ ability to understand, adapt to, and effectively utilize AI has therefore become a key professional competency3. However, the existing literature has primarily analyzed this competency in terms of technological skills, with relatively little attention paid to its metacognitive dimension. Grounded in the theoretical foundations of metacognition4 and supported by empirical findings on AI applications in education5, this ability enables teachers to critically reframe their strategies and optimize them in alignment with anticipated outcomes of interaction with intelligent technologies. Although research on AI in education has expanded rapidly, much of the existing measurement work has focused on teachers’ digital competence, technology acceptance, AI-related attitudes, or general readiness for technology use6, 7, 14, 15, 23. These domains are important, but they only partially address the capacities teachers need when working with AI-supported instructional systems. Less attention has been given to teachers’ ability to anticipate AI-related instructional changes, regulate teaching decisions through AI-generated feedback, and plan alternative educational responses under technological uncertainty1, 8, 9. This creates a measurement gap for a construct that is simultaneously metacognitive, AI-oriented, and future-focused. Previous instrument-development studies provide a useful foundation for this gap. For example, existing measures of teachers’ AI readiness and AI competence have advanced the assessment of teachers’ awareness, attitudes, self-efficacy, and practical skills for using AI in education34, 38, 39, 40. Similarly, technology acceptance and digital competence frameworks have clarified teachers’ willingness, perceived usefulness, and professional capacity to use educational technologies14, 15, 23, 35, 36. However, these instruments were not designed to measure the combined capacity to anticipate AI-related instructional change, regulate teaching through AI-generated feedback, design alternative future-oriented learning scenarios, and integrate professional judgment with algorithmic recommendations. This limitation indicates the need for a new instrument that is conceptually grounded in educational psychology and psychometrically validated for AI-integrated teaching contexts. To clarify this gap further, existing measures can be grouped into four related but distinct domains. First, digital competence and technology acceptance instruments mainly assess teachers’ technological ability, perceived usefulness, perceived ease of use, and willingness to adopt educational technologies14, 15, 23, 35, 36. Second, AI readiness and AI competence scales focus primarily on AI awareness, attitudes, self-efficacy, and practical skills for using AI tools in educational settings34, 38, 39, 40. Third, metacognitive and self-regulated learning measures assess monitoring and regulation processes, but they are not specifically designed for AI-supported instructional decision-making32, 37. Fourth, future-orientation and forward-looking assessment instruments address long-term thinking or data-informed learning support, but they do not directly assess teachers’ readiness to work with AI-generated predictions and recommendations41, 42. Therefore, the specific gap addressed in this study is the absence of a validated instrument that simultaneously measures teachers’ capacity to anticipate AI-related instructional change, regulate teaching through AI-generated feedback, design alternative future-oriented learning scenarios, and integrate professional judgment with algorithmic information. For the purposes of this study, Forward-Looking Metacognitive–AI Readiness is operationally defined as a multidimensional teacher capacity that combines anticipatory thinking, metacognitive regulation, AI-supported instructional decision-making, human–AI judgment integration, and long-term reflective adaptation. More specifically, it refers to the extent to which teachers can anticipate AI-related changes in teaching and learning, monitor and adjust instructional decisions using AI-generated information, design alternative future-oriented learning scenarios, balance algorithmic recommendations with professional judgment, and reflect on the longer-term instructional consequences of AI use. In measurement terms, this construct was operationalized through five observable dimensions that reflect teachers’ self-reported readiness to plan for AI-related change, adapt instruction using AI-supported information, reason with algorithmic outputs, combine human and AI-based judgment, and engage in long-term reflective practice. Study context: AI readiness and teacher professional practice in Jordan The need for a localized measure of Forward-Looking Metacognitive–AI Readiness is particularly important in educational systems where AI integration is still emerging and where teachers are expected to interpret, adapt, and contextualize digital innovations in practice. Jordan provides a relevant context for this purpose. In recent years, the country has invested in digital infrastructure and school smartization initiatives, creating initial conditions for the adoption of emerging technologies, including AI10, 11. However, the implementation of digital technology in Jordanian schools remains uneven. Challenges related to limited resources, unequal access to technology, infrastructure constraints, and variation in teachers’ digital competence continue to affect the extent to which AI-supported tools can be meaningfully integrated into teaching10, 12, 13. The educational culture in Jordan also gives teachers an important role in translating policy and technological change into classroom practice. Teachers are not only users of digital tools; they are often involved in selecting, adapting, and culturally contextualizing instructional resources for their students12. This is especially relevant in AI-integrated education, where teachers must evaluate algorithmic outputs, adapt recommendations to local classroom conditions, and balance technological possibilities with pedagogical judgment. In this sense, teacher readiness cannot be reduced to basic digital skill or general willingness to use technology. The Jordanian context also reflects broader regional concerns in the Middle East, where teacher professional development, digital competence, and readiness for educational innovation remain central issues in technology integration23. While regional and international frameworks provide useful guidance for teacher digital competence14, 23, there remains a need for instruments that capture the more specific capacities required for AI-supported teaching. These include the ability to anticipate AI-related instructional changes, regulate teaching through AI-generated feedback, design alternative learning scenarios, and reflect on the long-term consequences of AI use. For this reason, validating FLMAIRS among Jordanian teachers can provide context-sensitive evidence for professional development, school-level planning, and future comparative research in similar educational settings. Taken together, these developments point to a clear measurement need. Existing instruments provide valuable information about teachers’ digital competence, technology acceptance, AI literacy, or general readiness for technology use14, 15, 23, 34, 35, 36, 37, 38, 39, 40, but they do not fully capture the combined forward-looking, metacognitive, and AI-oriented capacities required for instructional decision-making. Accordingly, the present study aimed to develop and psychometrically evaluate the Forward-Looking Metacognitive–AI Readiness Scale (FLMAIRS) for AI-integrated education among Jordanian teachers. By doing so, the study seeks to provide a context-sensitive and psychometrically robust instrument for research, professional development, and educational policy planning. Theoretical framework To improve the conceptual clarity of FLMAIR, the construct was organized around four interrelated conceptual pillars rather than presented as a broad collection of theories: metacognitive monitoring and regulation, AI-supported instructional decision-making, educational foresight, and human–AI collaboration. These pillars draw on metacognition and self-regulated learning theory4, 16, 31, 32, technology acceptance, planned behavior, diffusion, and teacher digital competence frameworks14, 15, 20, 21, 22, 23, 33, future-oriented, scenario-planning, and complex adaptive systems perspectives9, 24, 25, 26, 27, and distributed cognition and human–AI collaboration1, 17. Together, they explain how teachers can move beyond basic technology use toward anticipatory, adaptive, and reflective engagement with AI systems in teaching. This framing provides the conceptual basis for developing FLMAIRS as a multidimensional measure of teachers’ readiness for AI-integrated education. The first pillar, metacognitive monitoring and regulation, draws on metacognition and self-regulated learning theory. From this perspective, teachers’ readiness for AI-integrated education involves more than knowing how to use digital tools; it includes the ability to