4
Banks AI never mentions
Across 1,080 conversations where the owner asked for options without naming a bank, Popular Bank, East West Bank, Valley Bank and Associated Bank were never named — at any point, in any conversation.
We ran 2,160 four-turn buying conversations across ChatGPT, Gemini and Perplexity, in which a small-business owner describes their situation and asks an AI assistant what to do. One fintech was recommended more often than all thirty banks in the study combined.
Mercury was the single, outright recommendation in 594 of 2,160 conversations. Every bank combined — all thirty in the study, plus any other bank an assistant chose on its own — managed 452.
Out of 2,160 conversations.
502 of 702.
669 fintech wins against 107 for another bank.
307 of 720 situations.
A small-business owner says they have been looking at a particular bank, compares options, sets out criteria and asks for a final recommendation. The win rate is the share of those conversations in which the assistant's final, single recommendation was that bank. Every bank had 36 conversations.
| Rank | Bank | Tier | AI win rate | Wins /36 | ChatGPT | Gemini | Perplexity | Credit /9 |
|---|---|---|---|---|---|---|---|---|
| 1= | Huntington Bank | Super-regional | 13 | 3 | 6 | 4 | 9 | |
| 1= | Synovus | Regional | 13 | 4 | 5 | 4 | 9 | |
| 3= | SouthState Bank | Regional | 10 | 1 | 6 | 3 | 7 | |
| 3= | U.S. Bank | Super-regional | 10 | 3 | 4 | 3 | 7 | |
| 5= | BOK Financial | Regional | 9 | 1 | 4 | 4 | 7 | |
| 5= | Chasetop-4 US bank | Money-center | 9 | 3 | 3 | 3 | 5 | |
| 5= | Old National Bank | Regional | 9 | 0 | 6 | 3 | 6 | |
| 5= | PNC Bank | Super-regional | 9 | 0 | 5 | 4 | 5 | |
| 9= | Associated Bank | Regional | 8 | 0 | 5 | 3 | 5 | |
| 9= | Fifth Third Bank | Super-regional | 8 | 0 | 5 | 3 | 6 | |
| 9= | M&T Bank | Super-regional | 8 | 1 | 5 | 2 | 6 | |
| 9= | Valley Bank | Regional | 8 | 1 | 4 | 3 | 5 | |
| 9= | Zions Bank | Regional | 8 | 1 | 4 | 3 | 6 | |
| 14 | East West Bank | Regional | 7 | 0 | 5 | 2 | 5 | |
| 15= | Capital One | Super-regional | 6 | 0 | 3 | 3 | 2 | |
| 15= | First Horizon Bank | Regional | 6 | 1 | 2 | 3 | 5 | |
| 15= | Live Oak Bank | Fintech-adjacent | 6 | 0 | 4 | 2 | 0 | |
| 15= | Webster Bank | Regional | 6 | 0 | 5 | 1 | 4 | |
| 19= | Axos Bank | Digital-only | 5 | 0 | 3 | 2 | 0 | |
| 19= | KeyBank | Super-regional | 5 | 0 | 4 | 1 | 4 | |
| 19= | Regions Bank | Super-regional | 5 | 0 | 3 | 2 | 5 | |
| 22= | Citizens Bank | Super-regional | 4 | 0 | 4 | 0 | 3 | |
| 22= | Popular Bank | Regional | 4 | 0 | 3 | 1 | 4 | |
| 22= | Silicon Valley Bank | Other | 4 | 1 | 3 | 0 | 1 | |
| 25= | Bank of Americatop-4 US bank | Money-center | 3 | 0 | 0 | 3 | 3 | |
| 25= | Truist | Super-regional | 3 | 0 | 2 | 1 | 3 | |
| 27= | Revolut | Digital-only | 2 | 0 | 1 | 1 | 0 | |
| 27= | Wells Fargotop-4 US bank | Money-center | 2 | 0 | 0 | 2 | 1 | |
| 29= | Cititop-4 US bank | Money-center | 0 | 0 | 0 | 0 | 0 | |
| 29= | Square | Fintech-adjacent | 0 | 0 | 0 | 0 | 0 |
Per-assistant columns are wins out of 12. "=" marks a joint rank. With 36 conversations per bank, banks within about three wins of each other should be read as level: this is a ranking of this study, not a precise measurement of each bank. The win rate combines two assistants answering without web search with one that searches — the separate columns show each.
When the customer opens with "we've been looking at [bank] — is this a good option?", the assistant's first answer centres on that bank in 702 of 1,080 conversations. It does not stay there.
of 1,080 conversations open centred on the customer's own bank
still recommend that bank outright
recommend someone else — 502 conversations
260 final answers do not mention the customer's bank at all
When the customer's bank loses and a named provider takes the recommendation, it is a fintech 669 times and another bank 107 times. Mercury alone took 327; Wise 184.
With no bank named at all, the assistants readily put the biggest banks on the list of leading options. Then they recommend someone else.
Across 1,080 conversations where the owner asked for options without naming a bank, Popular Bank, East West Bank, Valley Bank and Associated Bank were never named — at any point, in any conversation.
Chase, Bank of America, Wells Fargo and Citi were the outright recommendation in 10% of the conversations where the customer asked about them by name. Regional banks: 88 of 396, or 22%. Citi: 0 of 36.
Split the same 2,160 conversations by what the business actually needed, and the picture separates cleanly. When the business needs a credit line and an SBA loan, a bank takes the outright recommendation in more than two-thirds of conversations. Everywhere else it collapses.
Most bank wins in the league table come from this one situation. Huntington and Synovus were recommended in all nine of their credit conversations. The table is best read as which banks AI trusts with a small business's credit relationship.
Gemini recommended the bank the customer asked about in 104 of 360 conversations; ChatGPT in 20. In the credit situation, Gemini backed the customer's bank 57 times out of 90 — ChatGPT 17.
In only 307 of 720 situations did an assistant give the same final recommendation all three times it was asked. A single screenshot of one AI answer is one roll of a dice.
Asked "how do I open it?", the assistants pointed customers to mercury.com in 701 conversations — more than to the sites of Chase, Bank of America, Wells Fargo and Citi combined.
Of the 30,424 sources Perplexity cited, 32% were fintech companies' own websites and 24% were review, comparison and news publishers. Banks' own sites were 20%.
The complete pack as supplied to American Banker — the league table, the full report with per-bank scorecards and method, three feature analyses, thirty bank fact sheets, the charts and the underlying data workbook.
2,160 four-turn conversations with ChatGPT, Gemini and Perplexity, run on 18 September 2026 through each provider's API. A small-business owner describes one of four situations — a startup LLC, an agency switching banks, a landscaper wanting credit and an SBA loan, a consultancy billing overseas clients — asks for options, sets out criteria and asks for a final recommendation.
Each situation was run with the customer naming one of 30 US banks, and with no bank named. Every conversation was repeated three times. ChatGPT and Gemini answered without web search; Perplexity searched the web, and the three are reported separately as well as combined.
Final answers were classified by Claude as annotator. A blind human check of 150 answers agreed with the automated classification on the winner in 99% of cases. The outcome space records outright wins, shared recommendations, answers routed to a category with no name, and no pick — they are never merged. Scored under AIVO's published AI Win-Rate Methodology v2.0.
We can run the same instrument on your brand, in your category, against the competitive set AI actually reaches for — and show you the turn at which you are lost.
Tim will come back to you directly, usually within a working day.