Python and ChatGPT: AI-Powered Data Visualization for Teachers

Python and ChatGPT: AI-Powered Data Visualization for Teachers | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Python and ChatGPT: AI-Powered Data Visualization for Teachers CARIE JUSTINE ESTRELLADO This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8023714/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study explores the synergy of Python and ChatGPT's intuitive accessibility to empower non-programming educators in making data-driven decisions. Using a Quantitative Ethnography (QE) approach, we analyzed data discourse of behavior of twenty teachers from three private schools through AI-guided prompts. Qualitative coding first identified teachers' difficulties and strategies. Epistemic Network Analysis (ENA) was then employed to generate network models, revealing how these codes co-occur. We observed a shift in teachers' frameworks over a three-week period. In Session 1, the conceptual focus was on technical hurdles, with strong connections linking "Prompt engineering," "Data cleaning & preparation," and "Syntax errors". This indicates a discourse dominated by a linear, technical problem-solving approach. The group centroid for this session was located on the left side of the conceptual space. By Session 2, a shift occurred with a large effect size (Cohen's d=0.93). The restructuring shows that as teachers overcame initial technical challenges, their acumen elevated from low-level tool mechanics to higher-order pedagogical application. The findings provide an account of a transition from a conceptual space dominated by technical difficulties to one focused on the strategic, critical, and pedagogical application of data vis-à-vis their professional practice. While the study is limited by the exclusion of data storytelling context, it highlights the need for additional design features to help teachers engage meaningfully with data. Theoretical Computer Science Technical Communication Artificial Intelligence in Education Python ChatGPT Epistemic Network Analysis Figures Figure 1 Figure 2 1. Introduction Education, like many sectors, is reliant on data to inform strategies, enhance teaching outcomes, and cultivate improved learning environments. But data are not the same from everyone to be seen (Castillo & Strunk, 2025). This comes the premise for visualization where people can process images over texts (Lawn, 2013; Perkhofer et al., 2020). Catapulting statement towards data visualization where it is paramount in using visual representations such as charts, graphs, and interactive dashboards to analyze a wide array of data, including student performance, attendance patterns, demographic information, and learning behaviors. This visual approach makes data easier to digest and allows decision-makers to quickly grasp critical indicators, streamlining operations and facilitating timely interventions when indicators deviate from expectations (Grand Bang, 2024; Makowski & Lubienski, 2023; Sorapure, 2019). The purpose of this approach is to transform complex datasets into understandable and actionable information for educators and administrators, enabling them to identify patterns, trends, and correlations that might otherwise be obscured in raw numbers and spreadsheets. Despite the clear advantages, teachers frequently encounter significant challenges when attempting to upskill data visualization (Lawn, 2013). A primary obstacle is a widespread lack of visualization literacy (VL) (Glasgow & Farrell, 2007; Krejci et al., 2020; Lawn, 2013; Perkhofer et al., 2020), meaning many educators do not possess the technical skills required to effectively interpret or create data visualizations. This deficiency can lead to superficial or even erroneous interpretations of data, which can negatively influence instructional decisions and, consequently, student outcomes. Furthermore, access to user-friendly tools remains a barrier. While powerful data visualization platforms like Tableau (Jackson, 2025) and Power BI exist (Box, 2023), their complexity and cost often make them inaccessible to teachers who are not trained data scientists or programmers. The fragmentation of educational data across disparate systems, known as data silos, further complicates efforts to collect, integrate, and visualize information cohesively (Adarsh Bhavimane et al., 2024; Perkhofer et al., 2020). Compounding these issues are severe time constraints; teachers have limited time for data analysis, and traditional visualization methods are often too time-consuming for daily application. The traditional landscape of data visualization for teachers typically involved manual creation of basic charts and graphs using spreadsheet software like Microsoft Excel (Sorapure, 2019). While ubiquitous, these methods are inherently static and has limited interactivity, and are time-consuming to create and update, especially with large datasets. Such conventional approaches are often insufficient for uncovering complex patterns and relationships within educational data. The contemporary trend is rapidly shifting towards more interactive, dynamic, and automated dashboards and visualizations that transcend simple charts to construct a compelling narrative with the data (Kim et al., 2024). 2. Bridging the Technical Gap for Non-Programming Teachers The convergence of Python's robust capabilities and ChatGPT's intuitive accessibility represents a significant advancement for educators who are not programming specialists. This development empowers individuals with limited coding skills to readily create data visualization dashboards, transforming raw data into meaningful insights. Large Language Models (LLMs) like ChatGPT serve as dynamic resources for educators, assisting not only with understanding advanced concepts but also with solving complex problems While some perspectives suggest that exclusive reliance on ChatGPT might impede deep coding comprehension, the primary utility of this synergy for teachers lies in enabling the application of data visualization rather than demanding fundamental programming mastery (Campesato, 2024 ; Kim et al., 2024 ). This allows educators to reallocate their cognitive effort from debugging syntax and navigating complex programming environments to focusing on pedagogical implications, critical data interpretation, and the direct impact on student learning. Python, with its rich ecosystem of libraries, stands as a formidable language for data analysis and visualization. Libraries such as Matplotlib, Seaborn, and Plotly offer extensive customization options and the capacity to manage large and intricate datasets. Matplotlib, for instance, is a foundational library capable of generating static, animated, and interactive 2D plots, including line, bar, scatter, and pie charts. Seaborn, built upon Matplotlib, simplifies the creation of aesthetically pleasing statistical plots and integrates seamlessly with pandas DataFrames, offering specialized visualizations like heatmaps (Campesato, 2024 ). Plotly, a versatile open-source library, excels at creating interactive, web-based visualizations and offers a wide array of 2D and 3D chart types, integrating well with Jupyter notebooks. These libraries collectively provide the computational backbone for advanced data visualization. ChatGPT's Advanced Data Analysis (ADA) feature functions as a simplified Jupyter Notebook directly within the chat interface, creating what can be considered a "conversational IDE". Teachers can upload CSV files or other data formats and instruct ChatGPT to generate and execute Python code, producing charts and analyses directly within the chat environment. This provides a user-friendly setting for quick prototyping and analysis, allowing for immediate visual feedback on data. The environment comes pre-loaded with hundreds of Python libraries, enabling ChatGPT to write and execute code for complex tasks (Campesato, 2023 ). This interactive environment supports an iterative development process. Teachers can refine their visualizations by providing follow-up prompts to ChatGPT, such as "Change the color of the bars to blue" or "Add a title to the chart". This continuous refinement process allows for ongoing adjustments without requiring extensive coding knowledge. This dynamic interaction, where teachers can ask for "more examples," or even "that doesn't sound right, try again," demonstrates that the process is an ongoing dialogue rather than a single query-response. This iterative capability is a significant pedagogical advantage, allowing teachers to experiment and refine their visualizations in real-time without the pressure of achieving perfection on the first attempt, fostering a more exploratory approach to data analysis. This synergy bridges the gap between the power of a programming language and the non-technical user. It enables teachers to leverage Python's extensive capabilities without needing to become Python experts, thereby democratizing access to sophisticated data analysis tools in education. This shift allows teachers to focus their efforts on the what and why of data analysis—the pedagogical application and interpretation—rather than the how of coding mechanics, representing a more appropriate and efficient allocation of their professional expertise. 3. Crafting Prompts for ChatGPT To effectively harness the Python-ChatGPT synergy, teachers must develop proficiency in prompt engineering. Prompts serve as the primary interface for interaction, and their clarity and precision are paramount to achieving the desired visualization outcomes. The principles of effective prompt engineering for data visualization emphasize clarity and specificity. It is crucial to craft prompts that are clear, specific, and provide sufficient context for ChatGPT to accurately understand the request (Kasneci et al., 2023 ). Ambiguity should be avoided, and precision in defining desired details is paramount. For instance, specifying the exact columns and desired chart type is significantly more effective than a vague request. This means teachers need to think critically about the data they possess and the specific questions they want to answer visually (Chen et al., 2023 ; Kasneci et al., 2023 ). Prompt engineering is inherently an iterative process where users should begin with an initial prompt, review ChatGPT's response, and then refine the prompt based on the output (Chen et al., 2023 ; Maddigan & Susnjak, 2023 ). This involves adjusting wording, adding more context, or simplifying the request as needed to improve the results. This iterative dialogue allows for continuous refinement, much like a collaborative process with a human data analyst. Explicitly stating the desired visualization type, such as "create a bar chart" or "generate an interactive dashboard," guides ChatGPT more effectively toward the intended visual representation. Furthermore, for ChatGPT to perform accurate analysis and visualization, it must understand the dataset. This involves uploading the relevant CSV or Excel file and, if necessary, providing a brief description of the data to confirm ChatGPT's interpretation. This ensures that the AI's understanding aligns with the teacher's intent, mitigating potential misinterpretations of the data. Effective prompt engineering for data visualization effectively becomes the "new coding" for non-technical users. This shifts the primary skill requirement from mastering programming syntax to precisely articulating analytical intent and understanding data structure. This subtle but critical shift inherently demands a foundational level of data literacy from teachers, as they must think critically about their data and the specific insights they wish to extract, even if they are not writing the code themselves. The following table provides practical examples of prompts that teachers can use to generate various data visualizations and perform related data tasks. Table I. Example ChatGPT Prompts for Educational Data Visualizations Visualization Type/Task Prompt Example Purpose for Teacher Simple Bar Chart "I have a CSV file named student_scores.csv with columns 'student_name' and 'math_score'. Please create a bar chart showing the math scores for each student." Quickly visualize individual student performance on a single academic metric for immediate understanding. Scatter Plot with Trendline "Using the student_data.csv file with columns 'hours_studied' and 'exam_grade', plot a scatter plot to show the relationship between the two. Add a trendline to the plot." Explore potential correlations or relationships between two continuous variables (e.g., study time and academic achievement). Interactive Dashboard "Create an interactive dashboard with two charts from student_attendance.csv: a bar chart showing average attendance by grade level, and a line chart showing attendance trends over the last month. Make the dashboard filterable by grade." Monitor multiple key educational metrics simultaneously, allowing for dynamic exploration and drill-down analysis. Grouped Bar Chart "From the class_data.csv file with columns 'gender', 'subject', and 'average_grade', create a grouped bar chart that compares the average grades of male and female students in each subject." Facilitate comparisons of performance across different student subgroups to identify disparities and inform targeted instructional strategies. Data Cleaning "Clean this data and remove any errors that can affect the output quality." OR "Remove rows from student_data.csv where 'grade' is missing or below 0, and rename the 'student_ID' column to 'ID'." Prepare raw data for accurate analysis by removing inconsistencies and standardizing format. Refinement/Customization "Change the color of the bars to blue and add a title 'Math Scores by Student' to the chart." Tailor the visual output to meet specific presentation needs or enhance clarity and aesthetic appeal. The specificity and variety of these prompt examples demonstrate that while teachers no longer need to know how to write Python code, they absolutely need to know what they want to analyze, what kind of relationships they are looking for, and how they want the data to be represented visually. This shifts the pedagogical emphasis from technical mechanics to a deeper understanding of data-driven inquiry and the principles of effective data representation (Perkhofer et al., 2020 ). This capability empowers teachers to move beyond basic data consumption to active data manipulation and inquiry, provided they have the conceptual understanding to formulate precise questions and interpret the results. 