monitor instructional decisions, evaluate feedback, and adjust strategies in response to changing learning conditions4, 16, 31, 32. In AI-supported classrooms, this regulatory cycle becomes more anticipatory because teachers increasingly work with predictive feedback, learning analytics, and algorithmic recommendations1, 5. The second pillar concerns technology acceptance, digital competence, and pedagogical integration. TAM and TAM3 explain how perceived usefulness, perceived ease of use, and behavioral intention shape technology adoption15, 33, while TPACK emphasizes that meaningful technology use depends on the integration of technological, pedagogical, and content knowledge19. Similarly, DigCompEdu provides a competency-based foundation for understanding teachers’ digital professional practices14, 23. In FLMAIR, these perspectives are used not simply to assess technology use, but to clarify how teachers integrate AI into pedagogical decision-making. The third pillar is educational foresight. Future Time Perspective Theory and scenario-planning approaches explain how teachers’ orientation toward future outcomes can shape present instructional planning, especially under technological uncertainty and AI-supported planning contexts9, 25, 26, 74. This pillar is central to the forward-looking nature of FLMAIR, because AI readiness in this study is conceptualized as the capacity to anticipate possible instructional changes, prepare alternative learning pathways, and respond proactively to emerging educational scenarios. The fourth pillar is human–AI collaboration. Distributed Cognition Theory suggests that cognitive activity is distributed across people, tools, and environments, allowing AI systems to function as cognitive partners rather than merely external devices17, 18. In this study, this perspective supports the idea that teacher readiness involves balancing professional judgment with algorithmic feedback, rather than replacing human decision-making with automated outputs. Together, these four pillars provide the theoretical basis for a multidimensional scale that integrates AI readiness, metacognitive regulation, and educational foresight. In this study, these perspectives are not treated as separate theories to be tested independently. Rather, they are used as complementary foundations for defining the measurable dimensions of FLMAIR. The resulting construct focuses on how teachers anticipate AI-related changes, adapt instructional strategies using AI-supported information, design future-oriented learning scenarios, combine professional judgment with algorithmic feedback, and reflect on the long-term consequences of AI use in education1, 4, 9, 14, 16, 17, 25, 26. This synthesis provides the conceptual basis for the five-dimensional structure examined in the psychometric analyses. Operationally, the five FLMAIRS dimensions were defined as follows. AI Foresight Planning refers to teachers’ ability to use AI-supported information and simulations to prepare alternative instructional pathways for anticipated or unexpected educational changes. Adaptive AI-Oriented Instructional Regulation refers to teachers’ ability to adjust lesson sequencing, learning resources, assessment practices, and classroom activities in response to AI-generated feedback and real-time learning data. Algorithmic Anticipatory Reasoning refers to teachers’ ability to interpret algorithmic predictions, compare outputs from different AI-supported sources, and use them to anticipate students’ future learning needs. Human–AI Instructional Synergy refers to teachers’ ability to combine professional judgment with AI-generated recommendations in instructional decision-making. Long-Term AI-Reflective Practice refers to teachers’ ability to evaluate the longer-term effects of AI use on teaching quality, student development, and future instructional planning. Methods This study was conducted in Jordan in 2025 using an exploratory mixed-methods design to develop and psychometrically evaluate FLMAIRS. The qualitative phase informed item development, and the quantitative phase evaluated the scale’s factor structure, reliability, and validity. The overall procedure followed the scale-development framework proposed by Boateng et al., including item development, scale development, and scale evaluation28, 29. Step 1: Domain identification and item generation Domain identification was conducted to define the conceptual boundaries of FLMAIR and distinguish it from related constructs29, 30. In this study, FLMAIR was conceptualized as a multidimensional teacher capacity involving anticipatory thinking, metacognitive regulation, AI-supported instructional decision-making, educational foresight, and human–AI judgment integration. This conceptualization was author-constructed and informed by four complementary areas of literature: metacognition and self-regulated learning4, 16, 31, 32, technology acceptance and teacher digital competence14, 15, 23, 33, future-oriented and scenario-planning perspectives9, 25, 26, and distributed cognition and human–AI collaboration1, 17. These foundations guided the identification of the construct domain and the development of the initial item pool. The review of existing instruments showed that several related domains have already been measured, but usually in partial and separate ways. Instruments on teacher digital competence and technology acceptance provide important information about teachers’ ability and willingness to use educational technologies14, 15, 23, 35, 36. AI readiness and AI competence scales assess awareness, attitudes, self-efficacy, or practical skills for using AI in education34, 38, 39, 40. Metacognitive and self-regulated learning measures capture monitoring and regulation processes but are not designed for AI-supported teaching contexts32, 37. Similarly, future-orientation and forward-looking assessment approaches address long-term thinking or data-informed learning support but do not directly assess teachers’ readiness to work with AI systems41, 42. Taken together, these instruments provide valuable but fragmented coverage of the FLMAIR domain. They do not simultaneously assess teachers’ capacity to anticipate AI-related instructional change, regulate teaching through algorithmic feedback, design future-oriented scenarios, and integrate professional judgment with AI-generated information. This synthesis guided the development of FLMAIRS as a scale intended to capture the intersection of metacognitive regulation, AI-supported decision-making, educational foresight, and human–AI instructional synergy. Accordingly, the problem identification stage did not treat FLMAIRS as a replacement for existing instruments, but as a complementary measure designed to address an unmeasured intersection among established domains. Item generation To ensure conceptual comprehensiveness and contextual relevance, item generation combined deductive and inductive procedures. In the deductive phase, a structured literature review was conducted in Scopus, Web of Science, ERIC, PsycINFO, SpringerLink, and Google Scholar using combinations of terms related to metacognition, future-oriented thinking, AI readiness, teacher professional development, and AI in education. Eligible sources included theoretical or empirical studies in English or Arabic that addressed constructs conceptually related to FLMAIR in educational or educational-management contexts. After abstract and full-text screening, 48 studies were retained, and content analysis of these sources generated 28 initial items. In the inductive phase, semi-structured interviews were conducted with 22 secondary school teachers from different Jordanian provinces. Participants were selected purposively to maximize variation in teaching experience, subject area, and prior engagement with AI-supported tools. Interviews were conducted online, lasted approximately 45–75 min, and continued until data saturation was reached. The interview guide explored teachers’ experiences with AI-supported prediction, instructional adjustment, human–AI decision-making, and future-oriented planning. Interview transcripts were analyzed in MAXQDA Analytics Pro 2024 using conventional content analysis. Meaning units were identified, coded, grouped into subcategories, and then integrated into broader categories. Two coders reviewed the coding process, and disagreements were discussed until conceptual agreement was reached. Trustworthiness was supported through maximum variation sampling, member checking, audit trail documentation, parallel coding, reflective notes, and independent review, consistent with qualitative trustworthiness criteria43. This inductive analysis generated 40 items organized into 12 subcategories and five main categories. The deductive and inductive item pools were then compared to identify conceptual overlap, contextual gaps, and redundant wording. Items were retained when they were conceptually clear, measurable, and aligned with the FLMAIR domain. This integration resulted in an initial pool of 68 items. After review by the research team and two educational technology experts, 10 redundant or overlapping