Who Gets the Credit? Audit Workbook and Intercoder Reliability Sample — A Structured Audit of Google's AI-Generated Answers in India

This dataset accompanies the study "Who Gets the Credit? Source Attribution in Google's AI-Generated Answers Versus Organic Rankings: A Structured Audit from India." It contains two files: 1. Audit workbook — The complete audit instrument and link-level dataset. It covers 192 Google searches captured from Ahmedabad, India, balanced across four sectors and three search intents, recording 931 AI-generated citations and 960 organic results. The workbook has five sheets: Queries — The 192-query sampling frame: query ID, sector, search intent, and exact query text as entered. Coding — One row per search (n = 192): capture date and time, answer surface (AI Overview / AI Mode / None), the researcher's fold-visibility judgement, and formula-derived counts and completion status. Links — The citation-level dataset (n = 3,072 rows; 16 per search): every credited website, organic result, and sponsored result, with domain and source type extracted by formula and an in-top-five match flag. Websites — The shared domain-type register (779 distinct domains), classifying each domain into one of eight source-type categories, applied consistently across the dataset. Results — Live formula-driven summary statistics: prevalence, overlap, source-type composition, fold visibility, and the underlying statistical tests and data-quality checks. 2. Intercoder reliability sample — Supports the source-type reliability check reported in Section 6.7 of the paper. Contains a stratified random sample of 168 domains (drawn from the 762 domains appearing among credited or organic results in the audit workbook), each independently classified into one of eight source-type categories by two coders. The coders agreed on 144 of 168 domains (85.7%; Cohen's kappa = 0.79, 95% CI 0.72–0.87). No personal data were collected; all observations are of publicly accessible search results pages.