Can ChatGPT Kill User-Generated Q&A Platforms?

Large language models (LLMs), such as ChatGPT, are expected to significantly reshape learning, skill acquisition, and knowledge creation. Their impact on user-generated knowledge-sharing question and answer (Q&A) platforms has important implications for the comparative value and future coevolution of Q&A platforms and LLMs as well as for the future learning of LLMs. Using multiple empirical designs, we estimate that the launch of ChatGPT led to an average 14.09% reduction in the number of questions posted on Stack Overflow, an effect that intensifies over time and reaches a 27.88% decline by May 2023. Intriguingly, this decrease in quantity is accompanied by a substantial increase in the average quality of the remaining questions. Further analysis shows that these quantity and quality changes can be explained by two mechanisms: (1) uneven substitution by ChatGPT, where simpler and mid- to low-quality questions are more likely to disappear, and (2) direct quality improvement after accounting for the quantity reduction, suggesting positive spillovers from the time and search cost savings provided by ChatGPT assistance to improvements in question quality. Additional heterogeneity analyses reveal stronger effects on both question quantity and quality among inexperienced users; significant declines in both user retention and new user acquisition; and systematic variation across topics characterized by tenure, broadness, and depth. These findings delineate the coexistence boundary between LLMs and Q&A platforms, suggesting an evolution toward smaller but more specialized, high-expertise Q&A communities alongside growing welfare gains and value transfer to LLMs. They also offer actionable insights for platform managers and LLM developers on sustaining the value of user-generated content ecosystems while supporting long-run model improvement.

History: Juan Feng, Senior Editor; Jianqing Chen, Associate Editor.

Funding: J. Xue was supported by the Fundamental Research Funds for the Central Universities of Shanghai University of Finance and Economics. L. Wang was supported by the National Natural Science Foundation of China [Grant 72301259]. Y. Li was supported by the National Natural Science Foundation of China [Grants 72188101 and 72595864].

Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.0561.