items were removed and wording was refined. The preliminary 58-item version was then prepared for content and face validity assessment. The item development process is summarized in Table 1. Table 1Sample Item Development Process for the Algorithmic Anticipatory Reasoning Factor.Main CategorySubcategoryPrimary CategoryCodesFinal ItemQuotationAlgorithmic Anticipatory ReasoningPredicting the Technological Impacts on the Educational ProcessKnowledge-Driven Adaptation to Technological Advancements-Simulating Teaching Changes Using AI Data-Mapping Potential Outcomes of Educational Changes-Cross-Tool Comparison of Learning Trends-Aligning Instructional Design with Predictive OutputsI draw ideas from AI-related educational reports or news shared in teacher networks or professional groups, and using available AI tools, I simulate the potential impact of these changes on my teaching methods.When a significant educational change, such as a modification in textbook content or assessment methods, is discussed in the teacher group, I quickly use ChatGPT or similar tools to test how, if the change is implemented, assignments or class schedules should be adjusted.Algorithmic Anticipatory ReasoningIntegrating Multi-Source AI Data in Educational Decision-MakingComparative Analysis of Multi-Source AI Data-Simultaneous use of multiple AI tools to monitor progress-Comparing results to identify convergence and divergence of data-Evaluating the accuracy and currency of each source-Adjusting and refining the learning trajectory based on integrated dataWhen I use more than one AI tool in class, I compare their outputs (e.g., progress reports and motivation improvement suggestions) and adjust students’ learning paths accordingly.For example, one of the AI tools we use to monitor student progress may indicate that a student is falling behind. However, when I check their grades in the school system, the decline is not yet reflected in the report card. In such cases, I review the dates of assignments and tests to determine which information is more accurate and adjust the instructional plan accordingly.Algorithmic Anticipatory ReasoningAligning Policies and Technology in Educational PlanningPredicting the Impact of Policy Changes on AI Implementation-Monitoring changes in school regulations regarding AI-Reviewing classroom data recorded by AI-Identifying vulnerable areas in case of tool limitations-Preparing alternative instructional plans aligned with forecastsWhen school regulations or policies regarding the use of AI change, I review the AI-generated reports from my classes to anticipate how these changes will affect the use of these tools in my lessons.Before an official directive arrives for example, banning the use of a specific platform I analyze previous class data to assess what gaps its removal would create and determine which alternatives I should prepare to ensure the teaching process remains uninterrupted. Table 1Sample Item Development Process for the Algorithmic Anticipatory Reasoning Factor.Main CategorySubcategoryPrimary CategoryCodesFinal ItemQuotationAlgorithmic Anticipatory ReasoningPredicting the Technological Impacts on the Educational ProcessKnowledge-Driven Adaptation to Technological Advancements-Simulating Teaching Changes Using AI Data-Mapping Potential Outcomes of Educational Changes-Cross-Tool Comparison of Learning Trends-Aligning Instructional Design with Predictive OutputsI draw ideas from AI-related educational reports or news shared in teacher networks or professional groups, and using available AI tools, I simulate the potential impact of these changes on my teaching methods.When a significant educational change, such as a modification in textbook content or assessment methods, is discussed in the teacher group, I quickly use ChatGPT or similar tools to test how, if the change is implemented, assignments or class schedules should be adjusted.Algorithmic Anticipatory ReasoningIntegrating Multi-Source AI Data in Educational Decision-MakingComparative Analysis of Multi-Source AI Data-Simultaneous use of multiple AI tools to monitor progress-Comparing results to identify convergence and divergence of data-Evaluating the accuracy and currency of each source-Adjusting and refining the learning trajectory based on integrated dataWhen I use more than one AI tool in class, I compare their outputs (e.g., progress reports and motivation improvement suggestions) and adjust students’ learning paths accordingly.For example, one of the AI tools we use to monitor student progress may indicate that a student is falling behind. However, when I check their grades in the school system, the decline is not yet reflected in the report card. In such cases, I review the dates of assignments and tests to determine which information is more accurate and adjust the instructional plan accordingly.Algorithmic Anticipatory ReasoningAligning Policies and Technology in Educational PlanningPredicting the Impact of Policy Changes on AI Implementation-Monitoring changes in school regulations regarding AI-Reviewing classroom data recorded by AI-Identifying vulnerable areas in case of tool limitations-Preparing alternative instructional plans aligned with forecastsWhen school regulations or policies regarding the use of AI change, I review the AI-generated reports from my classes to anticipate how these changes will affect the use of these tools in my lessons.Before an official directive arrives for example, banning the use of a specific platform I analyze previous class data to assess what gaps its removal would create and determine which alternatives I should prepare to ensure the teaching process remains uninterrupted. Sample Item Development Process for the Algorithmic Anticipatory Reasoning Factor. -Simulating Teaching Changes Using AI Data -Mapping Potential Outcomes of Educational Changes -Cross-Tool Comparison of Learning Trends -Aligning Instructional Design with Predictive Outputs -Simultaneous use of multiple AI tools to monitor progress -Comparing results to identify convergence and divergence of data -Evaluating the accuracy and currency of each source -Adjusting and refining the learning trajectory based on integrated data -Monitoring changes in school regulations regarding AI -Reviewing classroom data recorded by AI -Identifying vulnerable areas in case of tool limitations -Preparing alternative instructional plans aligned with forecasts Step 2: Content validity Content and face validity were evaluated to ensure that the items adequately represented the FLMAIR construct and were understandable to the target population44. Content validity was assessed by a panel of 10 experts in psychometrics, educational measurement, teacher education, educational technology, metacognition, educational policy, and AI in education. Experts reviewed the items for relevance, clarity, appropriateness, and comprehensiveness. The Content Validity Ratio (CVR), Item-level Content Validity Index (I-CVI), and Scale-level Content Validity Index (S-CVI) were calculated. Lawshe’s CVR threshold of 0.62 and CVI thresholds of 0.78 for items and 0.90 for the scale were applied45, 46. One item was removed at this stage. Face validity was examined with 12 teachers from the target population and three experts in education and educational technology. Participants reviewed the items for clarity, simplicity, and comprehensibility. The Item Impact Score was also calculated using a five-point importance rating, with values above 1.5 considered acceptable47. Based on qualitative feedback and impact-score results, wording was refined and two additional items were removed. Phase 2: Scale development Step 3: Pre-testing and cognitive interviews Before the main survey administration, the draft instrument was pre-tested to examine item clarity, interpretability, response-option suitability, and alignment with the FLMAIR construct. Ten cognitive interviews were conducted with teachers from the target population across two rounds, following established guidance for improving questionnaire clarity and response processes48. Participants were asked to explain how they understood selected items and how they chose their responses. This process helped identify items that were interpreted differently from the intended meaning, were too long or complex, or created confusion because of wording or item order. Based on the cognitive interview findings, several items were reworded for clarity, response options were checked for suitability, and item order was adjusted to reduce potential confusion. Four items were removed because they were ambiguous or insufficiently aligned with the construct. The revised field-test version consisted of 51 items rated on a 5-point Likert scale ranging from 1 = Never to 5 = Always. Step 4: Survey administration, sampling, and participant recruitment The field-test version of the FLMAIRS was administered online to secondary school teachers in Jordan, following established recommendations for web-based survey administration49, 50. The target population consisted of teachers working in public and private schools who met the inclusion criteria. Participant recruitment was