Dataser for Accessible Web Content Generation Using LLMs: An Empirical Study on Prompting Strategies and Template-Guided Remediation
Description
This supplementary dataset provides full transparency and reproducibility for the empirical study titled "Accessible Web Content Generation Using LLMs: An Empirical Study on Prompting Strategies and Template-Guided Remediation". It contains all experimental data used to analyze the impact of different prompting strategies and input configurations in the automated remediation of web accessibility issues. The spreadsheet is organized into multiple sheets that document both manual and automated evaluations: Manual evaluation sheets report the results of expert assessments using the Barrier Walkthrough (BW) method, including severity levels (Minor, Significant, Critical) per variant and barrier. Summary graphs (bar and pie charts) aggregate this data for visualization and interpretation. Automated evaluation sheets include metrics obtained from accessibility tools such as WAVE and Lighthouse, reporting the number of errors, alerts, and compliance indicators across variants. Correlation sheets compare automated and manual results to assess alignment between tool-based evaluations and expert heuristics. Reference materials document the barrier definitions, scoring scales, and variant descriptions used in the study, ensuring methodological transparency. This dataset enables reproducibility, supports comparative analyses of prompt effectiveness, and provides a structured foundation for future research in AI-driven accessibility remediation.