Most published research on AI visibility in banking ranks banks by how often they are named. A digital marketing team planning next quarter's work needs to know which sources the engines read before naming a bank. Those are the pages the team can work on.
This report combines published studies with CB/I Digital's own reading of Ahrefs Brand Radar data, pulled on 1 October 2026, across ChatGPT, Perplexity, Google's AI Overviews and AI Mode, Gemini and Microsoft Copilot. It is the first edition. It uses Ahrefs' own index of questions built from search data, not a question set CB/I tracks for a bank. The method and its limits are set out near the end.
When an AI engine searches the web before answering, it names institutions from the sources it retrieves, alongside what its model already learned in training. A bank can influence the search step. If it is missing from the sources an engine retrieves, it has fewer opportunities to appear in the answer, even with a strong brand.
The platforms describe the retrieval step in their own documentation. Google says AI Overviews and AI Mode may use a "query fan-out" technique, issuing several related searches across subtopics before writing a response, and its models identify more supporting pages while the response is being generated. OpenAI said in its 2024 launch post that ChatGPT search draws on third-party search providers and on content from its partners. Perplexity searches the web in real time and shows numbered citations for each answer. Microsoft's documentation for its workplace version of Copilot says it turns the user's prompt into a short query and sends it to Bing.
Because the answer depends on what gets retrieved, the type of question matters more than the bank's size. New Media Advisors, an agency that analyzed more than 14,000 AI answers about banks and credit unions across 129 US markets, found credit unions taking roughly 61% of recommendations on consumer questions, while regional and community banks took 60% to 85% on business questions such as commercial banking relationships. The same study found AI citation counts correlating with Google top-three rankings at 0.81, which its authors describe as an association rather than proof of cause. Banks that rank in Google's top three more often also tend to be cited more often, though the study does not show that one causes the other.
Concentration also varies by product. In NP Digital's analysis of more than 68,000 AI answers about fintech brands, Klarna appeared in 91.8% of answers in buy now, pay later, while no lending or credit brand cleared 43.7% in its category. In the category dominated by one brand, the engines named it in almost every answer. Lending and credit answers named a wider range of brands.
Mostly publishers and comparison sites, with clear differences between the two. ChatGPT draws on a broader mix that includes banks and government agencies, while Perplexity relies almost entirely on financial publishers.
In the Ahrefs data for deposit and everyday banking questions, ChatGPT's most cited domains were NerdWallet, Bankrate, The Wall Street Journal and Forbes, followed by Capital One, Chase, the FDIC, the CFPB, Finder and Ally. Three of its top ten are banks and two are federal agencies. Perplexity's top ten were all publishers or forums, led by Bankrate and NerdWallet and including Yahoo Finance, CNBC, Business Insider, Reddit and Investopedia. In the eight Perplexity example answers we read, each cited ten links, against one to five on ChatGPT. That may spread Perplexity's citations across more publishers, though we did not measure it.
The table below extends the comparison to all six platforms. Each figure is that kind of source's share of all citations to the platform's 50 most cited domains, counting each domain once per answer.
| Platform | Most cited domain | Banks and credit unions | Publishers and comparison sites | Government agencies | Forums and video |
|---|---|---|---|---|---|
| ChatGPT | NerdWallet | 27% | 56% | 11% | 1% |
| Perplexity | Bankrate | 9% | 78% | 0% | 6% |
| Google AI Overviews | Bankrate, level with NerdWallet | 24% | 59% | 0% | 12% |
| Google AI Mode | Bankrate, level with NerdWallet | 18% | 63% | 0% | 9% |
| Gemini | Bankrate | 30% | 52% | 0% | 4% |
| Microsoft Copilot | Bankrate | 8% | 85% | 0% | 0% |
Source: CB/I Digital analysis of Ahrefs Brand Radar, US prompt index, deposit and everyday banking questions, pulled 1 October 2026. Rows do not sum to 100% because fintech, review and other sites are not shown.
