The adoption and efficacy of large language models in US consumer financial complaints

This research explores the impact of large language models (LLMs) on consumer complaints submitted to the US Consumer Financial Protection Bureau. Analysing 1,134,512 complaints from 2015 to 2024, we document a sharp increase in LLM usage following the release of ChatGPT. An instrumental variable analysis estimates that LLM usage increases the probability of obtaining favourable relief by 6.9 percentage points (95% confidence interval, (4.9, 8.9)). The analysis also reveals evidence of negative selection, where consumers otherwise prone to adverse outcomes are more likely to adopt LLMs. To further substantiate these findings and test the mechanism, we conducted three online controlled experiments (total N = 1,010 US participants); these demonstrate that LLMs can increase the likelihood of obtaining relief by enhancing the presentation of complaints without altering factual content. These findings suggest that LLMs can act as an equalizer, highlighting the need for policies that expand access to these technologies.