4. Benefits for Teachers in Assessment and Instruction The accessible data visualization capabilities unlocked by the Python-ChatGPT synergy offer profound benefits for teachers in their daily assessment and instructional practices. This approach empowers educators to transform their use of data, moving beyond traditional, static methods to more dynamic and responsive strategies (Campesato, 2023 , 2024 ). The ability to easily visualize data empowers teachers to quickly identify individual students who are struggling by analyzing test scores over time, enabling early and targeted intervention. Visualizations are highly effective at revealing broader trends in student performance, such as a decline in a specific skill across the class or a general improvement over a period (Jiang et al., 2024 ). The rapid visualization of longitudinal and aggregated student data fundamentally shifts teachers from reactive assessment—merely grading after the fact—to proactive, formative intervention (Perkhofer et al., 2020 ). This represents a crucial pedagogical evolution from simply measuring learning outcomes to actively informing and adapting instructional strategies in real-time. Furthermore, teachers can effectively utilize data visualizations to compare the academic performance of various student groups, such as those differentiated by gender, subject, or identified learning styles. The goal is to tailor lessons effectively, making learning more meaningful and engaging for students. Moreover, tools like ChatGPT can provide automated grading support, offering immediate assessments and feedback to students. This capability helps in generating a "robust dataset for teachers to analyze and better differentiate student learning levels," further integrating data into the assessment cycle (Campesato, 2023 ). This synergy facilitates a paradigm shift in educational assessment, transitioning from a predominantly summative, static evaluation to a formative, continuous feedback loop. Presumably, this integration of technology allows assessment to become a living, breathing component of instruction, rather than a separate, often delayed, activity, enabling a true "data-driven instruction" model where observations lead directly to improved teaching and learning. Despite the benefits of these technologies, they are still limited studies, and some of them focused on the how learners create and convey data visualizations (e.g., Binali et al., 2022 ; Börner et al., 2016 ; Lee et al., 2017 ). That is why these prompted researchers to: Objectives: To map and describe the characteristic epistemic networks representing how non-programming specialist teachers engage with and make sense of the synergies between Python and ChatGPT for AI-powered data visualization. To provide empirical direction, derived from the observed epistemic networks, for the design of socially-oriented data discussion map and upskilling initiatives for teachers. Methodology Quantitative Ethnography (QE) is grounded in the integration of qualitative and quantitative traditions, tracing its roots to the ethnographic study of meaning-making and the statistical modeling of those meanings. Theoretically, QE holds that human learning, expertise, and decision-making are best understood by examining the connections between concepts in context, not simply their isolated occurrences (Bowman et al., 2021 ). This stance aligns with the sociocultural view that knowledge is constructed through discourse and action, where meaning emerges from patterned relationships between ideas. In this study, QE’s adaptability made it possible to merge structured coding from transcribed teacher–AI interactions with the mathematical modeling of those connections through Epistemic Network Analysis (ENA), thereby preserving the richness of teachers’ discourse while enabling robust, reproducible analysis. ENA is originally designed to model networks of discourse emphasizing a mathematical method to represent individuals epistemic frames as comprehensive networks, including their epistemologies and beliefs, in an interpretable format (Csanadi et al., 2018 ; Goldfarb Cohen et al., 2024 ). Table II. Inclusion and exclusion criteria Inclusion criteria Exclusion criteria • Currently employed as a teacher. Participants must be actively teaching in one of the three participating private schools in Quezon during the study period. • A minimum of 2 years of teaching experience in a K-12 setting and frequent in using charts, graphs for their reports. • Willingness to engage with Technology. expressing a willingness to learn and regularly interact with new digital tools like Python (through AI guidance) and ChatGPT for data visualization. • Access to necessary devices. Confirmed access to a personal or school-provided computer/laptop with internet connectivity for the duration of the study. • Extensive prior programming experience . Teachers with more than 6 months of formal experience in Python programming (or similar object-oriented languages like Java, C++) are excluded. • Prior Advanced Data Visualization Tool Experience . Teachers with documented expertise (e.g., professional certification, regular use in their role) in advanced data visualization software beyond basic spreadsheet charting (e.g., Tableau, Power BI, D3.js) are excluded. • Administrative-only roles . Employed primarily in administrative roles with no regular teaching duties. • Inability to meet technical requirements . Experiencing chronic internet issues, incompatible devices) that prevent their consistent participation. The prompting techniques for a data visualizer using ChatGPT go beyond just asking for code, although has limitations (Skvortsova et al., 2025 ). They involve a structured, iterative approach that guides the AI through the entire data visualization process, from initial data understanding to the final design and interpretation. Table III. Prompting techniques used in the study Technique Description Example Prompts Contextualization Sets the stage by giving ChatGPT a specific role and objective. "Act as a data visualization expert and Python coder. My goal is to create a bar chart from this CSV data to show the number of students per grade level." Iteration Breaks the process into smaller, manageable steps rather than asking for a final product all at once. Initial Code: "Write the Python code to create a bar chart...". Debugging: "I ran the code, but I got a 'ValueError.' How can I fix this?". Customization: "The chart looks good, but the labels are overlapping. Can you add code to rotate them?" Critical Thinking Uses prompts to evaluate the AI's output and extract meaning from the visualization. Verification: "Can you explain what each line of the Python code you generated does?". Analysis: "Based on this chart, what are three potential insights I could ask my students?". Critique: "What is a better chart type to visualize this data, and why?" Interpretation and Criteria for the Three Tiers To provide empirical evidence for the three-tiered learning progression adapted from the Dreyfus Model (2004), a Latent Class Analysis (LCA) was conducted on the teachers' coded strategies. The analysis aimed to identify the optimal number of distinct, unobserved groups (latent classes) that best explain the observed patterns of strategy use among the participants. Model fit statistics, including the Bayesian Information Criterion (BIC), were used to compare a series of models within classes. The three-class model was found to be the most parsimonious and best-fitting model (BIC = 211.3), providing strong statistical justification for the existence of three distinct profiles of strategy use. The analysis revealed three empirically-derived classes, which were subsequently named to align with theoretical framework: Class 1. Non-Level Strategies. This class, representing approximately 35% of the participants (n = 7), was empirically characterized by a high probability of engaging in "Non-Level" strategies. The members of this class exhibited a high likelihood of trial and error (probability = 0.85) and external reliance (probability = 0.79). Concurrently, they showed a very low probability of employing more sophisticated strategies like proactive planning (probability < 0.05) or critical evaluation (probability < 0.03). This profile strongly aligns with the Novice stage of skill acquisition, where problem-solving is reactive and relies heavily on external guidance and unsystematic attempts. Class 2. Basic-Level Strategies. This class, comprising approximately 45% of the participants (n = 9), was defined by a moderate-to-high probability of using "Basic-Level" strategies. Individuals in this group were most likely to engage in systematic troubleshooting (probability = 0.89) and directed prompts (probability = 0.82). While they still showed a moderate probability of engaging in some "Non-Level" behaviors, their profile was distinguished by a significant and deliberate effort to diagnose and solve problems within the tool's ecosystem. This profile represents the Competent stage, where teachers have begun to internalize the rules and can apply them with emerging situational awareness, though they have not yet fully achieved mastery. Class 3. High-Level Strategies. This class, accounting for the remaining 20% of the participants (n = 4), was characterized by a high probability of using "High-Level" strategies. The profile for this group demonstrated a strong likelihood of critical evaluation (probability = 0.91) and proactive planning (probability = 0.88), as well as a high probability of customization and contextualization (probability = 0.75). This class showed a near-zero probability of using "Non-Level" strategies, indicating a fundamental shift in their approach. The empirically derived profile of this class provides strong evidence for the Proficient/Expert stage of skill acquisition, where teachers operate with a deep and intuitive understanding of the tools to achieve complex, purpose-driven goals. The results of the LCA thus provide a statistical foundation for qualitative findings, confirming that the teachers' engagement with AI-powered data visualization progressed through three distinct, empirically supported stages, which are consistent with the theoretical tenets of the Dreyfus Model of Skill Acquisition. Table IV. Derived-class classification Tier 3: High-Level Strategies Criteria Keywords/Phrases in Data Mastery and an Integrated, High-Level Understanding Proactive Planning: Articulating a clear goal and using tools to systematically achieve it, anticipating challenges. "I need to make this clear for my students, so I'll ask it to..." Critical Evaluation: Explicitly questioning AI output, verifying code accuracy, or discussing ethical implications of a visualization. "This chart tells me X, so I can use it to talk about Y in class" Customization and Contextualization: Modifying visualizations for a specific audience (e.g., students, parents) or pedagogical purpose. "I need to check if this calculation is correct." Iterative Design: Engaging in a cycle of generating, critiquing, and refining visualizations for clarity and impact. Tier 2: Basic-Level Strategies Criteria Keywords/Phrases in Data Emerging Understanding and a Systematic Approach Systematic Troubleshooting: Engaging in a structured process to solve a problem (e.g., re-reading error messages, re-prompting ChatGPT with slightly different wording). "I'm going to try to... " Internal Modification: Making small, deliberate changes to the code generated by ChatGPT to fix an error or get a different result. "I'll ask it to do this instead" Directed Prompts: Using more specific and detailed prompts that provide context or constraints. "It says this, so I need to change that." Connecting Concepts: Explicitly verbalizing the link between code and the visualization it produces. Tier 1: Non-Level Strategies Criteria Keywords/Phrases in Data Initial, Reactive, and Trial-and-Error Behaviors Simple Trial and Error: Randomly changing prompts or code without a clear understanding of the error. "It's not working" External Reliance: Immediately seeking help from others or simple Google searches without first trying to diagnose the problem. "I don't know what to do" Passive Use: Copying and pasting code without attempting to understand or modify its structure. "What should I type here?" Expressing Frustration: Discourse is dominated by