Evaluation of primary school teachers’ use and perceptions of artificial intelligence in primary school mathematics instruction: a mixed-methods study

Synthesis: Ciğerci and Uygun (2026) surveyed 302 primary school teachers in Türkiye during the 2024–2025 academic year and then interviewed 10 of them, using an explanatory sequential mixed-methods design: survey first, then interviews used to explain it. The survey applied the Turkish adaptation of a 25-item multidimensional scale covering willingness to use AI, attitudes toward AI, professional expectations and personal experiences. Teachers reported moderately positive overall perceptions (M = 3.62, SD = 0.61), with willingness (M = 3.98) and attitudes (M = 3.82) well above personal experiences (M = 3.00). Frequency of AI use and prior training were the variables most consistently associated with favorable perceptions, while gender, professional experience and educational background produced dimension-specific differences only. The interviews positioned AI as supportive, teacher-mediated work across planning, materials, visualization, problem solving, assessment, feedback and out-of-class learning, alongside concerns about infrastructure, reliability, privacy, age appropriateness, teacher competence and student over-reliance. The authors stress that the study documents perceptions and reported use, not instructional effectiveness.

Key Findings

• 302 primary school teachers in Türkiye completed the survey in the 2024–2025 academic year, and 10 trained teachers were interviewed to explain the patterns.
• Teachers reported moderately positive overall perceptions (M = 3.62, SD = 0.61), with willingness (M = 3.98) and attitudes (M = 3.82) above personal experience (M = 3.00).
• Frequency of AI use and prior AI training were most consistently associated with favorable perceptions; gender, experience and degree varied by dimension only.
• Interviews placed AI at planning, materials, visualization, problem solving, assessment, feedback and out-of-class learning, with ChatGPT, GeoGebra and Photomath mentioned most.
• Teachers raised concerns about infrastructure, access, reliability, privacy, age appropriateness and competence, and worried that ready-made answers erode persistence and reasoning.
• The authors frame the results as perceptions and reported use rather than evidence of effectiveness, and call for studies linking use to observed practice and outcomes.

What the survey contributed

The quantitative phase was a descriptive survey of 302 in-service primary teachers of Grades 1 to 4, recruited through voluntary online participation distributed via professional networks and institutional channels. Of these, 70.2% were female, 62.9% held a bachelor's degree, 19.2% (n = 58) had received prior training related to AI, and 17.9% reported never using AI. The instrument was a 25-item scale with four dimensions, all rated on a five-point Likert scale: willingness to use AI, attitudes toward AI, professional expectations toward AI and personal experiences with AI. Internal consistency was high, with Cronbach's alpha of 0.954 for the total scale, a value the authors also flag as a possible sign of item overlap. Group comparisons used independent-samples t-tests, one-way ANOVA and Tukey's HSD post-hoc tests, with Cohen's d and eta squared read alongside significance. The survey maps the distribution of self-reported perceptions and the background variables they track, without testing any instructional effect.

What the interviews added

Ten volunteer teachers were chosen by criterion-based purposive sampling from the 58 who had prior AI training, and interviewed in semi-structured Zoom sessions using a seven-question protocol reviewed by three experts. MaxQDA Analytics Pro 2024 supported coding through thematic analysis, and participants confirmed their transcripts before analysis began. Their accounts locate AI at specific points in instruction. Planning was named most often, alongside material development, attention-getting lesson openings, reinforcement, assessment and feedback, and out-of-class support. Teachers described generating questions aligned with learning objectives, explaining topics with ChatGPT, solving problems step by step with Photomath, concretizing fractions through geometric shapes in GeoGebra, and analyzing student responses to find gaps. They also named constraints the survey could not capture: missing smart boards, uneven internet and hardware access, unverified output, data privacy, the abstractness of AI for young children, and their own limited competence.

Where the two strands meet

The survey shows a favorable but shallow orientation: willingness (M = 3.98) and attitudes (M = 3.82) sit well above the midpoint while personal experience (M = 3.00) sits at it, and the strongest statistical patterns belong to prior training (t = 3.661, p = 0.001) and frequency of AI use (F = 41.280, p = 0.001) rather than to demographic categories. The interviews suggest why that gap can exist, since teachers can picture many uses of AI while still reporting infrastructure and competence barriers. Their accounts cast AI as teacher-mediated support rather than an autonomous tutor, and the authors read the pattern through the Technology Acceptance Model for willingness and Technological Pedagogical Content Knowledge (TPACK) for the pedagogical and content-specific knowledge needed to judge generated explanations. The contribution is descriptive: a portrait of perceived opportunity and constraint, not evidence that AI improves mathematics learning.

What this means for practice

• Treat prior AI training as the clearest lever: only 19.2% of these teachers had any, yet that group scored markedly higher across the dimensions, so structured professional training is the first thing schools can act on.
• Frequency of AI use separated teachers as strongly as training did, which favors sustained, repeated use over one-off introductions when building competence.
• Frame AI as support for planning, materials and feedback rather than as a replacement for teacher judgment, which is how the interviewed teachers themselves described their role.
• Because teachers raised age appropriateness and over-reliance alongside uneven access, pair any rollout with explicit guidance on what young children should and should not do with AI, and with equity checks on devices and connectivity.
• The authors document perceptions and reported use, not outcomes, so treat this as a starting point to test against observed practice rather than as proof that AI improves mathematics learning.

Limitations

• The sample was not designed to be statistically representative of primary school teachers in Türkiye, and voluntary online recruitment invites self-selection by teachers already interested in technology.
• Both strands rest on self-reported data, so reported uses may diverge from classroom practice and may be shaped by recall limits or social desirability.
• Because the design is cross-sectional, the links among training, frequency of use and positive perceptions cannot be read causally; teachers who already held favorable views may be more likely to seek both.
• The 10 interview participants all had prior AI training, so untrained teachers are not represented. The authors also note that the very high alpha may indicate item redundancy, and that no student achievement, reasoning or observed teaching quality was measured.

Connected Concepts

• Teacher AI Competency: competence and AI literacy named as barriers by interviewed teachers
• Teaching: AI framed as support for teacher judgment rather than a substitute
• Technological Pedagogical Content Knowledge (TPACK): interpretive lens for judging AI output in mathematics
• Workplace Learning: prior AI training as the variable most associated with favorable perceptions
• Formative Assessment: assessment and feedback as prominent perceived uses
• Digital Divide: unequal access to devices, internet and tools
• Generative AI: ChatGPT and other tools mentioned across instructional stages
• math education: the subject-specific context of the study

Connected Articles

• Talking mathematics with AI: Understanding teachers' motivation for utilizing chatbots: teachers' motivation for using mathematics chatbots
• Perceptions and Acceptance of Artificial Intelligence in Science Education Programmes: Voices of Pre-Service Science Teachers: positive perceptions alongside limited actual use
• Suitability of Artificial Intelligence Supported Lesson Plans from the Perspective of Science Education Experts: expert review of AI-supported lesson plans

Citation

Ciğerci, F. M., & Uygun, N. (2026). Evaluation of primary school teachers’ use and perceptions of artificial intelligence in primary school mathematics instruction: a mixed-methods study.