conducted through a stratified random sampling procedure with proportional allocation. First, the sampling frame was organized according to governorate of service, teaching level (lower secondary vs. upper secondary), and teaching experience (< 5 years, 5–15 years, > 15 years). The required number of participants from each stratum was determined proportionally to its share of the target population. Eligible teachers within each stratum were then invited to participate through official school communication channels and professional teacher networks. The online survey included an information sheet, an informed consent form, a brief demographic questionnaire, and the FLMAIRS items. Participation was voluntary, and teachers could proceed to the questionnaire only after providing online informed consent. No names or directly identifying information were collected. The survey was configured so that respondents could submit the form only after completing all items, which prevented missing item-level data. To reduce data-entry errors and support secure data collection, responses were collected through a protected online platform and stored in an encrypted file accessible only to the research team. For sample size determination, a 10:1 respondent-to-item ratio was applied29, 51. A total of 640 Jordanian teachers participated in the field test, including 333 men (52%) and 307 women (48%). Participants’ ages ranged from 28 to 45 years, with a mean age of 35.10 years (SD = 5.16). Inclusion criteria were: (1) at least one year of formal or contractual teaching experience in a Jordanian public or private school, (2) basic familiarity with digital technologies used in education, and (3) voluntary participation with online informed consent. Exclusion criteria were prior participation in the pilot or cognitive-interview phase and lack of access to the digital device required to complete the online scale. The dataset was randomly divided into two independent subsamples. EFA was conducted on one subsample of 320 teachers, and CFA was conducted on the second subsample of 320 teachers. This split-sample approach was used to reduce capitalization on chance and to examine whether the factor structure identified in the exploratory phase could be confirmed in an independent sample29. Step 5: Item reduction analysis Item reduction was conducted within the Classical Test Theory framework to identify items with weak statistical performance or limited alignment with the FLMAIR construct29. Items were examined using endorsement patterns, discrimination indices, inter-item correlations, and adjusted item–total correlations. Items were considered for removal or revision when endorsement proportions were outside the 0.40–0.60 range, discrimination coefficients were below 0.30, or inter-item/item–total correlations were below 0.3052, 53, 54, 55, 56. Because the online survey required responses to all items before submission, there were no missing item-level data and no imputation procedures were required57. Following these analyses, three low-performing items were removed and several items were refined before factor extraction. Step 6: Extraction of factors Before conducting EFA, the suitability of the data for factor analysis was examined using the Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy and Bartlett’s Test of Sphericity55, 58. EFA was then conducted using Principal Axis Factoring (PAF), which is appropriate when the aim is to identify latent constructs based on shared variance among items rather than to reduce observed variables only54, 60, 61. Because the FLMAIR dimensions were theoretically expected to be related, an oblique rotation method was used. The number of factors was determined through a multi-criteria approach that included eigenvalues, scree-plot inspection, parallel analysis, and conceptual interpretability59, 60, 61. Items with factor loadings below 0.40, weak item-level performance, or insufficient conceptual alignment with the construct were considered for removal or revision54, 60, 61. In addition to EFA, Exploratory Graph Analysis (EGA) was used as a complementary approach to examine the dimensional structure of the item network. In EGA, items are represented as nodes and partial correlations between items are represented as edges. The Gaussian Graphical Model was estimated with GLASSO regularization, and community detection was examined using the Walktrap algorithm62, 63, 64, 65. EGA was used to provide graph-theoretical evidence regarding dimensionality and to compare the emerging network structure with the factor structure obtained through classical factor-analytic procedures. Phase 3: Scale evaluation Step 7: Dimensionality testing Dimensionality was evaluated in the independent validation subsample using confirmatory factor analysis (CFA). CFA examined whether the five-factor structure identified through exploratory factor analysis (EFA) could be reproduced in a separate sample. Confirmatory Factor Analysis (CFA) CFA was conducted on the independent validation subsample to test whether the factor structure identified through EFA could be reproduced in a separate group of participants. The analysis followed an Independent Cluster Model (ICM), in which each item was specified to load on its intended factor and cross-loadings were fixed to zero66, 68. Both first-order and second-order CFA models were examined to evaluate whether the five dimensions functioned as distinct but related components of a broader FLMAIR construct. Model fit was evaluated using multiple fit indices rather than relying on a single criterion. The chi-square statistic and χ²/df ratio were reported alongside RMSEA, SRMR, CFI, and TLI, following common recommendations in CFA and structural equation modeling66, 69, 70. Because fit-index cutoffs should be interpreted in relation to model complexity, sample size, and theoretical plausibility, values around CFI/TLI ≥ 0.90, RMSEA ≤ 0.08, and SRMR ≤ 0.08 were considered indicative of acceptable fit, while more stringent criteria were treated as evidence of close fit rather than minimum requirements69, 70. Step 8: Tests of reliability Internal consistency was assessed using Cronbach’s alpha and McDonald’s omega, and composite reliability was calculated for each dimension, given recommendations to report omega alongside alpha in psychological and educational measurement71. Temporal stability was examined through a two-week test–retest procedure with a subsample of 30 teachers. ICC values were calculated for the total scale and each subdimension, with values above 0.75 interpreted as indicating good stability72. Step 9: Tests of validity Construct validity was examined through convergent and discriminant validity analyses. Convergent validity was assessed using Average Variance Extracted (AVE), with values of 0.50 or higher considered acceptable. Discriminant validity was evaluated using the Fornell–Larcker criterion, whereby the square root of each factor’s AVE should exceed its correlations with other factors73. These analyses were used to determine whether each FLMAIRS dimension was sufficiently represented by its items and distinguishable from the other dimensions. Data analysis Data were analyzed using SPSS 26, and RStudio. The total sample was randomly divided into two independent subsamples: one for EFA (n = 320) and one for CFA (n = 320). EFA was conducted using Principal Axis Factoring with Promax rotation in SPSS 26. The number of retained factors was evaluated using eigenvalues, scree-plot inspection, parallel analysis, factor loadings, and theoretical interpretability59, 60, 61. CFA was conducted in RStudio using Maximum Likelihood Estimation to test first-order and second-order five-factor models in the independent validation subsample66, 68, 70. Model fit was evaluated using χ²/df, RMSEA, SRMR, CFI, and TLI, with acceptable fit interpreted as CFI/TLI ≥ 0.90, RMSEA ≤ 0.08, and SRMR ≤ 0.0869, 70. Reliability was assessed using Cronbach’s alpha, McDonald’s omega, composite reliability, and ICC. Convergent and discriminant validity were examined using AVE and the Fornell–Larcker criterion. EGA was conducted in RStudio using a GLASSO network and community-detection approach62, 63, 64, 65. Results Exploratory factor analysis The sample was adequate for factor analysis (KMO = 0.908), and Bartlett’s test of sphericity was significant (χ² = 14,626.42, df = 990, p < 0.001), indicating that the correlation matrix was suitable for factor extraction. EFA identified a five-factor structure with eigenvalues greater than one (Fig. 1). After Promax rotation, the five factors collectively explained 58.29% of the total variance. The factors were labeled using simplified and practitioner-oriented names: AI Foresight Planning, Adaptive AI-Oriented Instructional Regulation, Algorithmic Anticipatory Reasoning, Human–AI Instructional Synergy, and Long-Term AI-Reflective Practice. Items 4, 11, and 44 were removed because their factor loadings were below 0.40 (Table 2). Fig. 1Scree plot. Scree plot. Table 2Results of Exploratory Factor Analysis for FLMAIRS (N = 320).ItemsFactorsh2Factor 1Factor 2Factor 3Factor 4Factor 520. For a