frustration and confusion rather than a problem-solving process. ENA utilization To extract the audio from each video segment, we used the ffmpeg library in Python. This method enables the extraction of the audio stream from the video, saving the audio from each segment as an individual file. Subsequently, the extracted audio files were used to generate transcription data. In this work, we utilized ConceptNet, a rich repository of common sense knowledge available in multiple languages. ConceptNet is a graph that links words and phrases with labeled weights, providing a deeper understanding of their meanings (Liu & Singh, 2004 ). To extract common-sense information, the text data of utterances is first converted into tokens, which are then processed through the knowledge graph (KG). Each related word and its corresponding token are treated as nodes, interconnected by edges in the graph. Nodes are embedded using ConceptNet Numberbatch Embeddings to form a graph representation. To analyze the relationships between nodes in the graph, we used a Graph Convolutional Network (GCN), which learns the embedding relations among all nodes from two sessions. The ENA1.7.0 Web Tool was used to quantitatively process the codes in Excel format (Marquart et al., 2018 ). A codebook in Excel consisting of a list of the codes for all the participants was generated, with the occurrence of each strategy or difficulty coded as 1, and absence as 0. Results and Discussion Codes Prompt Engineering for Desired Output The difficulty in crafting effective, clear prompts for ChatGPT to generate the specific type of Python code or data analysis they need. Teachers may not know what to ask for, resulting in generic or incorrect outputs that don't meet their pedagogical or reporting needs. Technical Literacy and Syntax Errors The difficulty in understanding and correctly writing the Python code provided by ChatGPT. Teachers may struggle with basic syntax, indentation, and the structure of a programming language, leading to frequent errors that they don't know how to debug. Data Cleaning and Preparation The struggle with preparing their raw school data (i.e., from spreadsheets) for analysis. Teachers may face issues with missing values, inconsistent formats, or the need to merge different data sources, and they may not know how to instruct ChatGPT to handle these tasks effectively. Conceptualizing Visualizations The difficulty in deciding which type of chart or graph (e.g., bar chart, scatter plot, pie chart) is most appropriate to tell the story of their data. They may understand the data, but lack the expertise in data visualization principles to choose an effective display. Verification of AI-Generated Code A lack of confidence or a sense of apprehension about using code generated by ChatGPT without fully understanding it. Teachers may worry about potential errors, data security risks, or the accuracy of the visualization without the ability to verify the underlying code themselves. Integrating Data Insights The difficulty in moving from a raw data visualization to a meaningful classroom application. Teachers may struggle to translate a chart or graph into a teachable moment, a student activity, or a strategic decision for their school reports. Notes. Network Layout. The position of the nodes in the two-dimensional space is not random; it is determined by a statistical process that plots the concepts based on their co-occurrence. The closer two nodes are in the network, the more frequently they were discussed together, suggesting a stronger conceptual link. The overall position of the entire network (the group centroid) within this space indicates the dominant conceptual focus of the group at that time. Nodes. Each node in the network represents a single, distinct code or theme identified from the qualitative data. the nodes correspond to the specific challenges teachers faced in their engagement with AI-powered data visualization. Edges (The Lines). The lines connecting the nodes represent a conceptual connection or co-occurrence between the coded themes. Line Thickness. The thickness of a line indicates the strength of the connection between the two nodes it links. Blue Network. Represents the conceptual framework of the teachers during Session 1, at the beginning of their learning journey. Green Network. Represents the evolved conceptual framework of the teachers during Session 2, after three weeks of engagement. The ENA reveals a fundamental shift in the teachers' conceptual framework over the three-week period. In Session 1, the blue network's most salient connections were centered on foundational, technical hurdles, with thick lines linking "Prompt engineering," "Data cleaning & preparation," and "Syntax errors." This indicates that teachers' initial struggles were dominated by a linear workflow of technical problems and solutions. By Session 2, the green network shows a dramatic re-structuring; the original strong connections have weakened, and a new, highly cohesive cluster has formed. The thickest lines now link "Conceptualizing visualizations," "Verification of AI-Generated Code," and "Integrating Data insights," signifying a shift in focus from low-level tool mechanics to higher-order pedagogical concerns. This shift indicates that the teachers have moved beyond the low-level, technical problems and are now wrestling with the higher-order, pedagogical challenges. They are no longer focused on "how to make the code run" but on "what chart should I make," "is this code and its output correct," and crucially, "how can I use this to teach." Of note, the strategies employed by the teachers in this study were categorized into three distinct tiers, representing a progression of skill acquisition and expertise adapted from the Dreyfus Model of Skill Acquisition (Dreyfus, 2004 ; Zhang & Wang, 2021 ). The result of the ENA provides an empirical account of how teachers' conceptual frameworks for AI-powered data visualization evolved over a three-week period interval from the data gathering. The blue network, representing the initial state in Session 1, is characterized by a central focus on foundational, technical challenges. The most prominent connections exist between 'Prompt engineering,' 'Data cleaning & preparation,' and 'Syntax errors,' indicating that teachers' initial discourse and problem-solving were dominated by the basic, mechanical hurdles of getting the tools to function. The group centroid for this session is located on the left side of the conceptual space, reflecting this technical-centric frame. A one-sample t-test was performed to assess the change between conditions. The results revealed a statistically significant difference along the X-axis (t(55.83) = 2.93, p < 0.01). The magnitude of this shift is substantial, as indicated by a large effect size (Cohen′s d = 0.93). This quantitative evidence confirms that the teachers' conceptual frame underwent a profound transformation from Session 1 to Session 2. The green network, representing the evolved state in Session 2, illustrates this shift. The group centroid has moved significantly to the right, signifying a new conceptual orientation. In this evolved frame, the strongest connections have coalesced around higher-order, pedagogical concerns. The most salient linkages now exist between 'Conceptualizing visualizations,' 'Verification of AI-Generated Code,' and 'Integrating Data insights.' This indicates that as teachers overcame the initial technical challenges, their focus elevated from tool mechanics to the critical and meaningful application of the data. The lack of a significant difference along the Y-axis suggests that this change was not a random reordering, but a focused progression along a primary dimension of their learning. When comparing two group centroids, if their 95% confidence intervals do not overlap, it provides strong visual evidence that the two groups are statistically distinct in their conceptual space (Bowman et al., 2021 ). In conclusion, the ENA models and supporting statistical analysis demonstrate a clear and substantial shift in the teachers' epistemic framework. They transitioned from a conceptual space dominated by technical difficulties to one focused on the strategic, critical, and pedagogical application of data. This finding provides a visual and empirical narrative of teacher empowerment, highlighting that true proficiency with AI-powered tools is achieved not by eliminating technical challenges, but by elevating one's engagement to the level of meaningful, contextualized application. This has significant implications for the design of future upskilling initiatives and educational tools, which must provide support for this critical conceptual leap. Reflexivity While the synergy between Python and ChatGPT offers transformative edge for data visualization in education, it is not without inherent risks, if wielded without proper understanding, can lead to unintended consequences. One observation was the tendency for some teachers to engage in "Vibe Coding." They would copy code generated by the AI and paste it directly into their environment without first checking it for logical errors or data integrity issues. Also, the visualizations often had subpar aesthetics due to poor whitespace management, resulting in cluttered charts that were difficult to interpret. The study highlights a fundamental ethical blind spot in the technology itself. The tool does not prevent the creation of misleading or biased visualizations, placing the full responsibility for ethical practice on the user. Another critical concern is the "black box" phenomenon associated with AI code generation. Relying on a tool like ChatGPT to generate code can create a scenario where the user does not fully comprehend the underlying logic or the intricate processes by which the AI arrives at its outputs. For example, if the AI is trained on biased data, its outputs, and consequently the visualizations, may perpetuate those biases without the user's awareness. Given these risks, the research emphasizes a strong imperative for comprehensive training. This training should extend beyond the mere technical skills of using Python and ChatGPT for visualization. It must focus on the principles of data literacy, ethical data visualization, and the ability to critically interpret visual data, encouraging a healthy skepticism toward statistics, promoting analytical thinking and the ability to identify potential biases or misleading claims (Lester, 2022 ). Furthermore, it must instill an understanding of ethical data practices, emphasizing transparency, fairness, and the avoidance of manipulating data or perpetuating stereotypes. Conclusion The integration of Python and ChatGPT for data visualization marks a pivotal in educational and technological possibilities, this paper opens inquiries related to the challenges of limited visualization literacy, lack of access to user-friendly tools, data fragmentation, and time constraints that have hindered educators from fully leveraging data for instructional improvement. However, the power of this technology necessitates a comprehensive data literacy training. Without a deep understanding of data principles, ethical visualization practices, and critical interpretation skills, there is a substantial risk of creating misleading visuals or blindly accepting AI-generated outputs. The "black box" nature of AI code generation unveils underlying logic to ensure accuracy, fairness, and accountability. Therefore, the full potential of this synergy can only be realized through robust professional development programs that equip teachers not just with technical proficiency, but with the critical thinking and ethical frameworks necessary to relate the data in education responsibly. References Adarsh Bhavimane, Rakshitha Shetty, Ghrutha Varsha Kurunji, Alex Tayenjam, & Dr. Pushparani M K. (2024). Data Visualization in Education: A Comprehensive Review. International Journal of Advanced Research in Science Communication and Technology , 503–509. https://doi.org/10.48175/IJARSCT-18676 Binali, T., Chang, C.-H., Chang, Y.-J., & Chang, H.-Y. (2022). High school and college students’ graph-interpretation competence in scientific and daily contexts of data visualization. Science & Education . https://doi.org/10.1007/s11191-022-00406-3 Börner, K., Maltese, A., Balliet, R. N., & Heimlich, J. (2016). Investigating aspects of data visualization literacy using 20 information visualizations and 273 science museum visitors. Information Visualization, 15 (3), 198–213. https://doi.org/10.1177/1473871615594652 Bowman, D., Swiecki, Z., Cai, Z., Wang, Y., Eagan, B., Linderoth, J., Shaffer, D. W., Ruis, A. R., & Lee, S. B. (2021). The Mathematical Foundations of Epistemic Network Analysis. In Advances in Quantitative Ethnography (Vol. 1312, pp. 91–105). Springer International Publishing AG. https://doi.org/10.1007/978-3-030-67788-6_7 Box, A. (2023). A Power BI Compendium: Answers to 65 Commonly Asked Questions on Power BI (1st ed.). Apress. https://doi.org/10.1007/978-1-4842-9765-0 Campesato, O. (2023). Python 3 Data Visualization Using ChatGPT/GPT-4 . Campesato, O. (2024). ChatGPT and Data Visualization. In Python 3 and Machine Learning Using ChatGPT/GPT-4 (pp. 1–1). Mercury Learning and Information. https://doi.org/10.1515/9781501520112-009 Castillo, W., & Strunk, K. K. (2025). How to QuantCrit: Applying Critical Race Theory to Quantitative Data in Education . Taylor & Francis. https://doi.org/10.4324/9781003429968 Chen, Z., Zhang, C., Wang, Q., Troidl, J., Warchol, S., Beyer, J., Gehlenborg, N., & Pfister, H. (2023). Beyond generating code: Evaluating GPT on a data visualization course. In 2023 IEEE VIS Workshop on Visualization Education, Literacy, and Activities (EduVis) (pp. 16–21). Csanadi, A., Eagan, B., Kollar, I. (2018). When coding-and-counting is not enough: using epistemic network analysis (ENA) to analyze verbal data in CSCL research. Intern. J. Comput.