{"slug": "jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete", "title": "JPMorgan Owns AI Answers. Here's how rivals can compete..", "summary": "JPMorganChase was cited in 28% of AI model answers to 31,500 consumer-banking queries, despite holding only 17% of U.S. consumer deposits, according to a 5W analysis of ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews from January through May 2026. The bank appeared more often than Bank of America, Wells Fargo, Citi and Capital One Financial combined, while 22 of the 75 largest banks had less than 0.3% citation share. \"Marketers and brands and companies have to understand, you're no longer talking to people; you're talking to machines,\" said 5W Chairman Ronn Torossian.", "body_md": "**Key insight:** JPMorganChase has an outsized influence on popular large language models like ChatGPT and Claude.**What's at stake:** Banks that do not appear in LLMs' answers lose the ability to reach new customers, especially Gen Zers.**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.\n\nJPMorganChase 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.\n\nThe 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.\n\nJPMorganChase 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.\n\nShowing 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.\n\n\"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.\n\n## Why JPMorganChase dominates\n\nExperts say JPMorganChase's brand recognition is a big part of its dominance in LLMs' answers.\n\nThe 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.\n\n\"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.\"\n\nThe banks that lag behind generate less content and marketing material, Thompson surmised.\n\n**Read more:**\n\n[Four factors that drove banks' blowout 2Q performance](https://www.americanbanker.com/news/four-factors-that-drove-banks-blowout-2q-performance)[Banks face a dilemma: More loan growth or better margins?](https://www.americanbanker.com/news/banks-face-a-dilemma-more-loan-growth-or-better-margins)[The top-performing 20 public banks with under $2B of assets in 2025](https://www.americanbanker.com/news/the-top-performing-20-public-banks-with-under-2b-of-assets-in-2025)[How a serial borrower fooled 7 banks out of $39 million](https://www.americanbanker.com/news/how-a-serial-borrower-fooled-seven-banks-out-of-39-million)\n\n\"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.'\"\n\nJPMorganChase 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.\n\nStudies 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.\n\nFor 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.\n\nThe 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.\"\n\n## A dead brand lives on in LLMs\n\nOne 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 did for Marcus. In 2020, for example, the bank spent an estimated $28 million on advertising for the brand, according to __Business Insider__** **\n\n\"They were doing a lot more content marketing than anybody else in the space,\" Torossian said. \"There's probably seven zillion pages out there talking about Marcus, and when they closed it, they didn't put out seven zillion pages. Their AI strategy was just much better.\"\n\nLarge language model training data \"doesn't do a really good job at looking at subtlety,\" Thompson pointed out. \"There was a lot of information about Marcus's creation, its offering, how good Goldman Sachs thought it was,\" provided to the frontier AI models, he said.\n\n\"And that information will be out there in perpetuity, unless the prompt asks, 'When did the model stop seeing a significant amount of information about Marcus as an ongoing offering?' Or 'When did the model recognize that Marcus no longer existed?' Unless you were affirmatively asking about the denouement or the ending of Marcus, the model wouldn't respond\" with that information.\n\nThis bug provides users with a distorted view of reality.\n\nFor a class Thompson was teaching about critical thinking while using generative AI, he created a 17-page prompt that spells out trusted versus untrusted sources, to give a model context when it answers questions. \"If you want to cognitively offload some of your original foundational thinking to a model, you could do it, and it gives you a framework, so the answers you're getting are vetted through a process that you would actually use,\" Thompson said. About 3,000 people have used his prompt since he published it three years ago.\n\nLarge language models are not holistic and don't have the kind of institutional knowledge humans have, said Siya Vansia, chief brand and innovation officer at ConnectOne Bank.