single topic, I design multiple skill-based learning paths, and AI selects the optimal path for each student based on their progress level.0.7540.69121. I prepare alternative learning scenarios so that if new technologies (e.g., augmented reality or AI-based real-time translation tools) are introduced, they can be immediately substituted for the current scenario.0.7080.65722. I use AI generative models to create simulated educational stories that illustrate long-term learning outcomes.0.7670.70923. I use AI-simulated data to predict potential events in the learning process (such as drops in motivation or sudden performance spikes) and prepare contingency plans accordingly.0.7450.70524. I design multiple AI-based contingency plans to address potential educational crises (such as internet outages or teacher absences).0.9610.83425. When AI indicates that a specific skill (e.g., digital teamwork or data analysis) will gain greater importance in the future, I incorporate related exercises into the classroom curriculum.0.8220.69126. If a new AI tool I use changes its language, style, or examples, I adapt the examples and exercises so that they remain understandable and relevant for my students.0.8220.78527. Before implementing an AI-assisted activity, I test it in different scenarios (e.g., small or large classes, varying student levels) to ensure it is suitable for all classroom conditions.0.8420.68828. I store lesson content (e.g., videos and slides) in formats compatible with future versions of the AI platform, so that when the system is updated, I can reuse the content with minimal changes.0.9540.78429. When a new AI tool (e.g., an automated assignment grading system or a personalized learning platform) is introduced in the classroom, I prepare a second version of the lesson plan that can be implemented without using that tool.0.6980.55410. When there is a sudden change in the weekly class schedule (e.g., a canceled in-person session or a time change), I adjust the lesson structure using the intelligent scheduling system to ensure the sequence of learning objectives is maintained.0.7760.66912. I update each student’s learning pathway based on new performance data (e.g., online test results or participation reports in virtual classes) analyzed by AI.0.8410.71413. I optimize group projects by adjusting team compositions based on AI analysis of students’ skills and preferred roles.0.6620.66414. I adjust the structure of assessments based on learning patterns (e.g., response times or error rates in different question types) identified by AI.0.6610.65115. I adjust the sequence of my instruction to maintain content continuity when the online learning platform alters the order or sections of the lesson.0.9100.63616. I use real-time AI data (e.g., activity completion rates or sudden drops in engagement) to redistribute educational resources among students.0.8620.69617. I sequence the teaching of concepts based on AI predictions of each topic’s difficulty level for different student groups.0.8060.74918. If AI indicates that students paid less attention to a part of a video or educational image, I replace that section with shorter examples or explanations.0.9350.84019. I dynamically redesign the curriculum in interaction with intelligent algorithms that personalize the learning pathways.0.6890.5791. Before changing the teaching method or adding a new activity, I use an AI predictive model to examine which skills or concepts will require the most reinforcement or repetition.0.8940.7622. Before modifying or removing any part of the lesson content, I use AI simulations to predict how this change will affect students’ performance in final exams and major assessments.0.9450.8193. When selecting an educational platform or tool, I use AI text and speech analysis algorithms to assess its impact on students’ cognitive skills and cultural interactions.0.8470.7485. I use machine learning models to predict which skills or concepts are likely to require reinforcement or review in the coming months.0.9460.8096. I use AI-analyzed learning patterns from the current term’s data to predict potential changes in students’ motivation or learning trajectories.0.5150.5867. I draw ideas from AI-related educational reports or news shared in teacher networks or professional groups, and use available AI tools to simulate the potential impact of these changes on my teaching methods.0.6180.6388. When I use more than one AI tool in class, I compare their outputs (e.g., progress reports and motivation improvement suggestions) and adjust students’ learning paths accordingly.0.6510.5159. When school regulations or policies regarding AI use change, I review AI-generated reports from my class to predict how these changes will affect the use of these tools in my lessons.0.6660.63230. I design learning activities so that AI and I can simultaneously and complementarily support student learning.0.9510.70431. I integrate exercise design ideas with the output of common mistake analyses identified by AI.0.8580.69632. When AI provides suggestions to improve students’ problem-solving, I compare and integrate them with traditional teaching techniques.0.7120.61233. I adjust my teaching methods based on progress or decline patterns identified by AI in the weekly reports.0.6760.53234. I complement my classroom assessments with analytical results from online tests processed by the AI system.0.6010.63835. I provide my classroom feedback in a way that aligns with AI performance indicators (e.g., percentage of mastery over concepts).0.7920.67436. I align the grading suggestions provided by AI with my own assessment decisions to reach a consensus.0.5390.49837. I use AI pattern recognition algorithms to identify students who need special support.0.4830.56638. I compare AI recommendations with my professional judgment to find a balance between data and experience.0.5610.39039. After each AI-driven activity, I save its analytical reports and, at the end of the term, compare students’ progress with the beginning-of-term results.0.6700.48540. I monitor changes in students’ study habits (e.g., use of non-classroom resources or consistency in completing assignments) throughout the term based on AI reports.0.6420.45841. I evaluate AI suggestions in terms of their impact on maintaining teaching quality in the coming months.0.6520.52442. I monitor the effects of AI use on students’ social skills (e.g., collaboration, turn-taking, conflict resolution) throughout the academic year.0.7910.53743. I use AI-generated reports to compare students’ critical thinking skills at the beginning and end of the term.0.5630.43945. I compare AI’s long-term predictions about my class’s strengths and weaknesses with students’ actual end-of-year performance, and based on the differences, I redesign next year’s instructional strategies.0.7750.72046. I periodically (e.g., monthly) assess the extent to which students rely on AI to complete their assignments.0.7050.60247. Based on AI analysis of the current term’s learning trends, I adjust the sequence of some chapters to optimize learning.0.8610.64948. When planning weekly lessons, I consider the potential effects of today’s AI use on learning activities in the following weeks.0.7520.542Eigenvalue6.595.754.814.434.63-Variance Explained Total (58.29%)14.6512.7910.719.8510.30-Factor 1: AI Foresight Planning; Factor 2: Adaptive AI-Oriented Instructional Regulation; Factor 3: Algorithmic Anticipatory Reasoning; Factor 4: Human–AI Instructional Synergy; Factor 5: Long-Term AI-Reflective Practice. Table 2Results of Exploratory Factor Analysis for FLMAIRS (N = 320).ItemsFactorsh2Factor 1Factor 2Factor 3Factor 4Factor 520. For a single topic, I design multiple skill-based learning paths, and AI selects the optimal path for each student based on their progress level.0.7540.69121. I prepare alternative learning scenarios so that if new technologies (e.g., augmented reality or AI-based real-time translation tools) are introduced, they can be immediately substituted for the current scenario.0.7080.65722. I use AI generative models to create simulated educational stories that illustrate long-term learning outcomes.0.7670.70923. I use AI-simulated data to predict potential events in the learning process (such as drops in motivation or sudden performance spikes) and prepare contingency plans accordingly.0.7450.70524. I design multiple AI-based contingency plans to address potential educational crises (such as internet outages or teacher absences).0.9610.83425. When AI indicates that a specific skill (e.g., digital teamwork or data analysis) will gain greater importance in the future, I incorporate related exercises into the classroom curriculum.0.8220.69126. If a new AI tool I use changes its language, style, or examples, I adapt the examples and exercises so that they remain understandable and relevant for my students.0.8220.78527. Before implementing an AI-assisted activity, I test it in different scenarios (e.g., small or large classes, varying student levels) to ensure it is suitable for all classroom conditions.0.8420.68828. I store lesson content (e.g., videos and slides) in formats