Banking Communications Has a New Job: Underwriting Trust

For most of a century, a bank's trustworthiness has been a shared production. The bank behaved, and supervision stood behind the behaving. Customers never had to evaluate a bank's conduct directly. Regulators did it for them, continuously.

That arrangement is being unwound. Regulatory oversight is being relaxed, fee limits are being reconsidered, capital requirements are being reduced, and supervision of crypto and AI activity is being loosened. Whatever the merits, reputation is now carrying risk that supervision used to absorb.

Our 2026 research with American Banker shows that while the banking industry's Reputation Score kept most of its gains from a strong 2025, every measure of support from the Informed General Public fell year over year: The industry’s Reputation Score dropped slightly from 70.9 to 70.5, but the IGP’s willingness to do business with banks, to trust them, and to give them the benefit of the doubt in a crisis all dropped meaningfully more.

The public is specific about what worries it. About 60% remain extremely concerned about changes to banking rules and regulations, naming reduced CFPB oversight and the removal of fee limits. Few trust banks to use AI ethically, which the study flags as a significant reputational risk in its own right.

Non-customers judge banks on Conduct

Regional banks lead both audiences. They score 79.5 with their own customers and 68.2 with people who bank elsewhere. Non-traditional banks follow closely at 79.3 and 66.1. Large banks trail both, scoring 75.7 with customers and 56.6 with everyone else, which puts them in the Weak band.

Nobody moved much. All three types gave back a little ground with their own customers, and the industry score slipped 0.4 to 70.5. Large banks broke the pattern once, adding 0.5 with non-customers while regional and non-traditional lost ground. They also gained between 1.0 and 1.7 points on all seven drivers with their own customers. They climbed from last place and stayed there.

That 11.6-point gap to regional banks among non-customers is what large banks have to close, and the weights explain why it is hard. Every driver in our model carries a weight as well as a score, measuring how much an audience cares. Customers rank Products & Services first and Conduct second. Non-customers flip them and put Conduct above everything else. So the audience that rates large banks lowest judges them mostly on Conduct, and Conduct is exactly where large banks underperform themselves. They score 71.3 there, against 76.5 on Products & Services and 75.8 on Performance.

Deregulation hits the same driver. Narrowing oversight, reconsidering fee limits and loosening crypto and AI supervision all shape whether people see a bank as fair, transparent and ethical. Capital requirements speak to solvency instead. Supervision has stopped underwriting the one dimension that non-customers weight most.

Banks have less benefit of the doubt to draw on

Benefit of the doubt functions as a reserve. A crisis draws down the audience's willingness to wait for facts, and a bank holds whatever it accumulated before the day began.

The size of that reserve varies more than most reputation measures do. In RepTrak's benchmark data, 73% of people would give a company in the Excellent band the benefit of the doubt during a crisis. For a company in the Poor band, the figure is 8%. That range is the difference between an incident that gets a hearing and one that gets a verdict, and it accumulates over years.

Why the reserve declined is open. Cost-of-living pressure, general institutional distrust and AI anxiety are all live candidates, and the survey collected ratings without asking for reasons. The position is what the study establishes: banks are being asked to carry more of their own reputational risk in a year when the buffer for absorbing it got smaller.

Trust in self-regulation runs out on AI

On deregulation, the survey asked directly. Respondents were told supervision would be reduced or eliminated, then asked how far they would trust their bank to hold the line anyway. Complete trust ranged from 40% on holding sufficient capital down to under 30% on using AI ethically, averaging 36% across the five areas tested.

The people who do extend that trust look very different in the data. Respondents with complete trust in a bank to self-regulate rated that bank in the mid-80s, inside the Excellent band. Respondents with no trust at all rated the same institutions roughly 40 points lower, with the gap running between 37 and 46 points depending on the area.

Treat that spread as description. Self-regulation trust and reputation are plausibly two readings of one underlying attitude. The prior wave lacks this cut, so whether the spread widened is unknown. The tightness of the association is the finding, and it indicates what reputation is now being asked to carry.

Where that trust runs out is unusually specific. It's lowest, in every single segment we measured, on the ethical use of AI. Every age group, both genders, all three political affiliations, with no exceptions. That's the ranking. Overall trust levels vary by segment, falling with age and running lower among women. Large banks were the least trusted bank type on self-regulation generally, and the distrust deepened on AI specifically since 2025.

Agreement at that level almost never occurs in this dataset. It makes the exposure a named gap in a named area, which is something a bank can act on. The risk data points the same way. Of the 18 scenarios we tested this year, failing to use AI ethically and failing to prevent or respond to AI-enabled fraud both scored above average, at an expected 8.5 and 8.4 points of reputation damage, placing both in the top seven. The most damaging scenarios are still human ones, led by inappropriate behavior by management at 10.0 points.

Conduct becomes something banks can compete on

Under the old arrangement, supervision set a floor. Every supervised institution carried roughly the same assurance that the basics were being checked. The downside was broadly similar across the sector, while the upside always varied. As that floor thins, the gap between a bank that can demonstrate its own conduct and one that can't opens on the downside too.

The study establishes where the exposure sits. Which interventions close it is a separate question, and what follows is advisory judgment from practice.

Start by treating benefit of the doubt as inventory, built while nothing is wrong. In practice, that means moving the conduct evidence base onto owned channels now:

• What governs the bank's use of AI

• How customer data is protected

• How an automated decision gets appealed

• What past remediation actually produced

Aim that material at non-customers, who weight Conduct above everything else. Then vary the emphasis by how customers use AI themselves, which the study measured separately. Likelihood of use falls with age, from 43% of 18-to-24s to 17% of over-65s, and runs lower among women than men. Under 45, the top uses are comparing products and understanding fees, so lead with what the tools do and where the limits sit. Over 65 and among women, distrust of AI tools outweighs trust, and checking suspicious transactions is a top-three use for the oldest group. Lead there with human escalation, fraud reimbursement, and the right to decline AI tools altogether.

The score is the slowest measure in the model, so it will be the last place this work shows up. Start before it moves.

Working with this data

RepTrak measures corporate reputation on a 0 to 100 scale across stakeholder groups and markets. Underneath the score sit seven drivers and 23 factors, each carrying a weight that shows how much an audience cares, and below those the supportive behaviors this piece has been describing. Our advisory team turns all of it into insights, recommendations and actions each wave.

The full 2026 rankings, for customers and non-customers, are in American Banker's coverage of the study.

If you want to know where your own benefit of the doubt stands while nothing is wrong, talk to us.