-Support. Collab. Learn 13, 419–438. https://doi.org/10.1007/s11412-018-9292-z Dreyfus, S. E. (2004). The Five-Stage Model of Adult Skill Acquisition. Bulletin of Science, Technology & Society , 24 (3), 177-181. https://doi.org/10.1177/0270467604264992 (Original work published 2004) Glasgow, N. A., & Farrell, T. S. C. (2007). What successful literacy teachers do : 70 research-based strategies for teachers, reading coaches, and instructional planners . Corwin Press. Goldfarb Cohen, S., Yanai, J. V., & Dishon, G. (2024). Modeling group discourse with epistemic network analysis: Unpacking connections, perspectives, and individual contributions. Journal of Science Education and Technology . https://doi.org/10.1007/s10956-024-10139-3 Grand Bang, C. (2024). Data Visualization and Presentation: How to Present and Communicate Data Effectively for Decision-Making. In Data-Driven Decision-Making for Business . Taylor & Francis Group. https://doi.org/10.4324/9781003457787-6 Hwang, M., Lee, K.-H., & Lee, H.-K. (2025). A word to the wise: Crafting impactful prompts for ChatGPT. System , 133 , 103756. https://doi.org/10.1016/j.system.2025.103756 Jackson, A. (2025). Learning AI tools in Tableau: Level up your data analytics and visualization capabilities with Tableau Pulse and Tableau Agent . O’Reilly Media, Inc. Jiang, T., Liu, E., Baig, T., & Li, Q. (2024). Enhancing decision‐making in higher education: Exploring the integration of ChatGPT and data visualization tools in data analysis. New Directions for Higher Education , 2024 (207), 15–29. https://doi.org/10.1002/he.20510 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 Kim, N. W., Ko, H.-K., Myers, G., & Bach, B. (2024). ChatGPT in Data Visualization Education: A Student Perspective. Proceedings of IEEE Symposium on Visual Languages and Human-Centric Computing, VL/HCC , 109–120. https://doi.org/10.1109/VL/HCC60511.2024.00022 Krejci, S. E., Ramroop-Butts, S., Torres, H. N., & Isokpehi, R. D. (2020). Visual Literacy Intervention for Improving Undergraduate Student Critical Thinking of Global Sustainability Issues. Sustainability , 12 (23), 10209. https://doi.org/10.3390/su122310209 Lawn, M. (2013). The rise of data in education systems: Collection, visualization and uses . Symposium Books. Lee, H. S., Mojica, G. F., Thrasher, E. P., & Baumgartner, P. (2022). Investigating data like a data scientist: Key practices and processes. Statistics Education Research Journal, 21(2), Article 3. https://doi.org/10.52041/serj.v21i2.41 Lee, S., Kim, S. H., & Kwon, B. C. (2017). VLAT: Development of a visualization literacy assessment test. IEEE Transactions on Visualization and Computer Graphics, 23 (1), 551–560. https://doi.org/10.1109/tvcg.2016.2598920 Lester, P. M. (2022). Data Visualization Ethics. In Visual Ethics (2nd ed., pp. 115–125). Routledge. https://doi.org/10.4324/9781003243045-11 Liu, H., & Singh, P. (2004). ConceptNet — A Practical Commonsense Reasoning Tool-Kit. BT Technology Journal , 22 (4), 211–226. https://doi.org/10.1023/B:BTTJ.0000047600.45421.6d Maddigan, P., & Susnjak, T. (2023). Chat2VIS: Generating Data Visualisations via Natural Language using ChatGPT, Codex and GPT-3 Large Language Models. IEEE Access , 11 , 1–1. https://doi.org/10.1109/ACCESS.2023.3274199 Makowski, M. B., & Lubienski, S. T. (2023). Classroom Data Visualization: Tracking Individuals During Group-Centered Instruction. Educational Researcher , 52 (3), 164–169. https://doi.org/10.3102/0013189X231158374 Marquart, C. L., Hinojosa, C., Swiecki, Z., Eagan, B., & Shaffer, D. W. (2018). Epistemic Network Analysis (Version 1.5.2) [Software] Perkhofer, L., Walchshofer, C., & Hofer, P. (2020). Does design matter when visualizing Big Data? An empirical study to investigate the effect of visualization type and interaction use. Journal of Management Control , 31 (1–2), 55–95. https://doi.org/10.1007/s00187-020-00294-0 Skvortsova, S., Symonenko, T., & Hnezdilova, K. (2025). Optimizing Routine Educational Tasks through Prompt Engineering: A Comparative Study of AI Chatbots. Proceedings of International Conference on Applied Innovation in IT, 13 (1), 35-42. Sorapure, M. (2019). Text, Image, Data, Interaction: Understanding Information Visualization. Computers and Composition , 54 , 102519. https://doi.org/10.1016/j.compcom.2019.102519 Zhang, S., & Wang, N. (2021). Dreyfus’ Model of Skill Acquisition from the Perspective of Phenomenology. Journal of Engineering Studies , 13 (4), 353–361. https://doi.org/10.3724/SP.J.1224.2021.00353 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8023714","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":539512146,"identity":"fa02cebe-ebf9-48c3-a852-e5fa3ab6f175","order_by":0,"name":"CARIE JUSTINE ESTRELLADO","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9klEQVRIiWNgGAWjYBADGT4GBgOGD2B2AgMDYwMB9QcYGHjYgFoYZ5CshZmHGC38s5sPf/64x4aHjf3wxse2bXcY+NlzDJgLd+DWInHnWJrEgWdpPGw8acXGuW3PGCR73hgwzzyDW4uBRI4Zw4EDh4EOyzGTzm07zGBwA2gLbxs+LfmfPxw48J+Hjf+NmbQlUIs9YS05DBIHDhzgYQNaJ80IskWCgBagX8wkzhxIBmp5VmzYc+4Zj8SZZwWHZ+LRAgyxxx8qDtjJ8fMnb3zwo+yOHH978sbHhXi0MEigcg+Ao+YwHg2YWsAkM14to2AUjIJRMNIAAIcDT89JOJoYAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-0495-7796","institution":"","correspondingAuthor":true,"prefix":"","firstName":"CARIE","middleName":"JUSTINE","lastName":"ESTRELLADO","suffix":""}],"badges":[],"createdAt":"2025-11-04 03:00:56","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":true,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8023714/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8023714/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95189709,"identity":"9f552a84-7f17-4e6b-9cc7-19d3b7a3914b","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1359023,"visible":true,"origin":"","legend":"","description":"","filename":"PythonandChatGPTAIPoweredDataVisualizationforTeachersCopy.docx","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/a749159166e8512f0e26594b.docx"},{"id":95189704,"identity":"edf5e61c-5c49-41cf-a726-464a8c8c5ea1","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8023714.json","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/32f609095b1ed28280abf9ae.json"},{"id":95189710,"identity":"e8602075-bc27-45f9-9b7b-258deadae73f","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":97247,"visible":true,"origin":"","legend":"","description":"","filename":"rs80237140enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/a7563431728540eb1e87ef8e.xml"},{"id":95189707,"identity":"420a7e7f-448f-4b9e-b430-453e1768820d","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":829555,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/a93e90dbe522f13f799862f9.png"},{"id":95189706,"identity":"0b8dd2d8-b2a9-4d50-98d0-65a15c50f49d","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":475482,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/7b3ed8f1c9f7c2d4e8cc8b74.png"},{"id":95189705,"identity":"4954cb81-bfcd-4d40-96cf-0e5aff4d952c","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":72191,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/0f3f679d2455f598d709452d.png"},{"id":95189708,"identity":"66df1696-f1d5-4add-9dfd-d50f681077f3","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26704,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/0c2e3775bf1f0c3cc38290de.png"},{"id":95228654,"identity":"0bf7688a-94ba-4ac1-9b9b-dd574144a653","added_by":"auto","created_at":"2025-11-05 16:34:01","extension":"xml","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":95698,"visible":true,"origin":"","legend":"","description":"","filename":"rs80237140structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/a8c42e34cfcc145a40582a9e.xml"},{"id":95189712,"identity":"41088099-9757-4049-8dca-dfa8754dfd23","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"html","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":102307,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/ad8bba890d18fe42756b37c8.html"},{"id":95189703,"identity":"ff445d07-d336-4a5f-9a34-04d3afc038c0","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":150469,"visible":true,"origin":"","legend":"\u003cp\u003eUnnumbered image in the Results and Discussion section.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/0b66d39d3791cd3972dc5dfc.png"},{"id":95189702,"identity":"cbedc649-7900-407c-81b1-5f92e705002f","added_by":"auto","created_at":"2025-11-05 10:03:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":122856,"visible":true,"origin":"","legend":"\u003cp\u003eUnnumbered image in the Results and Discussion section.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/35715d817d4d0ef1435a5d46.png"},{"id":95230785,"identity":"64ed39f7-9e74-4922-b92d-fdd26e68bb90","added_by":"auto","created_at":"2025-11-05 16:38:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":968773,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8023714/v1/818329b1-53a4-42e0-a9da-c6233b4c4449.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003ePython and ChatGPT: AI-Powered Data Visualization for Teachers\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eEducation, like many sectors, is reliant on data to inform strategies, enhance teaching outcomes, and cultivate improved learning environments. But data are not the same from everyone to be seen (Castillo \u0026amp; Strunk, 2025). This comes the premise for visualization where people can process images over texts (Lawn, 2013; Perkhofer et al., 2020). Catapulting statement towards data visualization where it is paramount in using visual representations such as charts, graphs, and interactive dashboards to analyze a wide array of data, including student performance, attendance patterns, demographic information, and learning behaviors. This visual approach makes data easier to digest and allows decision-makers to quickly grasp critical indicators, streamlining operations and facilitating timely interventions when indicators deviate from expectations (Grand Bang, 2024; Makowski \u0026amp; Lubienski, 2023; Sorapure, 2019).\u003c/p\u003e\n\u003cp\u003eThe purpose of this approach is to transform complex datasets into understandable and actionable information for educators and administrators, enabling them to identify patterns, trends, and correlations that might otherwise be obscured in raw numbers and spreadsheets. Despite the clear advantages, teachers frequently encounter significant challenges when attempting to upskill data visualization (Lawn, 2013). A primary obstacle is a widespread lack of visualization literacy (VL) (Glasgow \u0026amp; Farrell, 2007; Krejci et al., 2020; Lawn, 2013; Perkhofer et al., 2020), meaning many educators do not possess the technical skills required to effectively interpret or create data visualizations. This deficiency can lead to superficial or even erroneous interpretations of data, which can negatively influence instructional decisions and, consequently, student outcomes. Furthermore, access to user-friendly tools remains a barrier. While powerful data visualization platforms like Tableau (Jackson, 2025) and Power BI exist (Box, 2023), their complexity and cost often make them inaccessible to teachers who are not trained data scientists or programmers. The fragmentation of educational data across disparate systems, known as data silos, further complicates efforts to collect, integrate, and visualize information cohesively (Adarsh Bhavimane et al., 2024; Perkhofer et al., 2020). Compounding these issues are severe time constraints; teachers have limited time for data analysis, and traditional visualization methods are often too time-consuming for daily application.\u003c/p\u003e\n\u003cp\u003eThe traditional landscape of data visualization for teachers typically involved manual creation of basic charts and graphs using spreadsheet software like Microsoft Excel (Sorapure, 2019). \u0026nbsp;While ubiquitous, these methods are inherently static and has limited interactivity, and are time-consuming to create and update, especially with large datasets. Such conventional approaches are often insufficient for uncovering complex patterns and relationships within educational data. The contemporary trend is rapidly shifting towards more interactive, dynamic, and automated dashboards and visualizations that transcend simple charts to construct a compelling narrative with the data (Kim et al., 2024).\u0026nbsp;\u003c/p\u003e"},{"header":"2. Bridging the Technical Gap for Non-Programming Teachers","content":"\u003cp\u003eThe convergence of Python's robust capabilities and ChatGPT's intuitive accessibility represents a significant advancement for educators who are not programming specialists. This development empowers individuals with limited coding skills to readily create data visualization dashboards, transforming raw data into meaningful insights. Large Language Models (LLMs) like ChatGPT serve as dynamic resources for educators, assisting not only with understanding advanced concepts but also with solving complex problems While some perspectives suggest that exclusive reliance on ChatGPT might impede deep coding comprehension, the primary utility of this synergy for teachers lies in enabling the \u003cem\u003eapplication\u003c/em\u003e of data visualization rather than demanding fundamental programming mastery (Campesato, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kim et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This allows educators to reallocate their cognitive effort from debugging syntax and navigating complex programming environments to focusing on pedagogical implications, critical data interpretation, and the direct impact on student learning.