compatible with future versions of the AI platform, so that when the system is updated, I can reuse the content with minimal changes.0.9540.78429. When a new AI tool (e.g., an automated assignment grading system or a personalized learning platform) is introduced in the classroom, I prepare a second version of the lesson plan that can be implemented without using that tool.0.6980.55410. When there is a sudden change in the weekly class schedule (e.g., a canceled in-person session or a time change), I adjust the lesson structure using the intelligent scheduling system to ensure the sequence of learning objectives is maintained.0.7760.66912. I update each student’s learning pathway based on new performance data (e.g., online test results or participation reports in virtual classes) analyzed by AI.0.8410.71413. I optimize group projects by adjusting team compositions based on AI analysis of students’ skills and preferred roles.0.6620.66414. I adjust the structure of assessments based on learning patterns (e.g., response times or error rates in different question types) identified by AI.0.6610.65115. I adjust the sequence of my instruction to maintain content continuity when the online learning platform alters the order or sections of the lesson.0.9100.63616. I use real-time AI data (e.g., activity completion rates or sudden drops in engagement) to redistribute educational resources among students.0.8620.69617. I sequence the teaching of concepts based on AI predictions of each topic’s difficulty level for different student groups.0.8060.74918. If AI indicates that students paid less attention to a part of a video or educational image, I replace that section with shorter examples or explanations.0.9350.84019. I dynamically redesign the curriculum in interaction with intelligent algorithms that personalize the learning pathways.0.6890.5791. Before changing the teaching method or adding a new activity, I use an AI predictive model to examine which skills or concepts will require the most reinforcement or repetition.0.8940.7622. Before modifying or removing any part of the lesson content, I use AI simulations to predict how this change will affect students’ performance in final exams and major assessments.0.9450.8193. When selecting an educational platform or tool, I use AI text and speech analysis algorithms to assess its impact on students’ cognitive skills and cultural interactions.0.8470.7485. I use machine learning models to predict which skills or concepts are likely to require reinforcement or review in the coming months.0.9460.8096. I use AI-analyzed learning patterns from the current term’s data to predict potential changes in students’ motivation or learning trajectories.0.5150.5867. I draw ideas from AI-related educational reports or news shared in teacher networks or professional groups, and use available AI tools to simulate the potential impact of these changes on my teaching methods.0.6180.6388. When I use more than one AI tool in class, I compare their outputs (e.g., progress reports and motivation improvement suggestions) and adjust students’ learning paths accordingly.0.6510.5159. When school regulations or policies regarding AI use change, I review AI-generated reports from my class to predict how these changes will affect the use of these tools in my lessons.0.6660.63230. I design learning activities so that AI and I can simultaneously and complementarily support student learning.0.9510.70431. I integrate exercise design ideas with the output of common mistake analyses identified by AI.0.8580.69632. When AI provides suggestions to improve students’ problem-solving, I compare and integrate them with traditional teaching techniques.0.7120.61233. I adjust my teaching methods based on progress or decline patterns identified by AI in the weekly reports.0.6760.53234. I complement my classroom assessments with analytical results from online tests processed by the AI system.0.6010.63835. I provide my classroom feedback in a way that aligns with AI performance indicators (e.g., percentage of mastery over concepts).0.7920.67436. I align the grading suggestions provided by AI with my own assessment decisions to reach a consensus.0.5390.49837. I use AI pattern recognition algorithms to identify students who need special support.0.4830.56638. I compare AI recommendations with my professional judgment to find a balance between data and experience.0.5610.39039. After each AI-driven activity, I save its analytical reports and, at the end of the term, compare students’ progress with the beginning-of-term results.0.6700.48540. I monitor changes in students’ study habits (e.g., use of non-classroom resources or consistency in completing assignments) throughout the term based on AI reports.0.6420.45841. I evaluate AI suggestions in terms of their impact on maintaining teaching quality in the coming months.0.6520.52442. I monitor the effects of AI use on students’ social skills (e.g., collaboration, turn-taking, conflict resolution) throughout the academic year.0.7910.53743. I use AI-generated reports to compare students’ critical thinking skills at the beginning and end of the term.0.5630.43945. I compare AI’s long-term predictions about my class’s strengths and weaknesses with students’ actual end-of-year performance, and based on the differences, I redesign next year’s instructional strategies.0.7750.72046. I periodically (e.g., monthly) assess the extent to which students rely on AI to complete their assignments.0.7050.60247. Based on AI analysis of the current term’s learning trends, I adjust the sequence of some chapters to optimize learning.0.8610.64948. When planning weekly lessons, I consider the potential effects of today’s AI use on learning activities in the following weeks.0.7520.542Eigenvalue6.595.754.814.434.63-Variance Explained Total (58.29%)14.6512.7910.719.8510.30-Factor 1: AI Foresight Planning; Factor 2: Adaptive AI-Oriented Instructional Regulation; Factor 3: Algorithmic Anticipatory Reasoning; Factor 4: Human–AI Instructional Synergy; Factor 5: Long-Term AI-Reflective Practice. Results of Exploratory Factor Analysis for FLMAIRS (N = 320). Factor 1: AI Foresight Planning; Factor 2: Adaptive AI-Oriented Instructional Regulation; Factor 3: Algorithmic Anticipatory Reasoning; Factor 4: Human–AI Instructional Synergy; Factor 5: Long-Term AI-Reflective Practice. Confirmatory factor analysis The fit indices obtained from the CFA for the first- and second-order five-factor models indicate acceptable fit for both models. In the CFA model, the fit indices were RMSEA < 0.08, CFI and TLI > 0.90, and SRMR < 0.08, all of which fall within the acceptable range. These results support the adequacy of the hypothesized factor structure in this population (Table 3). The factor loadings of the items in the first-order five-factor CFA model show that all 45 FLMAIRS items had factor loadings greater than 0.4 and were significant at p < 0.001 (Fig. 2). The second-order CFA results are presented in Fig. 3. Table 3Fit Indices of the First- and Second-Order Confirmatory Factor Analysis Models of FLMAIRS.Fit IndexCriteriaFirst-order modelSecond-order modelχ²-1134.231139.62df-935940P-< 0.001< 0.001χ²/df< 51.211.21RMSEA< 0.080.026 (0.020–0.031)0.026 (0.020–0.031)TLI> 0.90.9740.974CFI> 0.90.9750.975SRMR< 0.080.0410.045 Table 3Fit Indices of the First- and Second-Order Confirmatory Factor Analysis Models of FLMAIRS.Fit IndexCriteriaFirst-order modelSecond-order modelχ²-1134.231139.62df-935940P-< 0.001< 0.001χ²/df< 51.211.21RMSEA< 0.080.026 (0.020–0.031)0.026 (0.020–0.031)TLI> 0.90.9740.974CFI> 0.90.9750.975SRMR< 0.080.0410.045 Fit Indices of the First- and Second-Order Confirmatory Factor Analysis Models of FLMAIRS. Fig. 2First-order confirmatory factor analysis of FLMAIRS using the revised five-factor labels. First-order confirmatory factor analysis of FLMAIRS using the revised five-factor labels. Fig. 3Second-order confirmatory factor analysis of FLMAIRS using the revised five-factor labels. Second-order confirmatory factor analysis of FLMAIRS using the revised five-factor labels. Reliability The results of the reliability analysis indicated that the FLMAIRS demonstrated high reliability across the five dimensions of teachers’ forward-looking metacognitive readiness for AI-integrated education. All statistical indices, including the intraclass correlation coefficient (ICC: 0.752–0.851), Cronbach’s alpha (0.897–0.931), McDonald’s omega (0.900–0.932), and composite reliability (CR: 0.906–0.932), fell within acceptable ranges (Table S1). The highest values of Cronbach’s alpha, McDonald’s omega, and composite reliability were observed for Long-Term AI-Reflective Practice (α = 0.931; ω = 0.932; CR = 0.932), while the lowest values were reported for Human–AI Instructional Synergy (α = 0.897; ω = 0.900; CR = 0.909). Regarding ICC, the highest value was found for AI Foresight Planning (ICC = 0.851), whereas the lowest value was observed for Algorithmic Anticipatory Reasoning (ICC = 0.752). Overall, all indices remained within acceptable ranges, confirming the reliability of the FLMAIRS. Convergent and discriminant validity As shown in Table S2, the AVE