Gender-Reversed Simulated AI Career-Counseling Sessions (Korean): Transcripts, Posture Coding and Run Records
Description
This dataset accompanies the article “Can AI Become My Counselor? An Exploratory Audit of Gender-Differentiated Career Guidance.” It contains 400 simulated career-counseling sessions in Korean between a language-model counselor and language-model clients (gpt-5.6-terra). The clients were built from 100 synthetic personas based on Nemotron-Personas-Korea (NVIDIA, 2026; CC-BY-4.0). Each persona presented once as a woman and once as a man in each of two studies. In Study 1, the client stated their gender. In Study 2, gender could only be inferred from the name and gender-marking words. Study_Data_Revision5.xlsx contains: the persona records as given to the client model; the full transcripts; the 71,811 counselor units with their analysis codes under the final coding manual (Revision 5); the manual itself and a crosswalk from the pilot categories; the codes of both human coders on the development, held-out and supplementary samples; the AI coding run log; the counselor and client prompts with their SHA-256 hashes; the session records; the manipulation-check outputs for both studies; the session-level category rates and within-persona differences used in the paired analysis. Run_records.zip contains the original run records of every session, including the exact input sent to each agent at each call and the raw model responses. analysis_scripts.py reproduces the agreement statistics and the paired analyses reported in the article (Tables 2–4). README.txt lists each file with its SHA-256 hash. All persona records are synthetic, and none corresponds to a living person.