\u003c/p\u003e\u003cp\u003ePython, with its rich ecosystem of libraries, stands as a formidable language for data analysis and visualization. Libraries such as Matplotlib, Seaborn, and Plotly offer extensive customization options and the capacity to manage large and intricate datasets. Matplotlib, for instance, is a foundational library capable of generating static, animated, and interactive 2D plots, including line, bar, scatter, and pie charts. Seaborn, built upon Matplotlib, simplifies the creation of aesthetically pleasing statistical plots and integrates seamlessly with pandas DataFrames, offering specialized visualizations like heatmaps (Campesato, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Plotly, a versatile open-source library, excels at creating interactive, web-based visualizations and offers a wide array of 2D and 3D chart types, integrating well with Jupyter notebooks. These libraries collectively provide the computational backbone for advanced data visualization.\u003c/p\u003e\u003cp\u003eChatGPT's Advanced Data Analysis (ADA) feature functions as a simplified Jupyter Notebook directly within the chat interface, creating what can be considered a \"conversational IDE\". Teachers can upload CSV files or other data formats and instruct ChatGPT to generate and execute Python code, producing charts and analyses directly within the chat environment. This provides a user-friendly setting for quick prototyping and analysis, allowing for immediate visual feedback on data. The environment comes pre-loaded with hundreds of Python libraries, enabling ChatGPT to write and execute code for complex tasks (Campesato, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis interactive environment supports an iterative development process. Teachers can refine their visualizations by providing follow-up prompts to ChatGPT, such as \"Change the color of the bars to blue\" or \"Add a title to the chart\". This continuous refinement process allows for ongoing adjustments without requiring extensive coding knowledge. This dynamic interaction, where teachers can ask for \"more examples,\" or even \"that doesn't sound right, try again,\" demonstrates that the process is an ongoing dialogue rather than a single query-response. This iterative capability is a significant pedagogical advantage, allowing teachers to experiment and refine their visualizations in real-time without the pressure of achieving perfection on the first attempt, fostering a more exploratory approach to data analysis.\u003c/p\u003e\u003cp\u003eThis synergy bridges the gap between the power of a programming language and the non-technical user. It enables teachers to leverage Python's extensive capabilities without needing to become Python experts, thereby democratizing access to sophisticated data analysis tools in education. This shift allows teachers to focus their efforts on the \u003cem\u003ewhat\u003c/em\u003e and \u003cem\u003ewhy\u003c/em\u003e of data analysis\u0026mdash;the pedagogical application and interpretation\u0026mdash;rather than the \u003cem\u003ehow\u003c/em\u003e of coding mechanics, representing a more appropriate and efficient allocation of their professional expertise.\u003c/p\u003e"},{"header":"3. Crafting Prompts for ChatGPT","content":"\u003cp\u003eTo effectively harness the Python-ChatGPT synergy, teachers must develop proficiency in prompt engineering. Prompts serve as the primary interface for interaction, and their clarity and precision are paramount to achieving the desired visualization outcomes.\u003c/p\u003e\u003cp\u003eThe principles of effective prompt engineering for data visualization emphasize clarity and specificity. It is crucial to craft prompts that are clear, specific, and provide sufficient context for ChatGPT to accurately understand the request (Kasneci et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Ambiguity should be avoided, and precision in defining desired details is paramount. For instance, specifying the exact columns and desired chart type is significantly more effective than a vague request. This means teachers need to think critically about the data they possess and the specific questions they want to answer visually (Chen et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kasneci et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003ePrompt engineering is inherently an iterative process where users should begin with an initial prompt, review ChatGPT's response, and then refine the prompt based on the output (Chen et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Maddigan \u0026amp; Susnjak, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This involves adjusting wording, adding more context, or simplifying the request as needed to improve the results. This iterative dialogue allows for continuous refinement, much like a collaborative process with a human data analyst.\u003c/p\u003e\u003cp\u003eExplicitly stating the desired visualization type, such as \"create a bar chart\" or \"generate an interactive dashboard,\" guides ChatGPT more effectively toward the intended visual representation. Furthermore, for ChatGPT to perform accurate analysis and visualization, it must understand the dataset. This involves uploading the relevant CSV or Excel file and, if necessary, providing a brief description of the data to confirm ChatGPT's interpretation. This ensures that the AI's understanding aligns with the teacher's intent, mitigating potential misinterpretations of the data.\u003c/p\u003e\u003cp\u003eEffective prompt engineering for data visualization effectively becomes the \"new coding\" for non-technical users. This shifts the primary skill requirement from mastering programming syntax to precisely articulating analytical intent and understanding data structure. This subtle but critical shift inherently demands a foundational level of data literacy from teachers, as they must think critically about their data and the specific insights they wish to extract, even if they are not writing the code themselves.\u003c/p\u003e\u003cp\u003eThe following table provides practical examples of prompts that teachers can use to generate various data visualizations and perform related data tasks.\u003c/p\u003e\u003cp\u003eTable I. Example ChatGPT Prompts for Educational Data Visualizations\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVisualization Type/Task\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePrompt Example\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePurpose for Teacher\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSimple Bar Chart\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"I have a CSV file named student_scores.csv with columns 'student_name' and 'math_score'. Please create a bar chart showing the math scores for each student.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eQuickly visualize individual student performance on a single academic metric for immediate understanding.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eScatter Plot with Trendline\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"Using the student_data.csv file with columns 'hours_studied' and 'exam_grade', plot a scatter plot to show the relationship between the two. Add a trendline to the plot.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExplore potential correlations or relationships between two continuous variables (e.g., study time and academic achievement).\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInteractive Dashboard\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"Create an interactive dashboard with two charts from student_attendance.csv: a bar chart showing average attendance by grade level, and a line chart showing attendance trends over the last month. Make the dashboard filterable by grade.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMonitor multiple key educational metrics simultaneously, allowing for dynamic exploration and drill-down analysis.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGrouped Bar Chart\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"From the class_data.csv file with columns 'gender', 'subject', and 'average_grade', create a grouped bar chart that compares the average grades of male and female students in each subject.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFacilitate comparisons of performance across different student subgroups to identify disparities and inform targeted instructional strategies.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eData Cleaning\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"Clean this data and remove any errors that can affect the output quality.\" OR \"Remove rows from student_data.csv where 'grade' is missing or below 0, and rename the 'student_ID' column to 'ID'.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrepare raw data for accurate analysis by removing inconsistencies and standardizing format.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRefinement/Customization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\"Change the color of the bars to blue and add a title 'Math Scores by Student' to the chart.\"\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTailor the visual output to meet specific presentation needs or enhance clarity and aesthetic appeal.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe specificity and variety of these prompt examples demonstrate that while teachers no longer need to know \u003cem\u003ehow\u003c/em\u003e to write Python code, they absolutely need to know \u003cem\u003ewhat\u003c/em\u003e they want to analyze, \u003cem\u003ewhat\u003c/em\u003e kind of relationships they are looking for, and \u003cem\u003ehow\u003c/em\u003e they want the data to be represented visually. This shifts the pedagogical emphasis from technical mechanics to a deeper understanding of data-driven inquiry and the principles of effective data representation (Perkhofer et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This capability empowers teachers to move beyond basic data consumption to active data manipulation and inquiry, provided they have the conceptual understanding to formulate precise questions and interpret the results.\u003c/p\u003e"},{"header":"4. Benefits for Teachers in Assessment and Instruction","content":"\u003cp\u003eThe accessible data visualization capabilities unlocked by the Python-ChatGPT synergy offer profound benefits for teachers in their daily assessment and instructional practices. This approach empowers educators to transform their use of data, moving beyond traditional, static methods to more dynamic and responsive strategies (Campesato, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe ability to easily visualize data empowers teachers to quickly identify individual students who are struggling by analyzing test scores over time, enabling early and targeted intervention. Visualizations are highly effective at revealing broader trends in student performance, such as a decline in a specific skill across the class or a general improvement over a period (Jiang et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The rapid visualization of longitudinal and aggregated student data fundamentally shifts teachers from reactive assessment\u0026mdash;merely grading after the fact\u0026mdash;to proactive, formative intervention (Perkhofer et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This represents a crucial pedagogical evolution from simply \u003cem\u003emeasuring\u003c/em\u003e learning outcomes to actively \u003cem\u003einforming\u003c/em\u003e and \u003cem\u003eadapting\u003c/em\u003e instructional strategies in real-time.\u003c/p\u003e\u003cp\u003eFurthermore, teachers can effectively utilize data visualizations to compare the academic performance of various student groups, such as those differentiated by gender, subject, or identified learning styles. The goal is to tailor lessons effectively, making learning more meaningful and engaging for students. Moreover, tools like ChatGPT can provide automated grading support, offering immediate assessments and feedback to students. This capability helps in generating a \"robust dataset for teachers to analyze and better differentiate student learning levels,\" further integrating data into the assessment cycle (Campesato, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This synergy facilitates a paradigm shift in educational assessment, transitioning from a predominantly summative, static evaluation to a formative, continuous feedback loop. Presumably, this integration of technology allows assessment to become a living, breathing component of instruction, rather than a separate, often delayed, activity, enabling a true \"data-driven instruction\" model where observations lead directly to improved teaching and learning.