values for all dimensions exceeded the recommended threshold of 0.50, ranging from 0.51 to 0.61. This indicates adequate convergent validity for all constructs. The highest AVE value was observed for Long-Term AI-Reflective Practice (0.61), while the lowest was observed for Human–AI Instructional Synergy (0.51), both remaining above the acceptable cutoff. For discriminant validity, the square root of AVE for each dimension was greater than its inter-factor correlations with other dimensions, in line with the Fornell–Larcker criterion. Specifically, √AVE values ranged from 0.71 to 0.78, while inter-factor correlations ranged from 0.09 to 0.25. These results confirm that each construct is empirically distinct from the others. Overall, the AVE results and correlation matrix provide evidence supporting both convergent and discriminant validity of the FLMAIRS scale. Exploratory graph analysis The exploratory graph analysis was conducted using the GLASSO algorithm with EBIC model selection (γ = 0.5). The analysis was based on 45 nodes representing the retained FLMAIRS items. The resulting network showed a high level of connectivity among items, with 990 non-zero edges and an average edge weight of 0.022. This pattern indicates that the items were broadly interrelated, although the generally small average edge weight suggests that most associations were weak. Community-detection results were interpreted in relation to the theoretically and empirically supported five-factor structure. Overall, the EGA findings provided complementary evidence that the FLMAIRS items form an interconnected but multidimensional structure consistent with the factor-analytic results (Fig. 4). Fig. 4Dimensionality results from exploratory graph analysis of FLMAIRS. Dimensionality results from exploratory graph analysis of FLMAIRS. Discussion This study developed and evaluated the Forward-Looking Metacognitive–AI Readiness Scale (FLMAIRS) for AI-integrated education among Jordanian teachers. Overall, the findings support FLMAIRS as a multidimensional instrument for assessing how teachers anticipate AI-related instructional change, regulate teaching decisions using AI-supported information, combine professional judgment with algorithmic feedback, and reflect on the longer-term consequences of AI use. The five-factor structure identified in the exploratory phase and supported in the confirmatory analyses suggests that teachers’ AI readiness is not limited to technical skill or general acceptance of technology. Rather, it includes a broader set of metacognitive, adaptive, and future-oriented capacities needed for meaningful AI integration in teaching. AI foresight planning AI Foresight Planning accounted for 14.65% of the total variance and refers to teachers’ ability to use AI-supported information, simulations, and anticipated trends to prepare alternative instructional pathways. This dimension reflects a proactive form of readiness in which teachers do not simply react to technological change after it occurs, but use AI-related information to plan possible scenarios, prepare contingency options, and adapt teaching plans in advance. Theoretically, this factor is closely aligned with future time perspective and scenario-planning approaches, which emphasize the role of future-oriented thinking in present decision-making9, 25, 26. In the context of AI-integrated education, this dimension suggests that teachers’ readiness depends partly on their ability to imagine alternative instructional futures and prepare flexible responses before classroom disruption occurs. Adaptive AI-oriented instructional regulation Adaptive AI-Oriented Instructional Regulation explained 12.79% of the total variance and captures teachers’ ability to adjust lesson sequencing, learning resources, assessment practices, and classroom activities in response to AI-generated feedback and real-time learning data. This dimension extends the logic of self-regulated learning to teachers’ professional practice, where monitoring and adjustment are supported by algorithmic information16. It also reflects the pedagogical integration emphasized in TPACK, because AI-related information becomes meaningful only when teachers can connect it to instructional goals, content demands, and students’ learning needs19. Practically, this factor indicates whether teachers can use AI-supported data to make timely and context-sensitive instructional adjustments. Algorithmic anticipatory reasoning Algorithmic Anticipatory Reasoning accounted for 10.71% of the total variance and refers to teachers’ ability to interpret al.gorithmic predictions, compare outputs from different AI-supported sources, and use these outputs to anticipate students’ future learning needs. This dimension highlights the reasoning process involved in working with AI-generated information. Teachers must evaluate whether algorithmic outputs are relevant, plausible, and consistent with classroom evidence rather than accepting them automatically. This interpretation is consistent with predictive processing and distributed cognition perspectives, which emphasize prediction, feedback, and the distribution of cognitive activity across people and tools8, 17. In practical terms, this factor reflects teachers’ capacity to transform AI outputs into informed educational judgments. Human–AI instructional synergy Human–AI Instructional Synergy explained 9.85% of the total variance and refers to teachers’ ability to combine professional judgment with AI-generated recommendations in instructional decision-making. This dimension emphasizes that AI readiness is not equivalent to dependence on automated systems. Instead, it involves a balanced form of collaboration in which teachers use AI as a source of information while maintaining pedagogical responsibility and contextual judgment. Distributed Cognition Theory provides a useful foundation for this interpretation because it views cognition as distributed across individuals, tools, and environments17, 18. Within AI-integrated classrooms, this factor reflects the teacher’s ability to coordinate human expertise and algorithmic support in ways that enhance, rather than replace, professional decision-making. Long-term AI-reflective practice Long-Term AI-Reflective Practice accounted for 10.30% of the total variance and refers to teachers’ ability to evaluate the longer-term effects of AI use on teaching quality, student development, and future instructional planning. Unlike short-term adjustment, this dimension focuses on reflection across extended time horizons. Teachers with higher levels of this readiness are expected to compare AI-generated predictions with later classroom outcomes, reconsider instructional strategies, and use long-term feedback to improve future practice. This dimension is consistent with future-oriented metacognition and reflective professional learning, as it links AI-supported evidence with sustained pedagogical reflection4, 9, 16. It therefore captures the long-term evaluative component of teachers’ readiness for AI-integrated education. The pattern obtained from the confirmatory factor analysis for both the first-order and second-order five-factor models provided robust evidence of an acceptable fit for the FLMAIRS factorial structure. According to commonly accepted criteria in the psychometric literature, the obtained fit indices fell within the recommended ranges; specifically, CFI and TLI exceeded the 0.90 threshold, indicating the model’s strong capacity to represent the observed covariance relationships based on the hypothesized factor structure54. This aligns with the perspective of Marsh et al., suggesting model adequacy even under conditions of relative complexity, provided that factor loadings are significant and homogeneous across their respective dimensions68. Moreover, the SRMR value below 0.08 indicates a satisfactory correspondence between the predicted and observed covariance matrices69. Such a pattern of results, especially when examined alongside the simultaneous fit of first-order and second-order models, indicates that the scale’s dimensions not only possess independent explanatory power but also function effectively as part of a higher-order construct. According to Putnick and Bornstein, this characteristic serves as a key indicator of the multidimensional coherence of a measurement instrument67. Overall, the combination of fit indices and the pattern of factor loadings support the construct validity of FLMAIRS and indicates that the instrument adequately represents the construct of Forward-Looking Metacognitive–AI Readiness within the target population. The findings from reliability and validity analyses provided strong evidence for the psychometric robustness of the FLMAIRS. According to widely accepted psychometric criteria, instruments that demonstrate both high internal consistency and satisfactory temporal stability exhibit greater reliability in replicating results56, 72. In this study, the values of internal and composite indices across all dimensions exceeded the recommended thresholds for high-quality measurement tools, indicating a coherent and well-aligned structure