By Minkyu Shin
Have a complaint about a business? AI may be a lot more effective at explaining your grievances than you are.
My colleagues and I analyzed more than one million consumer complaints filed with the federal Consumer Financial Protection Bureau, separating out complaints that were likely written or edited by AI. And we found that they had a clear advantage over the ones likely prepared by people alone: AI-assisted complaints were 6.9 percentage points more likely to receive a relief decision from financial firms.
Giving an explanation
Many consumers turn to the Consumer Financial Protection Bureau after trying to resolve an issue directly with a financial firm. The bureau collects complaints and forwards almost all of them to the companies, requiring a response.
But the bureau asks for a narrative about the complaint: what happened, when it happened, what the company did, and what the consumer wants done about it. That can be a huge barrier for people who aren't confident writers, or have to write in a second language -- exactly the kind of situations that would lead people to use large language models.
So, just how many people were turning to AI? And what kind of results were they getting? To find out, my colleagues -- Jin Kim and Jiwoong Shin -- and I took one million complaints filed to the bureau, from March 2015 through March 2024, and plugged them into a commercial AI-detection tool. The software sorted the complaints into two groups: likely AI and likely human.
As of March 2024, it found, nearly one in 10 complaints filed were likely written or edited by AI -- up from essentially none before ChatGPT was released to the public on Nov. 30, 2022. And -- by using ZIP Code information -- we found that the increases in likely AI complaints were larger in areas where a greater share of residents speak a language other than English at home, indicating potential language barriers to writing a narrative.
Our research found that the AI-aided complaints got better responses: They had a 6.9-percentage-point better chance of getting either monetary or nonmonetary relief, such as correcting a credit report, than those written by a human.
We found the same pattern in three online experiments in which participants played the role of complaint handlers and were randomly shown either an original complaint or an AI-edited version of the same complaint. Across experiments, AI-assisted complaints were more likely to receive favorable responses.
Presentation matters
Why does AI get better results? When decision makers face uncertainty and ambiguity about core issues, the presentation of a complaint can carry significant signaling value. Qualities such as fluency, clarity and professionalism act as a channel of credibility.
Consider this likely human original complaint submitted to the bureau. In the original, the consumer writes, "I got aggravated with it and decided to just pay the total balance I owed..." After we ran it through ChatGPT, it became: "Frustrated by these issues, I decided to pay off the total balance I owed..."
Later, "Now their calling me saying I owe $110 i'm assuming it was for late fees!" became: "Now, I am being contacted by the company, saying that I owe $110.00, which I assume is for additional late fees."
The facts didn't change, just the presentation: The revised version is clearer about the sequence of events, more precise in its wording, and more conventionally professional in tone. Those small changes can make a big difference when complaints are reviewed at scale.
Our research offers an important implication for companies. As AI-assisted writing spreads, prose quality becomes a weaker signal of case quality -- and firms that equate "well written" with "more legitimate" may be unwittingly favoring AI-assisted complaints. Complaints should be judged based on hard evidence like documents, transaction records and clear procedures -- not on how polished the writing is.
Minkyu Shin is an assistant professor of marketing at the City University of Hong Kong. He can be reached at reports@wsj.com.
(END) Dow Jones Newswires
March 19, 2026 15:00 ET (19:00 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments