Uses WildChat
Prototypical Human-AI Collaboration Behaviors from LLM-Assisted Writing in the Wild
EMNLP 2025 · First arXiv submission
Studies how users steer writing across successive turns. The authors label follow-up types, identify recurring collaboration patterns with principal component analysis, and relate those patterns to writing intents such as brainstorming, technical drafting, and long-form composition.
Data used
Approximately 68,000 English writing sessions from WildChat-1M, filtered to at least two user utterances; compared with writing sessions from Bing Copilot.
Findings
Seven collaboration patterns account for 80–85% of the variance in the analyzed follow-up representations; explicit positive or negative feedback appears in only a small minority of sessions.
Sections 3–6 and Appendix B: WildChat subset, analysis, and shared patterns across the two services. Read the source.
Dates and versions
First arXiv version. ArXiv version history.
EMNLP 2025 proceedings paper.
Cite this paper
Sheshera Mysore, Debarati Das, Hancheng Cao, Bahareh Sarrafzadeh. 2025. Prototypical Human-AI Collaboration Behaviors from LLM-Assisted Writing in the Wild. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing.
Show BibTeX
@inproceedings{pathswriting2025,
title = {{Prototypical Human-AI Collaboration Behaviors from LLM-Assisted Writing in the Wild}},
author = {Mysore, Sheshera and Das, Debarati and Cao, Hancheng and Sarrafzadeh, Bahareh},
year = {2025},
booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
publisher = {Association for Computational Linguistics},
pages = {16819--16846},
doi = {10.18653/v1/2025.emnlp-main.852},
url = {https://aclanthology.org/2025.emnlp-main.852/}
}