\u003c/p\u003e\u003cp\u003eDespite the benefits of these technologies, they are still limited studies, and some of them focused on the how learners create and convey data visualizations (e.g., Binali et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; B\u0026ouml;rner et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Lee et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). That is why these prompted researchers to:\u003c/p\u003e\u003cp\u003eObjectives:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eTo map and describe the characteristic epistemic networks representing how non-programming specialist teachers engage with and make sense of the synergies between Python and ChatGPT for AI-powered data visualization.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eTo provide empirical direction, derived from the observed epistemic networks, for the design of socially-oriented data discussion map and upskilling initiatives for teachers.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003eQuantitative Ethnography (QE) is grounded in the integration of qualitative and quantitative traditions, tracing its roots to the ethnographic study of meaning-making and the statistical modeling of those meanings. Theoretically, QE holds that human learning, expertise, and decision-making are best understood by examining the \u003cem\u003econnections\u003c/em\u003e between concepts in context, not simply their isolated occurrences (Bowman et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). This stance aligns with the sociocultural view that knowledge is constructed through discourse and action, where meaning emerges from patterned relationships between ideas. In this study, QE\u0026rsquo;s adaptability made it possible to merge structured coding from transcribed teacher\u0026ndash;AI interactions with the mathematical modeling of those connections through Epistemic Network Analysis (ENA), thereby preserving the richness of teachers\u0026rsquo; discourse while enabling robust, reproducible analysis. ENA is originally designed to model networks of discourse emphasizing a mathematical method to represent individuals epistemic frames as comprehensive networks, including their epistemologies and beliefs, in an interpretable format (Csanadi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Goldfarb Cohen et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTable II. Inclusion and exclusion criteria\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabb\" border=\"1\"\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eInclusion criteria\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExclusion criteria\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u0026bull; \u003cb\u003eCurrently employed as a teacher.\u003c/b\u003e Participants must be actively teaching in one of the three participating private schools in Quezon during the study period.\u003c/p\u003e\u003cp\u003e\u0026bull; A minimum of \u003cb\u003e2 years of teaching experience\u003c/b\u003e in a K-12 setting and frequent in using charts, graphs for their reports.\u003c/p\u003e\u003cp\u003e\u0026bull; \u003cb\u003eWillingness to engage with Technology.\u003c/b\u003e expressing a willingness to learn and \u003cem\u003eregularly interact\u003c/em\u003e with new digital tools like Python (through AI guidance) and ChatGPT for data visualization.\u003c/p\u003e\u003cp\u003e\u0026bull; \u003cb\u003eAccess to necessary devices.\u003c/b\u003e Confirmed access to a personal or school-provided computer/laptop with internet connectivity for the duration of the study.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u0026bull; \u003cb\u003eExtensive prior programming experience\u003c/b\u003e. Teachers with more than 6 months of formal experience in Python programming (or similar object-oriented languages like Java, C++) are excluded.\u003c/p\u003e\u003cp\u003e\u0026bull; \u003cb\u003ePrior Advanced Data Visualization Tool Experience\u003c/b\u003e. Teachers with documented expertise (e.g., professional certification, regular use in their role) in advanced data visualization software beyond basic spreadsheet charting (e.g., Tableau, Power BI, D3.js) are excluded.\u003c/p\u003e\u003cp\u003e\u0026bull; \u003cb\u003eAdministrative-only roles\u003c/b\u003e. Employed primarily in administrative roles with no regular teaching duties.\u003c/p\u003e\u003cp\u003e\u0026bull; \u003cb\u003eInability to meet technical requirements\u003c/b\u003e. Experiencing chronic internet issues, incompatible devices) that prevent their consistent participation.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe prompting techniques for a data visualizer using ChatGPT go beyond just asking for code, although has limitations (Skvortsova et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). They involve a structured, iterative approach that guides the AI through the entire data visualization process, from initial data understanding to the final design and interpretation.\u003c/p\u003e\u003cp\u003eTable III. Prompting techniques used in the study\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabc\" border=\"1\"\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTechnique\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDescription\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eExample Prompts\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eContextualization\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSets the stage by giving ChatGPT a specific role and objective.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"Act as a data visualization expert and Python coder. My goal is to create a bar chart from this CSV data to show the number of students per grade level.\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIteration\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBreaks the process into smaller, manageable steps rather than asking for a final product all at once.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eInitial Code: \"Write the Python code to create a bar chart...\". Debugging: \"I ran the code, but I got a 'ValueError.' How can I fix this?\". Customization: \"The chart looks good, but the labels are overlapping. Can you add code to rotate them?\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCritical Thinking\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eUses prompts to evaluate the AI's output and extract meaning from the visualization.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eVerification: \"Can you explain what each line of the Python code you generated does?\". Analysis: \"Based on this chart, what are three potential insights I could ask my students?\". Critique: \"What is a better chart type to visualize this data, and why?\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\n\u003ch3\u003eInterpretation and Criteria for the Three Tiers\u003c/h3\u003e\n\u003cp\u003eTo provide empirical evidence for the three-tiered learning progression adapted from the Dreyfus Model (2004), a Latent Class Analysis (LCA) was conducted on the teachers' coded strategies. The analysis aimed to identify the optimal number of distinct, unobserved groups (latent classes) that best explain the observed patterns of strategy use among the participants. Model fit statistics, including the Bayesian Information Criterion (BIC), were used to compare a series of models within classes. The three-class model was found to be the most parsimonious and best-fitting model (BIC\u0026thinsp;=\u0026thinsp;211.3), providing strong statistical justification for the existence of three distinct profiles of strategy use.\u003c/p\u003e\u003cp\u003eThe analysis revealed three empirically-derived classes, which were subsequently named to align with theoretical framework:\u003c/p\u003e\u003cp\u003eClass 1. Non-Level Strategies. This class, representing approximately 35% of the participants (n\u0026thinsp;=\u0026thinsp;7), was empirically characterized by a high probability of engaging in \"Non-Level\" strategies. The members of this class exhibited a high likelihood of trial and error (probability\u0026thinsp;=\u0026thinsp;0.85) and external reliance (probability\u0026thinsp;=\u0026thinsp;0.79). Concurrently, they showed a very low probability of employing more sophisticated strategies like proactive planning (probability\u0026thinsp;\u0026lt;\u0026thinsp;0.05) or critical evaluation (probability\u0026thinsp;\u0026lt;\u0026thinsp;0.03). This profile strongly aligns with the Novice stage of skill acquisition, where problem-solving is reactive and relies heavily on external guidance and unsystematic attempts.\u003c/p\u003e\u003cp\u003eClass 2. Basic-Level Strategies. This class, comprising approximately 45% of the participants (n\u0026thinsp;=\u0026thinsp;9), was defined by a moderate-to-high probability of using \"Basic-Level\" strategies. Individuals in this group were most likely to engage in systematic troubleshooting (probability\u0026thinsp;=\u0026thinsp;0.89) and directed prompts (probability\u0026thinsp;=\u0026thinsp;0.82). While they still showed a moderate probability of engaging in some \"Non-Level\" behaviors, their profile was distinguished by a significant and deliberate effort to diagnose and solve problems within the tool's ecosystem. This profile represents the Competent stage, where teachers have begun to internalize the rules and can apply them with emerging situational awareness, though they have not yet fully achieved mastery.\u003c/p\u003e\u003cp\u003eClass 3. High-Level Strategies. This class, accounting for the remaining 20% of the participants (n\u0026thinsp;=\u0026thinsp;4), was characterized by a high probability of using \"High-Level\" strategies. The profile for this group demonstrated a strong likelihood of critical evaluation (probability\u0026thinsp;=\u0026thinsp;0.91) and proactive planning (probability\u0026thinsp;=\u0026thinsp;0.88), as well as a high probability of customization and contextualization (probability\u0026thinsp;=\u0026thinsp;0.75). This class showed a near-zero probability of using \"Non-Level\" strategies, indicating a fundamental shift in their approach. The empirically derived profile of this class provides strong evidence for the Proficient/Expert stage of skill acquisition, where teachers operate with a deep and intuitive understanding of the tools to achieve complex, purpose-driven goals.\u003c/p\u003e\u003cp\u003eThe results of the LCA thus provide a statistical foundation for qualitative findings, confirming that the teachers' engagement with AI-powered data visualization progressed through three distinct, empirically supported stages, which are consistent with the theoretical tenets of the Dreyfus Model of Skill Acquisition.\u003c/p\u003e\u003cp\u003eTable IV. Derived-class classification\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabd\" border=\"1\"\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTier 3: High-Level Strategies\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCriteria\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKeywords/Phrases in Data\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eMastery and an Integrated, High-Level Understanding\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eProactive Planning: Articulating a clear goal and using tools to systematically achieve it, anticipating challenges.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"I need to make this clear for my students, so I'll ask it to...\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCritical Evaluation: Explicitly questioning AI output, verifying code accuracy, or discussing ethical implications of a visualization.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"This chart tells me X, so I can use it to talk about Y in class\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCustomization and Contextualization: Modifying visualizations for a specific audience (e.g., students, parents) or pedagogical purpose.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"I need to check if this calculation is correct.\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eIterative Design: Engaging in a cycle of generating, critiquing, and refining visualizations for clarity and impact.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTier 2: Basic-Level Strategies\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCriteria\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKeywords/Phrases in Data\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eEmerging Understanding and a Systematic Approach\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSystematic Troubleshooting: Engaging in a structured process to solve a problem (e.g., re-reading error messages, re-prompting ChatGPT with slightly different wording).\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"I'm going to try to... \"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eInternal Modification: Making small, deliberate changes to the code generated by ChatGPT to fix an error or get a different result.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"I'll ask it to do this instead\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDirected Prompts: Using more specific and detailed prompts that provide context or constraints.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"It says this, so I need to change that.\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eConnecting Concepts: Explicitly verbalizing the link between code and the visualization it produces.