among items, consistent with Kline’s perspective54. External stability, assessed through retesting, further suggested that the instrument delivers a stable representation of the intended constructs even across different time points. As emphasized by DeVellis and Thorpe, this characteristic is crucial for longitudinal and cross-cultural applications of the tool52. From the perspective of convergent validity, achieving an AVE greater than 0.50 for each dimension indicates that the shared variance between each dimension and its indicators exceeds the variance attributable to measurement error58. This finding aligns with the recommendations of Fornell and Larcker, suggesting that each dimension is effectively explained by its respective items73. In terms of discriminant validity, full compliance with the Fornell–Larcker criterion indicates that the defined dimensions exhibit minimal conceptual overlap with one another73, thereby preventing the intrusion of irrelevant variance that could threaten construct validity75. This situation ensures that item development was conducted with precision, focusing on the conceptual boundaries of each dimension, and that the initial theoretical model is accurately represented in the factor structure. Overall, the combination of reliability and convergent–discriminant validity evidence indicates that FLMAIRS is a suitable instrument for assessing teachers’ Forward-Looking Metacognitive–AI Readiness and provides sufficient measurement quality, reliability, and interpretive accuracy for use in educational research and evidence-informed policy planning. The EGA results provided complementary network-based evidence for the internal structure of FLMAIRS. The network showed broad connectivity among the retained items, suggesting that the dimensions share a common conceptual foundation related to teachers’ readiness for AI-integrated education. At the same time, the EGA findings were interpreted alongside the EFA and CFA results rather than treated as independent evidence of dimensionality. The convergence between the network-based interpretation and the five-factor structure supports the view that FLMAIRS is an interconnected but multidimensional instrument. However, because the estimated network was highly dense and the average edge weight was small, these findings should be interpreted cautiously and replicated in future samples, in line with recommendations for interpreting psychological networks and their accuracy76. Conclusion The findings indicate that FLMAIRS has a robust multidimensional psychometric structure for assessing teachers’ forward-looking metacognitive readiness for AI-integrated education. EFA supported a five-factor structure consisting of AI Foresight Planning, Adaptive AI-Oriented Instructional Regulation, Algorithmic Anticipatory Reasoning, Human–AI Instructional Synergy, and Long-Term AI-Reflective Practice. This structure was further supported by CFA, with acceptable fit for both first-order and second-order models. Reliability, test–retest stability, convergent validity, and discriminant validity results provided additional evidence for the quality and interpretability of the scale. Overall, FLMAIRS offers a context-sensitive tool for assessing how teachers anticipate AI-related instructional changes, regulate teaching with AI-supported feedback, integrate professional judgment with algorithmic outputs, and reflect on the long-term implications of AI use in education. Strengths and limitations This study has several methodological strengths. First, FLMAIRS was developed through an exploratory mixed-methods design that combined theoretical evidence from the literature with teachers’ lived experiences of AI-supported education. This combination strengthened both the conceptual coverage and contextual relevance of the item pool. Second, the scale-development process followed recognized psychometric procedures, including expert review, content and face validity assessment, cognitive interviewing, item reduction, EFA, CFA, reliability assessment, convergent and discriminant validity testing analysis. Third, the quantitative phase used a large and well-structured sample of 640 teachers. The use of proportional stratified sampling, good gender balance, and independent subsamples for EFA and CFA strengthened the internal robustness of the findings and reduced the risk of sample-specific factor solutions. These features support the use of FLMAIRS within the Jordanian teacher population while still requiring further validation in other cultural and educational contexts. However, several limitations should be considered when interpreting the findings. First, the data were collected at a single time point. This is important because Forward-Looking Metacognitive–AI Readiness is conceptualized as a future-oriented and dynamic construct. Although the present study provides initial evidence for the factor structure, reliability, and validity of FLMAIRS, it cannot show how teachers’ readiness changes over time, how stable the construct remains across different phases of AI implementation, or how sensitive the scale is to professional-development interventions. Future research should therefore use longitudinal designs to examine the temporal development of FLMAIRS scores, test–retest stability over longer intervals, and responsiveness to AI literacy programs or teacher-training interventions. Second, the sample consisted only of teachers working in Jordan. This focus was appropriate for developing a culturally and contextually grounded instrument, particularly because the study aimed to examine AI readiness within the Jordanian educational setting. However, it limits the extent to which the findings can be generalized internationally. Educational systems differ in their AI policies, digital infrastructure, teacher-preparation models, professional-development opportunities, and cultural expectations regarding technology use. Therefore, future studies should validate FLMAIRS in other Middle Eastern and international contexts, examine whether the five-factor structure is replicated, and test measurement invariance across countries, school systems, languages, and cultural groups. Third, both phases of data collection relied on self-reported information. The qualitative interviews depended on teachers’ descriptions of their experiences with AI-supported instruction, and the quantitative survey depended on teachers’ self-ratings of their readiness. Although these approaches were appropriate for scale development and psychometric evaluation, they may be influenced by social desirability, recall bias, inaccurate self-perceptions, or participants’ tendency to overestimate their readiness for AI-integrated education. Future research should therefore examine FLMAIRS alongside external and behavioral indicators, such as classroom observations, AI-use logs, performance-based tasks, supervisor or peer evaluations, and evidence of actual AI-supported instructional practice. Such designs would help establish stronger criterion-related and ecological validity. Practical implications FLMAIRS can be used by researchers, school leaders, and policymakers to assess teachers’ readiness for AI-integrated education in a more targeted way. Because the scale measures five related but distinct dimensions, it can help identify whether teachers need support in AI foresight planning, adaptive instructional regulation, algorithmic reasoning, human–AI instructional synergy, or long-term reflective practice. At the policy and professional-development level, FLMAIRS may help identify areas where teachers require additional training before AI-supported tools are implemented at scale. For example, low scores in AI Foresight Planning may indicate the need for training on scenario planning and the design of alternative instructional pathways. Lower scores in Human–AI Instructional Synergy may suggest a need for professional development focused on interpreting AI recommendations while maintaining teachers’ pedagogical judgment. At the school level, the scale can support needs assessment, mentoring, and targeted capacity-building. Teachers with stronger readiness profiles may serve as peer mentors or lead teachers in AI-related professional learning communities. At the research level, FLMAIRS can be used in pretest–posttest designs to evaluate AI literacy workshops, professional-development programs, or school-level technology initiatives. The scale may also facilitate future research examining relationships between forward-looking metacognitive–AI readiness and other educational variables, such as technology integration, instructional innovation, and teacher professional development. Finally, the scale may inform the design of AI-supported educational systems. Dimension-level results can help developers and instructional designers understand which types of AI functions teachers may be prepared to use and where additional scaffolding is needed. In this way, FLMAIRS can contribute to more evidence-informed, context-sensitive, and teacher-centered implementation of AI in education. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2