Managerial AI Optimism in US Technology Earnings Calls: An Educational Dataset

What this dataset is. An educational dataset of constructed measures of managerial AI communication, derived from the quarterly earnings calls of 20 US-listed technology companies across 2024Q4–2025Q3. It provides passage-level and company-quarter scores for Managerial AI Optimism and six AI-claim dimensions, rule-based AI-relevance flags, an independently human-coded benchmark of 456 passages, and the prompts and coding rules used to produce them. What this dataset is not. It is not a transcript corpus and is not a source of earnings-call transcripts. The dataset contains short excerpts selected for their relevance to the constructs being measured, and that text has been adjusted and reformatted for educational use. It does not reproduce complete or verbatim calls. Anyone needing transcripts should obtain them from an appropriate source; the contribution here is the accumulated and constructed metrics, together with the human-coded benchmark against which they can be assessed. Intended uses. Teaching and demonstrating measurement workflow design, LLM-assisted text coding, inter-coder reliability, criterion-related and construct validity, invariance testing, and evaluation on held-out data. Suitable for research-methods teaching, workshops and tutorials, methodological demonstration, and as a benchmark for comparing coding instruments. Contents. passages.csv — 9,636 three-sentence scoring passages with AI-relevance flags and all constructed scores. company_quarter_variables.csv — 80 company-quarter rows; the aggregated analytical variables. gold_sample.csv — the 456-passage human-coded benchmark. prompts_construction.csv — prompts, model settings and coding rules used to construct the measures. prompts_validation.csv — prompt variants and perturbations used in validation, with the change applied to each. Limitations. Because the excerpts were selected for construct relevance rather than sampled to represent the calls, and because the text has been edited for teaching, the dataset does not support substantive conclusions about the companies named in it. It is a teaching and methodological resource. Compiled and made available by the author, who holds the rights to the constructed measures and to this compilation. All rights reserved.