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTier 1: Non-Level Strategies\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCriteria\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eKeywords/Phrases in Data\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"3\" rowspan=\"4\"\u003e\u003cp\u003eInitial, Reactive, and Trial-and-Error Behaviors\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSimple Trial and Error: Randomly changing prompts or code without a clear understanding of the error.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"It's not working\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExternal Reliance: Immediately seeking help from others or simple Google searches without first trying to diagnose the problem.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e\"I don't know what to do\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePassive Use: Copying and pasting code without attempting to understand or modify its structure.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003e\"What should I type here?\"\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eExpressing Frustration: Discourse is dominated by frustration and confusion rather than a problem-solving process.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eENA\u003c/b\u003e utilization\u003c/p\u003e\u003cp\u003eTo extract the audio from each video segment, we used the \u003cem\u003effmpeg\u003c/em\u003e library in Python. This method enables the extraction of the audio stream from the video, saving the audio from each segment as an individual file. Subsequently, the extracted audio files were used to generate transcription data.\u003c/p\u003e\u003cp\u003eIn this work, we utilized ConceptNet, a rich repository of common sense knowledge available in multiple languages. ConceptNet is a graph that links words and phrases with labeled weights, providing a deeper understanding of their meanings (Liu \u0026amp; Singh, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). To extract common-sense information, the text data of utterances is first converted into tokens, which are then processed through the knowledge graph (KG). Each related word and its corresponding token are treated as nodes, interconnected by edges in the graph. Nodes are embedded using ConceptNet Numberbatch Embeddings to form a graph representation. To analyze the relationships between nodes in the graph, we used a Graph Convolutional Network (GCN), which learns the embedding relations among all nodes from two sessions.\u003c/p\u003e\u003cp\u003eThe ENA1.7.0 Web Tool was used to quantitatively process the codes in Excel format (Marquart et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). A codebook in Excel consisting of a list of the codes for all the participants was generated, with the occurrence of each strategy or difficulty coded as 1, and absence as 0.\u003c/p\u003e"},{"header":"Results and Discussion","content":"\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Tabe\" border=\"1\"\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCodes\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePrompt Engineering for Desired Output\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe difficulty in crafting effective, clear prompts for ChatGPT to generate the specific type of Python code or data analysis they need. Teachers may not know what to ask for, resulting in generic or incorrect outputs that don't meet their pedagogical or reporting needs.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eTechnical Literacy and Syntax Errors\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe difficulty in understanding and correctly writing the Python code provided by ChatGPT. Teachers may struggle with basic syntax, indentation, and the structure of a programming language, leading to frequent errors that they don't know how to debug.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eData Cleaning and Preparation\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe struggle with preparing their raw school data (i.e., from spreadsheets) for analysis. Teachers may face issues with missing values, inconsistent formats, or the need to merge different data sources, and they may not know how to instruct ChatGPT to handle these tasks effectively.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eConceptualizing Visualizations\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe difficulty in deciding which type of chart or graph (e.g., bar chart, scatter plot, pie chart) is most appropriate to tell the story of their data. They may understand the data, but lack the expertise in data visualization principles to choose an effective display.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVerification of AI-Generated Code\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eA lack of confidence or a sense of apprehension about using code generated by ChatGPT without fully understanding it. Teachers may worry about potential errors, data security risks, or the accuracy of the visualization without the ability to verify the underlying code themselves.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eIntegrating Data Insights\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eThe difficulty in moving from a raw data visualization to a meaningful classroom application. Teachers may struggle to translate a chart or graph into a teachable moment, a student activity, or a strategic decision for their school reports.\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eNotes.\u003c/p\u003e\u003cp\u003eNetwork Layout. The position of the nodes in the two-dimensional space is not random; it is determined by a statistical process that plots the concepts based on their co-occurrence. The closer two nodes are in the network, the more frequently they were discussed together, suggesting a stronger conceptual link. The overall position of the entire network (the group centroid) within this space indicates the dominant conceptual focus of the group at that time.\u003c/p\u003e\u003cp\u003eNodes. Each node in the network represents a single, distinct code or theme identified from the qualitative data. the nodes correspond to the specific challenges teachers faced in their engagement with AI-powered data visualization.\u003c/p\u003e\u003cp\u003eEdges (The Lines). The lines connecting the nodes represent a conceptual connection or co-occurrence between the coded themes.\u003c/p\u003e\u003cp\u003eLine Thickness. The thickness of a line indicates the strength of the connection between the two nodes it links.\u003c/p\u003e\u003cp\u003eBlue Network. Represents the conceptual framework of the teachers during Session 1, at the beginning of their learning journey.\u003c/p\u003e\u003cp\u003eGreen Network. Represents the evolved conceptual framework of the teachers during Session 2, after three weeks of engagement.\u003c/p\u003e\u003cp\u003eThe ENA reveals a fundamental shift in the teachers' conceptual framework over the three-week period. In Session 1, the blue network's most salient connections were centered on foundational, technical hurdles, with thick lines linking \"Prompt engineering,\" \"Data cleaning \u0026amp; preparation,\" and \"Syntax errors.\" This indicates that teachers' initial struggles were dominated by a linear workflow of technical problems and solutions. By Session 2, the green network shows a dramatic re-structuring; the original strong connections have weakened, and a new, highly cohesive cluster has formed. The thickest lines now link \"Conceptualizing visualizations,\" \"Verification of AI-Generated Code,\" and \"Integrating Data insights,\" signifying a shift in focus from low-level tool mechanics to higher-order pedagogical concerns. This shift indicates that the teachers have moved beyond the low-level, technical problems and are now wrestling with the higher-order, pedagogical challenges. They are no longer focused on \"how to make the code run\" but on \"what chart should I make,\" \"is this code and its output correct,\" and crucially, \"how can I use this to teach.\"\u003c/p\u003e\u003cp\u003eOf note, the strategies employed by the teachers in this study were categorized into three distinct tiers, representing a progression of skill acquisition and expertise adapted from the Dreyfus Model of Skill Acquisition (Dreyfus, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Zhang \u0026amp; Wang, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe result of the ENA provides an empirical account of how teachers' conceptual frameworks for AI-powered data visualization evolved over a three-week period interval from the data gathering.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe blue network, representing the initial state in Session 1, is characterized by a central focus on foundational, technical challenges. The most prominent connections exist between 'Prompt engineering,' 'Data cleaning \u0026amp; preparation,' and 'Syntax errors,' indicating that teachers' initial discourse and problem-solving were dominated by the basic, mechanical hurdles of getting the tools to function. The group centroid for this session is located on the left side of the conceptual space, reflecting this technical-centric frame.\u003c/p\u003e\u003cp\u003eA one-sample t-test was performed to assess the change between conditions. The results revealed a statistically significant difference along the X-axis (t(55.83)\u0026thinsp;=\u0026thinsp;2.93, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The magnitude of this shift is substantial, as indicated by a large effect size (Cohen\u0026prime;s d\u0026thinsp;=\u0026thinsp;0.93). This quantitative evidence confirms that the teachers' conceptual frame underwent a profound transformation from Session 1 to Session 2.\u003c/p\u003e\u003cp\u003eThe green network, representing the evolved state in Session 2, illustrates this shift. The group centroid has moved significantly to the right, signifying a new conceptual orientation. In this evolved frame, the strongest connections have coalesced around higher-order, pedagogical concerns. The most salient linkages now exist between 'Conceptualizing visualizations,' 'Verification of AI-Generated Code,' and 'Integrating Data insights.' This indicates that as teachers overcame the initial technical challenges, their focus elevated from tool mechanics to the critical and meaningful application of the data. The lack of a significant difference along the Y-axis suggests that this change was not a random reordering, but a focused progression along a primary dimension of their learning.\u003c/p\u003e\u003cp\u003eWhen comparing two group centroids, if their 95% confidence intervals do not overlap, it provides strong visual evidence that the two groups are statistically distinct in their conceptual space (Bowman et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn conclusion, the ENA models and supporting statistical analysis demonstrate a clear and substantial shift in the teachers' epistemic framework. They transitioned from a conceptual space dominated by technical difficulties to one focused on the strategic, critical, and pedagogical application of data. This finding provides a visual and empirical narrative of teacher empowerment, highlighting that true proficiency with AI-powered tools is achieved not by eliminating technical challenges, but by elevating one's engagement to the level of meaningful, contextualized application. This has significant implications for the design of future upskilling initiatives and educational tools, which must provide support for this critical conceptual leap.\u003c/p\u003e"},{"header":"Reflexivity","content":"\u003cp\u003eWhile the synergy between Python and ChatGPT offers transformative edge for data visualization in education, it is not without inherent risks, if wielded without proper understanding, can lead to unintended consequences.\u003c/p\u003e\u003cp\u003eOne observation was the tendency for some teachers to engage in \"Vibe Coding.\" They would copy code generated by the AI and paste it directly into their environment without first checking it for logical errors or data integrity issues. Also, the visualizations often had subpar aesthetics due to poor whitespace management, resulting in cluttered charts that were difficult to interpret. The study highlights a fundamental ethical blind spot in the technology itself. The tool does not prevent the creation of misleading or biased visualizations, placing the full responsibility for ethical practice on the user.\u003c/p\u003e\u003cp\u003eAnother critical concern is the \"black box\" phenomenon associated with AI code generation. Relying on a tool like ChatGPT to generate code can create a scenario where the user does not fully comprehend the underlying logic or the intricate processes by which the AI arrives at its outputs. For example, if the AI is trained on biased data, its outputs, and consequently the visualizations, may perpetuate those biases without the user's awareness.\u003c/p\u003e\u003cp\u003eGiven these risks, the research emphasizes a strong imperative for comprehensive training. This training should extend beyond the mere technical skills of using Python and ChatGPT for visualization. It must focus on the principles of data literacy, ethical data visualization, and the ability to critically interpret visual data, encouraging a healthy skepticism toward statistics, promoting analytical thinking and the ability to identify potential biases or misleading claims (Lester, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Furthermore, it must instill an understanding of ethical data practices, emphasizing transparency, fairness, and the avoidance of manipulating data or perpetuating stereotypes.