When Stereotypes GTG: The Impact of Predictive Text Suggestions on Gender Bias in Human-AI Co-Writing
By: Connor Baumler <baumler@cs.umd.edu> and Hal Daumé III
@inproceedings{10.1145/3772318.3790733,
author = {Baumler, Connor and Daum\'{e}, Hal, III},
title = {When Stereotypes GTG: The Impact of Predictive Text Suggestions on Gender Bias in Human-AI Co-Writing},
year = {2026},
isbn = {9798400722783},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3772318.3790733},
doi = {10.1145/3772318.3790733},
booktitle = {Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems},
articleno = {682},
numpages = {44},
keywords = {Co-writing, predictive text, stereotyping},
location = {
},
series = {CHI '26}
}
This repository contains writing scenarios, model prompts, and sample generation code. The human-facing prefix of each writing scenario and the corresponding model prompt (and potential pre-determined first set of suggestions) can be found in prompts.tsv. The prompts used to collect model annotations can be found in hypotheses.csv. example_generation.py includes the code for generating predictive text suggestions and can be run as-is to see example generations on one scenario.