Government sources appear among the top citations only on ChatGPT, where the FDIC and the CFPB are both in the top ten. Reddit and YouTube matter on Google's engines, where they make up 12% of the top citations in AI Overviews and 9% in AI Mode, against less than 1% on ChatGPT. Banks are cited most on Gemini and ChatGPT and least on Perplexity and Copilot. Gemini's top ten include PNC, Chase and SoFi; ChatGPT's include Capital One, Chase and Ally.
Published studies point the same way once their questions are taken into account. New Media Advisors found independent third parties providing about 62% of citations on national banking questions and 52% at city level, with institution-owned websites at 38% nationally and closer to 48% locally. The PR firm 5W, using national prompts, found bank-owned domains below 7% of banking citations, on a measure that blends citations with mentions. Yext, a listings software company, found 47% of citations on location-based financial services queries coming from first-party websites and another 41% from directory listings. NP Digital, looking across industries, found 82% of citations on commercial prompts going to third parties and 3% to owned pages. The studies differ in method and each publisher sells related services, but together they suggest that the closer a question gets to a specific bank or place, the more weight the bank's own pages carry.
The ChatGPT examples we read were consistent with that pattern. Questions that named a bank, such as whether a particular bank's checking account was a good choice, were often answered with the bank's own page as the only link. Generic questions about the best account or the highest rate went to publishers. We did not count that split, so it remains an observation from the sample rather than a measured result.
The product and location in the question determine which sources are relevant to the answer. Rate-driven products such as CDs and savings are answered mostly from publishers that compare rates. Relationship products such as business checking draw more on the banks themselves. Questions that name a place bring in local institutions that national answers leave out.
The product split in the Ahrefs data is wide. On ChatGPT, banks accounted for 52% of the top citations for business checking and 35% for checking, against 11% for savings and 9% for CDs. Mortgage answers on ChatGPT drew only 3% of their top citations from banks, and the CFPB was the most cited domain for mortgage questions. On Google's AI Overviews, CD answers drew 96% of their top citations from publishers, and single publishers appeared in a large share of answers. Investopedia was cited in 72 of 129 CD answers, Bankrate in 117 of 193 home equity line of credit answers, and NerdWallet in 40 of 65 business checking answers.
Place works through location settings and local sources. Google says it uses the user's current location to show nearby results even when a search does not name a place, and Perplexity says it falls back to the user's network location when no location is set. New Media Advisors ran the same prompts in 13 markets and found that markets with a dominant home-market bank lean local. In Utah, Bank of Utah led the state at roughly 60% visibility and eight of the ten most cited domains were Utah bank sites. In Minneapolis, US Bank reached 68% visibility.
Our own local sample was small, between 22 and 91 answers per platform, and most of its questions named a state rather than a city. Even so, local questions brought in 38 domains that did not appear in any platform's top 50 for general banking questions, 13 of them credit unions or regional banks. One Florida credit union was cited on all four platforms we tested for local questions. On ChatGPT, local answers also shifted toward rating and directory sites such as credit union rating pages and J.D. Power. Measured over each slice's 25 most cited domains, their share rose from none in general answers to 19% in local ones, while the share going to banks fell from 23% to 7%.
A large bank's national product pages compete with different sources than its branch and market pages do. Local questions are where New Media Advisors found regional banks and credit unions doing best, and in our sample they brought in institutions that national answers left out. Branch and market pages, local terms and directory listings are what local answers can draw on.