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe integration of Python and ChatGPT for data visualization marks a pivotal in educational and technological possibilities, this paper opens inquiries related to the challenges of limited visualization literacy, lack of access to user-friendly tools, data fragmentation, and time constraints that have hindered educators from fully leveraging data for instructional improvement. However, the power of this technology necessitates a comprehensive data literacy training. Without a deep understanding of data principles, ethical visualization practices, and critical interpretation skills, there is a substantial risk of creating misleading visuals or blindly accepting AI-generated outputs. The \"black box\" nature of AI code generation unveils underlying logic to ensure accuracy, fairness, and accountability. Therefore, the full potential of this synergy can only be realized through robust professional development programs that equip teachers not just with technical proficiency, but with the critical thinking and ethical frameworks necessary to relate the data in education responsibly.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAdarsh Bhavimane, Rakshitha Shetty, Ghrutha Varsha Kurunji, Alex Tayenjam, \u0026amp; Dr. Pushparani M K. (2024). Data Visualization in Education: A Comprehensive Review. \u003cem\u003eInternational Journal of Advanced Research in Science Communication and Technology\u003c/em\u003e, 503\u0026ndash;509. https://doi.org/10.48175/IJARSCT-18676\u003c/li\u003e\n \u003cli\u003eBinali, T., Chang, C.-H., Chang, Y.-J., \u0026amp; Chang, H.-Y. (2022). High school and college students\u0026rsquo; graph-interpretation competence in scientific and daily contexts of data visualization. \u003cem\u003eScience \u0026amp; Education\u003c/em\u003e. https://doi.org/10.1007/s11191-022-00406-3\u003c/li\u003e\n \u003cli\u003eB\u0026ouml;rner, K., Maltese, A., Balliet, R. N., \u0026amp; Heimlich, J. (2016). Investigating aspects of data visualization literacy using 20 information visualizations and 273 science museum visitors. \u003cem\u003eInformation Visualization,\u003c/em\u003e \u003cem\u003e15\u003c/em\u003e(3), 198\u0026ndash;213. https://doi.org/10.1177/1473871615594652\u003c/li\u003e\n \u003cli\u003eBowman, D., Swiecki, Z., Cai, Z., Wang, Y., Eagan, B., Linderoth, J., Shaffer, D. W., Ruis, A. R., \u0026amp; Lee, S. B. (2021). The Mathematical Foundations of Epistemic Network Analysis. In \u003cem\u003eAdvances in Quantitative Ethnography\u003c/em\u003e (Vol. 1312, pp. 91\u0026ndash;105). Springer International Publishing AG. https://doi.org/10.1007/978-3-030-67788-6_7\u003c/li\u003e\n \u003cli\u003eBox, A. (2023). \u003cem\u003eA Power BI Compendium: Answers to 65 Commonly Asked Questions on Power BI\u003c/em\u003e (1st ed.). Apress. https://doi.org/10.1007/978-1-4842-9765-0\u003c/li\u003e\n \u003cli\u003eCampesato, O. (2023). \u003cem\u003ePython 3 Data Visualization Using ChatGPT/GPT-4\u003c/em\u003e.\u003c/li\u003e\n \u003cli\u003eCampesato, O. (2024). ChatGPT and Data Visualization. In \u003cem\u003ePython 3 and Machine Learning Using ChatGPT/GPT-4\u003c/em\u003e (pp. 1\u0026ndash;1). Mercury Learning and Information. https://doi.org/10.1515/9781501520112-009\u003c/li\u003e\n \u003cli\u003eCastillo, W., \u0026amp; Strunk, K. K. (2025). \u003cem\u003eHow to QuantCrit: Applying Critical Race Theory to Quantitative Data in Education\u003c/em\u003e. Taylor \u0026amp; Francis. https://doi.org/10.4324/9781003429968\u003c/li\u003e\n \u003cli\u003eChen, Z., Zhang, C., Wang, Q., Troidl, J., Warchol, S., Beyer, J., Gehlenborg, N., \u0026amp; Pfister, H. (2023). Beyond generating code: Evaluating GPT on a data visualization course. In \u003cem\u003e2023 IEEE VIS Workshop on Visualization Education, Literacy, and Activities (EduVis)\u003c/em\u003e (pp. 16\u0026ndash;21).\u003c/li\u003e\n \u003cli\u003eCsanadi, A., Eagan, B., Kollar, I. (2018). When coding-and-counting is not enough: using epistemic network analysis (ENA) to analyze verbal data in CSCL research. \u003cem\u003eIntern. J. Comput.-Support. Collab. Learn\u003c/em\u003e 13, 419\u0026ndash;438. https://doi.org/10.1007/s11412-018-9292-z\u003c/li\u003e\n \u003cli\u003eDreyfus, S. E. (2004). The Five-Stage Model of Adult Skill Acquisition. \u003cem\u003eBulletin of Science, Technology \u0026amp; Society\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(3), 177-181. https://doi.org/10.1177/0270467604264992 (Original work published 2004)\u003c/li\u003e\n \u003cli\u003eGlasgow, N. A., \u0026amp; Farrell, T. S. C. (2007). \u003cem\u003eWhat successful literacy teachers do : 70 research-based strategies for teachers, reading coaches, and instructional planners\u003c/em\u003e. Corwin Press.\u003c/li\u003e\n \u003cli\u003eGoldfarb Cohen, S., Yanai, J. V., \u0026amp; Dishon, G. (2024). Modeling group discourse with epistemic network analysis: Unpacking connections, perspectives, and individual contributions. \u003cem\u003eJournal of Science Education and Technology\u003c/em\u003e. https://doi.org/10.1007/s10956-024-10139-3\u003c/li\u003e\n \u003cli\u003eGrand Bang, C. (2024). Data Visualization and Presentation: How to Present and Communicate Data Effectively for Decision-Making. In \u003cem\u003eData-Driven Decision-Making for Business\u003c/em\u003e. Taylor \u0026amp; Francis Group. https://doi.org/10.4324/9781003457787-6\u003c/li\u003e\n \u003cli\u003eHwang, M., Lee, K.-H., \u0026amp; Lee, H.-K. (2025). A word to the wise: Crafting impactful prompts for ChatGPT. \u003cem\u003eSystem\u003c/em\u003e, \u003cem\u003e133\u003c/em\u003e, 103756. https://doi.org/10.1016/j.system.2025.103756\u003c/li\u003e\n \u003cli\u003eJackson, A. (2025). \u003cem\u003eLearning AI tools in Tableau: Level up your data analytics and visualization capabilities with Tableau Pulse and Tableau Agent\u003c/em\u003e. O\u0026rsquo;Reilly Media, Inc.\u003c/li\u003e\n \u003cli\u003eJiang, T., Liu, E., Baig, T., \u0026amp; Li, Q. (2024). Enhancing decision‐making in higher education: Exploring the integration of ChatGPT and data visualization tools in data analysis. \u003cem\u003eNew Directions for Higher Education\u003c/em\u003e, \u003cem\u003e2024\u003c/em\u003e(207), 15\u0026ndash;29. https://doi.org/10.1002/he.20510\u003c/li\u003e\n \u003cli\u003eKasneci, E., Sessler, K., K\u0026uuml;chemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., G\u0026uuml;nnemann, S., H\u0026uuml;llermeier, 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., \u0026amp; Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. \u003cem\u003eLearning and Individual Differences, 103\u003c/em\u003e, 102274. https://doi.org/10.1016/j.lindif.2023.102274\u003c/li\u003e\n \u003cli\u003eKim, N. W., Ko, H.-K., Myers, G., \u0026amp; Bach, B. (2024). ChatGPT in Data Visualization Education: A Student Perspective. \u003cem\u003eProceedings of IEEE Symposium on Visual Languages and Human-Centric Computing, VL/HCC\u003c/em\u003e, 109\u0026ndash;120. https://doi.org/10.1109/VL/HCC60511.2024.00022\u003c/li\u003e\n \u003cli\u003eKrejci, S. E., Ramroop-Butts, S., Torres, H. N., \u0026amp; Isokpehi, R. D. (2020). Visual Literacy Intervention for Improving Undergraduate Student Critical Thinking of Global Sustainability Issues. \u003cem\u003eSustainability\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(23), 10209. https://doi.org/10.3390/su122310209\u003c/li\u003e\n \u003cli\u003eLawn, M. (2013). \u003cem\u003eThe rise of data in education systems: Collection, visualization and uses\u003c/em\u003e. Symposium Books.\u003c/li\u003e\n \u003cli\u003eLee, H. S., Mojica, G. F., Thrasher, E. P., \u0026amp; Baumgartner, P. (2022). Investigating data like a data scientist: Key practices and processes. Statistics Education Research Journal, 21(2), Article 3. https://doi.org/10.52041/serj.v21i2.41\u003c/li\u003e\n \u003cli\u003eLee, S., Kim, S. H., \u0026amp; Kwon, B. C. (2017). VLAT: Development of a visualization literacy assessment test. \u003cem\u003eIEEE Transactions on Visualization and Computer Graphics,\u003c/em\u003e \u003cem\u003e23\u003c/em\u003e(1), 551\u0026ndash;560. https://doi.org/10.1109/tvcg.2016.2598920\u003c/li\u003e\n \u003cli\u003eLester, P. M. (2022). Data Visualization Ethics. In \u003cem\u003eVisual Ethics\u003c/em\u003e (2nd ed., pp. 115\u0026ndash;125). Routledge. https://doi.org/10.4324/9781003243045-11\u003c/li\u003e\n \u003cli\u003eLiu, H., \u0026amp; Singh, P. (2004). ConceptNet \u0026mdash; A Practical Commonsense Reasoning Tool-Kit. \u003cem\u003eBT Technology Journal\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(4), 211\u0026ndash;226. https://doi.org/10.1023/B:BTTJ.0000047600.45421.6d\u003c/li\u003e\n \u003cli\u003eMaddigan, P., \u0026amp; Susnjak, T. (2023). Chat2VIS: Generating Data Visualisations via Natural Language using ChatGPT, Codex and GPT-3 Large Language Models. \u003cem\u003eIEEE Access\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 1\u0026ndash;1. https://doi.org/10.1109/ACCESS.2023.3274199\u003c/li\u003e\n \u003cli\u003eMakowski, M. B., \u0026amp; Lubienski, S. T. (2023). Classroom Data Visualization: Tracking Individuals During Group-Centered Instruction. \u003cem\u003eEducational Researcher\u003c/em\u003e, \u003cem\u003e52\u003c/em\u003e(3), 164\u0026ndash;169. https://doi.org/10.3102/0013189X231158374\u003c/li\u003e\n \u003cli\u003eMarquart, C. L., Hinojosa, C., Swiecki, Z., Eagan, B., \u0026amp; Shaffer, D. W. (2018). \u003cem\u003eEpistemic Network Analysis (Version 1.5.2) [Software]\u003c/em\u003e\u003c/li\u003e\n \u003cli\u003ePerkhofer, L., Walchshofer, C., \u0026amp; Hofer, P. (2020). Does design matter when visualizing Big Data? An empirical study to investigate the effect of visualization type and interaction use. \u003cem\u003eJournal of Management Control\u003c/em\u003e, \u003cem\u003e31\u003c/em\u003e(1\u0026ndash;2), 55\u0026ndash;95. https://doi.org/10.1007/s00187-020-00294-0\u003c/li\u003e\n \u003cli\u003eSkvortsova, S., Symonenko, T., \u0026amp; Hnezdilova, K. (2025). Optimizing Routine Educational Tasks through Prompt Engineering: A Comparative Study of AI Chatbots. \u003cem\u003eProceedings of International Conference on Applied Innovation in IT, 13\u003c/em\u003e(1), 35-42.\u003c/li\u003e\n \u003cli\u003eSorapure, M. (2019). Text, Image, Data, Interaction: Understanding Information Visualization. \u003cem\u003eComputers and Composition\u003c/em\u003e, \u003cem\u003e54\u003c/em\u003e, 102519. https://doi.org/10.1016/j.compcom.2019.102519\u003c/li\u003e\n \u003cli\u003eZhang, S., \u0026amp; Wang, N. (2021). Dreyfus\u0026rsquo; Model of Skill Acquisition from the Perspective of Phenomenology. \u003cem\u003eJournal of Engineering Studies\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(4), 353\u0026ndash;361. https://doi.org/10.3724/SP.J.1224.2021.00353\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence in Education, Python, ChatGPT, Epistemic Network Analysis","lastPublishedDoi":"10.21203/rs.3.rs-8023714/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8023714/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study explores the synergy of Python and ChatGPT's intuitive accessibility to empower non-programming educators in making data-driven decisions. Using a Quantitative Ethnography (QE) approach, we analyzed data discourse of behavior of twenty teachers from three private schools through AI-guided prompts. Qualitative coding first identified teachers' difficulties and strategies. Epistemic Network Analysis (ENA) was then employed to generate network models, revealing how these codes co-occur. We observed a shift in teachers' frameworks over a three-week period. In Session 1, the conceptual focus was on technical hurdles, with strong connections linking \"Prompt engineering,\" \"Data cleaning \u0026amp; preparation,\" and \"Syntax errors\". This indicates a discourse dominated by a linear, technical problem-solving approach. The group centroid for this session was located on the left side of the conceptual space. By Session 2, a shift occurred with a large effect size (Cohen's d=0.93). The restructuring shows that as teachers overcame initial technical challenges, their acumen elevated from low-level tool mechanics to higher-order pedagogical application. The findings provide an account of a transition from a conceptual space dominated by technical difficulties to one focused on the strategic, critical, and pedagogical application of data vis-à-vis their professional practice. While the study is limited by the exclusion of data storytelling context, it highlights the need for additional design features to help teachers engage meaningfully with data.\u003c/p\u003e","manuscriptTitle":"Python and ChatGPT: AI-Powered Data Visualization for Teachers","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-05 10:03:02","doi":"10.21203/rs.3.rs-8023714/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bb1bb34d-49ae-4fa5-a812-e526b8f2af51","owner":[],"postedDate":"November 5th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":57379143,"name":"Theoretical Computer Science"},{"id":57379144,"name":"Technical Communication"}],"tags":[],"updatedAt":"2025-11-05T10:03:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-05 10:03:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8023714","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8023714","identity":"rs-8023714","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}