K-12 teachers worldwide experience varying levels of "cultural distance" when using GenAI tools, ranging from seamless alignment in communication tasks to complete breakdown when local languages and cultural contexts are absent from training data, revealing how global AI defaults systematically favor dominant cultures over marginalized educational communities.
Objective: This study introduced and explored the concept of "cultural distance" - defined as the gap between GenAI's default cultural repertoire and the situated demands of teaching practice. The researchers aimed to understand how K-12 teachers across diverse global contexts experience and navigate misalignments between GenAI outputs and their local classroom needs, moving beyond binary assessments of AI bias to examine the spectrum of user effort required for adaptation.
Methods: The researchers conducted 30 in-depth, semi-structured interviews with K-12 teachers who actively used GenAI in their teaching practice, with 10 participants each from South Africa, Taiwan, and the United States. These regions were selected to represent a gradient of cultural distance, differing in linguistic diversity, educational traditions, resource availability, and domestic AI development presence. Interviews lasted 60-90 minutes and were conducted remotely via Zoom in participants' primary teaching languages. The study used reflexive thematic analysis to identify patterns of alignment and misalignment, developing the cultural distance framework inductively from teachers' accounts of their everyday GenAI experiences across various tasks including communication, lesson design, assessment creation, and culturally responsive teaching.
Key Findings: The analysis revealed a three-level spectrum of cultural distance. At the low-distance level, GenAI outputs aligned smoothly with teachers' needs in routine communication tasks (drafting emails, parent communications) and brainstorming engaging instructional activities, requiring only minimal editing while often exceeding expectations. At the mid-distance level, teachers encountered partial misalignments requiring significant effort, particularly in generating assessment questions (which often had mismatched difficulty levels or overly formal vocabulary) and creating culturally relevant learning activities (where teachers had to extensively rework prompts or supplement outputs with local knowledge). At the high-distance level, teachers faced structural barriers where adaptation was nearly impossible, including when local languages and indigenous traditions were missing from training data (resulting in grammatically incorrect or completely absent responses) and when policy-level controls blocked queries deemed sensitive by either educational institutions or GenAI companies themselves.
Implications: This research demonstrates that AI alignment is not binary but exists on a spectrum of cultural distance that determines how much effort users must invest to make global AI systems locally useful. The findings reveal that current GenAI tools systematically favor high-resource languages and dominant cultural contexts, placing disproportionate adaptation burdens on teachers in marginalized communities. For educational practice, this means recognizing that successful GenAI integration requires not just technological access but also significant cultural labor that is unevenly distributed. For AI development, companies must prioritize expanding training data to include underrepresented languages and cultural contexts, while designing more transparent interfaces that communicate system limitations rather than producing misleading outputs. For policy, the research highlights needs for investment in local AI research capabilities, clearer governance around content restrictions, and recognition that alignment requires both technical improvements and redistribution of adaptation responsibility.
Limitations: The study's scope was limited to 30 teachers across three specific regions and cannot represent global diversity in K-12 education. The focus on chat-based GenAI tools may not capture experiences with more deeply integrated educational technologies. The interview-based methodology captured teachers' self-reported experiences rather than observational data or usage logs. Additionally, the cultural distance framework was developed as an exploratory concept rather than a closed typology, requiring further testing and refinement across different domains and user groups. The research also primarily examined current GenAI limitations rather than tracking how these dynamics might evolve as AI systems improve.
Future Directions: The researchers recommend extending the cultural distance framework to additional regions with different technological infrastructures and cultural contexts, exploring how the concept applies to non-educational domains such as journalism or healthcare, and conducting longitudinal studies to track how cultural distance changes as AI systems develop. Future work should also incorporate multi-stakeholder perspectives including students and policymakers, examine more deeply integrated AI educational technologies beyond chatbots, and test specific interventions designed to reduce cultural distance for underrepresented communities.
Title and Authors: "Bridging Cultural Distance Between Models Default and Local Classroom Demands: How Global Teachers Adopt GenAI to Support Everyday Teaching Practices" by Ruiwei Xiao, Qing Xiao, Xinying Hou, Hanqi Jane Li, Phenyo Phemelo Moletsane, Hong Shen, and John Stamper.
Published On: September 13, 2025 (arXiv preprint)
Published By: Submitted to ACM Conference on Human Factors in Computing Systems, published as arXiv preprint arXiv:2509.10780v1