\n\n\"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,\" Vansia told American Banker. \"Let's say I want to learn about what determines the size of an ocean wave. You would think that it would go deep and search all the facts and give me a summary, but that's not really what's happening. It's all dependent on what's available on the internet, and then if no one's vlogged about it, you won't get a good answer.\"\n\n## How banks can become more AI-visible\n\nBanks that are AI-visible have mastered the art of impressing large language models, which is called generative engine optimization, or GEO. Companies can improve their GEO by communicating better with large language models, Torossian said.\n\nThey can rewrite their websites, for instance, to have schema, FAQs, headers and images that the large language models can read and understand, he said. Though websites only account for about 7% of LLM citations, according to the 5W study, they're one of the few things a company can completely control.\n\n\"We estimate that more than 90% of American businesses need to redo their websites to be more LLM-friendly and to tell a consistent story,\" Torossian said. Because they are heavily regulated, some banks don't give out much information on their websites, he noted.\n\nConnectOne Bank, which is not connected to 5W, is a case in point. The New Jersey bank recently upgraded its website infrastructure to become more LLM-friendly, using software from Webflow.\n\nBut Vansia treads carefully when changing ConnectOne's website content, to make sure it's in compliance with bank regulations and also educational for consumers and clients, with proper disclosures.\n\n\"I can't focus so much on AI that I confuse the person when they're on the site,\" she said. \"But we are doing a lot of research to learn how we can better word titles and such. The good thing about a website is you can experiment a little bit.\"\n\nTorossian advises CEOs to write blogs and posts on Facebook, Twitter and LinkedIn, and to take strong positions on issues in those forums.\n\nAt ConnectOne Bank, CEO Frank Sorrentino has had a blog for years. Vansia recently started turning these pieces into Forbes Advisor contributions.\n\n\"We get a lot of feedback from our clients on it,\" she said. \"They like thought leadership.\"\n\n\"I think there's a unique opportunity in the way community and regional banks can show up with a local presence and drive value that way, because there's a lot of noise,\" Vansia said. \"We have a CEO blog. We're investing in marketing infrastructure and tools, marketing automation. We rolled out a podcast. We continue to develop our thought leaders. If I can engage our client base and our prospects, and continue with community building, that becomes even more valuable in the age of AI.\n\n\"We cannot take our eye off this ball,\" she said. \"We test and learn, test and learn, especially as everything evolves.\"\n\nMentions of a bank on external websites also matter. About 40% of AI engines' information comes from Reddit and Wikipedia, according to Torossian.\n\nVansia is considering creating a Wikipedia page for ConnectOne Bank. She is also thinking about starting to post or advertise on Reddit or ChatGPT.\n\nAmong publishers, AI models cite Bankrate, Investopedia and NerdWallet most often. \"The brand authority of the U.S. banking industry has been outsourced to a handful of media platforms,\" the 5W report stated.\n\nAI engines also draw answers from the Consumer Financial Protection Bureau's complaint database and the Federal Deposit Insurance Corp.'s consumer data resources, the 5W study found.\n\nThe blogging site Medium also ranks high with AI models.\n\n\"The elites are over and finished,\" Torossian said. \"The barnyard is open, and everybody can come and play. Medium, Wikipedia, Reddit — these are not the Ivy League towers shaping everything.\"", "url": "https://wpnews.pro/news/jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete", "canonical_source": "https://www.americanbanker.com/news/jpmorgan-owns-ai-answers-heres-how-rivals-can-compete", "published_at": "2026-07-22 17:04:10+00:00", "updated_at": "2026-07-22 17:23:02.169407+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products"], "entities": ["JPMorganChase", "5W", "ChatGPT", "Claude", "Gemini", "Perplexity", "Google AI Overviews", "Ronn Torossian"], "alternates": {"html": "https://wpnews.pro/news/jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete", "markdown": "https://wpnews.pro/news/jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete.md", "text": "https://wpnews.pro/news/jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete.txt", "jsonld": "https://wpnews.pro/news/jpmorgan-owns-ai-answers-here-s-how-rivals-can-compete.jsonld"}}