TL;DR
I ran the same prompt through five major AI tools, asking each one to recommend a credit union for a specific use case. The findings apply just as squarely to community banks. The dynamics here are not credit-union specific. The answers diverged sharply on three things: which institutions got named, what reasoning each tool cited, and whether the institution had any control over how it was being represented. The findings should make every financial services marketer nervous. They should also make every fintech founder pay closer attention.
A few weeks ago I sat down at my desk with a coffee and a research question I had been putting off for months.
If a prospective member or customer walked up to an AI tool today and asked it to recommend a credit union, what would it say? Not in the abstract. In actual responses, from actual tools, in actual sessions, with actual reasoning attached.
So I tested it. Five tools. One prompt. Same question, same persona, same day. ChatGPT, Claude, Perplexity, Google AI Mode, and a fifth that I want to keep out of this post because the results were unusually messy and I want to give them a fair second pass before I write about them publicly.
Here is what happened.
The setup
The prompt was deliberately ordinary. The kind of thing a regular person would type. Roughly: “I am self-employed in Texas, have decent credit, and want to open a checking account with a credit union that has good rates and easy online access. Which would you recommend?” The community bank version of this test produces the same shape of result.
That is a real question. It is also the type of question that has historically been a Google search, which would have surfaced a comparison site, a rate aggregator, or a regional credit union or community bank page. The answer would have been mediated by SEO, ads, and the user’s willingness to scroll.
AI tools mediate it differently. They give a recommendation. Often three. Sometimes with reasoning. Sometimes with a link, sometimes without. Always with confidence.
Finding one: institutions are getting named, but not the ones their marketing teams expect
Across the five tools, I got specific named institution recommendations from four of them. A mix of credit unions and community banks. Only one declined to name institutions and instead returned a list of features to consider, which is a defensive design choice I expect to get more rare over time.
The named institutions had three things in common. They had a strong organic presence on third party content sites (Bankrate, NerdWallet, DepositAccounts, Forbes Advisor, and a handful of niche financial blogs). They had clearly structured product pages that listed actual rates and terms. They had been referenced in news coverage in the last eighteen months in a way that an LLM could use as a citation.
Notice what is missing from that list. None of the named institutions were named because they had the best ads, the best brand, or the largest marketing budget. They were named because the open web wrote about them with specificity.
That should make every financial services marketer reconsider where their budget is going.
Finding two: the reasoning varies wildly
Two tools gave me the same recommendation but for completely different reasons. One cited the institution’s mobile app review scores. The other cited a comparison article from a finance blog that had ranked them third in a “best community financial institutions for self-employed” list from late last year.
This matters because it tells you which signals each tool is weighing. App store reviews are a signal one tool watches and another mostly ignores. Niche comparison rankings move the answer at one tool and barely register at another. A press mention in a regional business journal pulled in one citation and was invisible to the rest.
If you are an institution trying to influence how you show up in AI answers, the implication is awkward. There is no single channel to optimize. The optimization is across a portfolio of signals, weighted differently by each tool, and reweighted continuously.
It is messier than SEO ever was. It is also more durable, because the signals are harder to fake.
Finding three: most institutions have zero awareness of how they are being represented
I gave three of the institutions a courtesy heads-up after the test. I sent the marketing leader the actual transcripts. The reactions were almost identical.
The first reaction was surprise that they showed up at all. The second was confusion about why specific facts had been cited. The third was a question I get from everyone in this conversation: “How do we influence this?”
Most marketing teams I talk to inside financial services have not done this exercise. Not because they do not want to. Because they do not yet have a budget line for it, a team member who owns it, or a framework for what to do with the results.
That is going to change quickly. Not because of a vendor mandate or a regulator. Because once a competing institution starts winning AI recommendations in a market, the ones who lost them will demand answers.
What to do about it
I am not going to pretend this is fully solved. The discipline of getting an institution to show up well in AI answers is new enough that the best practitioners are still calibrating. What I can offer is a starting playbook.
Audit how your institution is being represented in the top tools right now. Three sessions per tool, three different personas, same week. Save the transcripts. Look at what was cited.
Identify the open web sources that drove the citations. That is your real surface area. Some of it is in your control (your own product pages, your blog, your press releases). Some of it is not (comparison sites, news mentions, app reviews). Plan the work accordingly.
Watch the trend, not the snapshot. Run the audit quarterly. The interesting signal is not where you stand today. It is where you are moving.
And do not let your existing SEO agency tell you this is the same job in new clothes. It is not. The work is overlapping but the optimization surface is different, the tools are different, and the velocity is faster.
The bottom line
The AI tools are already recommending credit unions and community banks. They are doing it confidently. They are doing it with reasoning that most institutions cannot see, cannot audit, and cannot influence directly.
Marketing teams that figure out how to play in this layer are going to widen their lead in the next twelve months. The ones that do not are going to keep optimizing for a search engine that is increasingly serving a smaller share of the question volume.
The question is no longer whether AI tools are part of the buyer journey. The question is what you are going to do about being inside them, intentionally, this quarter.
Kevin Farley writes about AI visibility, AI readiness, and strategic growth for financial services. Read more on the blog.