JPMorgan owns AI answers. Here's how rivals can compete.. | American Banker

Author Follow Us whatsapp<br>twitter<br>youtube<br>facebook<br>linkedin

© 2026 Arizent. All rights reserved.

Sections

Show Search

Search Query Submit Search

AI Home

Research

Innovation

AI Risk

Agentic AI and Personalization

Insights

Top Stories

Artificial Intelligence

JPMorgan owns AI answers. Here's how rivals can compete..<br>By Penny Crosman CloseText

About Penny<br>twitter<br>pennycrosman<br>mailto<br>penny.crosman@arizent.com<br>linkedin<br>pennycrosman

Published July 22, 2026, 12:48 p.m. EDT<br>Updated July 22, 2026, 12:50 p.m. EDT

8 Min Read

Facebook<br>Twitter<br>LinkedIn<br>Email

Key insight: JPMorganChase has an outsized influence on popular large language models like ChatGPT and Claude.<br>What's at stake: Banks that do not appear in LLMs' answers lose the ability to reach new customers, especially Gen Zers.<br>Expert quote: "AI is more like collective knowledge or collective intelligence, not really like artificial intelligence, and so it just aggregates the mass of what's out there and distills it," said Siya Vansia, chief brand and innovation officer at ConnectOne Bank.<br>JPMorganChase has an outsized presence in the output of large language models, according to a recently released analysis that offers lessons for smaller banks on how to gain more visibility in the age of AI.<br>The market research and marketing firm 5W studied how often the largest U.S. banks appeared in answers to 31,500 consumer-banking queries pasted into ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews from January through May 2026.<br>Processing Content<br>JPMorganChase was cited in 28% of the models' answers, though its market share of U.S. consumer deposits is only around 17%. It was cited more times than Bank of America, Wells Fargo, Citi and Capital One Financial combined. Twenty-two of the 75 largest banks had less than 0.3% citation share.

Showing up in AI answers is increasingly important for banks as consumers shift from Google searches to AI models (including Google AI Overviews) when they need information. An RFIGlobal study released this week found that 49% of U.S. adults use AI for general information and research, and 80% of Gen Zers and millennials use it for at least one banking task. About 51% of U.S. consumers use AI to research products at least once a week, according to an Emarketer study to be published next week.<br>"Marketers and brands and companies have to understand, you're no longer talking to people; you're talking to machines," 5W Chairman Ronn Torossian told American Banker.

Why JPMorganChase dominates<br>Experts say JPMorganChase's brand recognition is a big part of its dominance in LLMs' answers.<br>The large language models in the study "are trained on pretty much the corpus of the internet," said John A. Thompson, professor at the University of Michigan, where he teaches courses on AI.

"I think Chase is mentioned much more widely and deeply and broadly than the other banks, so the training data says that there's something about Chase that's more ubiquitous and happens at a higher frequency than the other banks," said Thompson, who is not connected to 5W. "I don't think that there's any engineering or malfeasance going on in there, other than the training data is probably biased towards Chase, and therefore it shows up in the results."<br>The banks that lag behind generate less content and marketing material, Thompson surmised.<br>"Obviously their marketing engines are anemic compared to what Chase is doing," he said. "This is a war of volume. If they want their numbers to rise and maybe even rival Chase at some point, they have to put out more content rather than sitting back as a bank and saying, 'Hey, come to our site and look at our rates.'"<br>JPMorganChase is also a prolific publisher of long research reports, which gives it an edge, especially with models like Gemini and Perplexity that appreciate extensive detail and lots of sources. Capital One shows up well in these two models because it posts long-form, detailed, source-heavy content, Torossian theorizes.<br>Studies of large language models' output are challenging to conduct because the models are sycophantic — they aim to tell users what they likely want to hear, based on their past activity.<br>For the 5W study, every query was run in a fresh, logged-out instance — "no account, no memory, no chat history, no personalization signal, each isolated," Torossian said.<br>The prompts were neutral, he said. "We don't ask, 'Is JPMorgan the best bank for X?' We ask the questions a real buyer asks — 'Best bank for a small business checking account,' 'Who has the best mortgage rates?' 'Safest bank for a large deposit.' There's no brand priming."

A dead brand lives on in LLMs<br>One surprise of the 5W study is that Goldman Sachs' Marcus brand appeared frequently in AI answers, cited in 31% of consumer banking responses, even though it was mostly shut down in 2022. This is a testament to the amount of marketing Goldman...