Plan by the type of question, since each type is answered from a different kind of source. The table below maps the patterns in this report to where a bank's effort is likely to count.
| Question type | Sources the engines tend to cite | Where a bank's effort counts |
|---|---|---|
| About the bank by name, such as whether its account is a good choice | The bank's own product pages, when they can be read and answer the question | Product pages with a self-contained answer passage, open to the crawlers each platform uses |
| Best account or highest rate in a category | NerdWallet, Bankrate, Forbes, Investopedia and similar publishers | Accurate, current listings on those publishers, and a product competitive enough to be included |
| Rules and protections, such as deposit insurance or mortgage terms | Government sources on ChatGPT, publishers elsewhere | Wording on the bank's pages that matches FDIC and CFPB terminology |
| Experiences and opinions | Reddit and YouTube, especially in Google's AI features | Useful video and genuine participation, covered in our article on how banks earn citations beyond their website |
| Local, such as the best bank in a state or city | Local banks, credit unions, rating sites and directories | Branch and market pages, local terms and consistent directory listings |
The rules and protections row is Your Money or Your Life content in Google's terms, where accuracy carries the most weight, so wording there belongs in compliance review. Our guide to measuring a bank's AI search visibility explains which indicator shows progress for each row. Our article on why banks are losing the first customer conversation to AI explains the role of the bank's own site as the definitive source for facts only it can provide, and the role of comparison sites in answering comparison questions. Our guide to running an AI visibility audit for a bank shows how to check which of these patterns applies to a bank's own priority products.
The CB/I Digital figures come from Ahrefs Brand Radar's US prompt index, which holds questions Ahrefs collects from search data and the answers it records on each platform. We filtered for questions that mention savings accounts, checking accounts, certificates of deposit, money market accounts, business checking, home equity lines of credit and similar everyday banking terms, then counted, for each platform, how many answers cited each domain. A domain cited several times in one answer counts once. The source mix is computed over each platform's 50 most cited domains, so the long tail of smaller sites, which likely includes more local banks, is not included.
The filters are imperfect. A few matched questions were not about deposit products, the local filter caught a handful of unrelated questions, and Perplexity's US results also cited some UK consumer finance sites. We classified each domain by hand, counting chartered online banks such as Ally and SoFi as banks. The answers in the index lean toward higher-volume questions, and the data is a snapshot that will change, since AI answers shift from week to week. The answers in the index were collected before the pull date, and cited page titles mention July and August 2026.
The published studies cited here are vendor research, from two agencies, a PR firm and a listings software company, and each has an interest in the results. Ahrefs, whose volatility study appears in the FAQ, also sells the tool that supplied our data. Their methods differ and are only partly public, so their figures should be compared by direction rather than combined.
The source map in this report is general. For a bank, the useful version covers its own priority products and markets, and shows which domains shape each answer and whether the bank's pages are among them. CB/I Digital builds that map as part of our AI search work for banks and turns it into a list of pages and listings to fix.
That work is part of our SEO & AI Search services. If your priority products do not seem to fit the platform patterns in this report, we can compare your categories against them and show where your bank differs.
Why do the cited sources change from one week to the next?
AI answers are rebuilt each time and can draw on different sources. An Ahrefs study found that the content of AI Overviews changed every 2.15 days on average, with close to half of the cited sources entirely new over the period studied. One week's reading can reflect that churn, so changes are better read across repeated monthly runs, as our guide to measuring a bank's AI search visibility explains.
Why does ChatGPT cite the FDIC and CFPB so often?
In our data, government sources made up 29% of ChatGPT's top citations for mortgage questions and 20% for home equity lines, against about 11% for savings and checking. We did not test why ChatGPT favors them. For a bank, the implication is that its own explanations of insurance coverage and loan terms should be consistent with how those agencies describe them, and any wording about deposit insurance should go through compliance review.
Does paying for a comparison site listing get a bank cited?
No study we found shows that. AI answers cite the editorial pages of comparison sites, and whether a paid placement affects those pages depends on each publisher's policies. Paid listings are also advertising the bank is responsible for, so their claims need the same review as the bank's own copy. Our article on keeping bank content compliant while optimizing for AI search covers third parties that publish on a bank's behalf.
Copyright © 2026 CB/I DIGITAL INC. All rights reserved | Privacy Policy