# What Happens When We Prompt the Model, Not the People? Evaluating LLMs as Survey Estimators

Status: under-review
Visibility: metadata-only
Status label: Submitted to Public Opinion Quarterly
Authors: with Trent D. Buskirk

## Summary

This paper evaluates LLM-based estimates as direct survey estimators rather than as synthetic respondents or synthetic microdata. It benchmarks model-generated estimates against Survey of Consumer Finances quantities and emphasizes substitution risk when rough point-estimate similarity is mistaken for survey-compatible accuracy.

## Public Files

- Metadata: https://jyl19.github.io/papers/what-happens-when-we-prompt-the-model-not-the-people-evaluating-llms-as-survey-estimators/metadata.json
- PDF: not yet staged for public release