The reputational cost of a bad chatbot experience

Since 2024, virtually every large US bank has put an AI layer between its customers and staff. Chatbots drive interactions in the app; voice bots answer the phone lines; and automated verification runs across the whole system.

We wanted to explore the degree to which banking customers were happy or not with their bot interactions. Specifically, had the number of customer complaints risen as AI presence increased? And, what kinds of interactions caused the most dissatisfaction?

We analyzed the public review record of 100 large US banks, including 2,672 snippets across 57 institutions with Trustpilot profiles. From archived snapshots of those pages, we reconstructed a time series of 3,735 dated, full-text reviews logged between January 2023 and March 2026.

For comparison, we also ran Revolut and Monzo (two digital challengers from the UK, with 458,000 combined reviews) through the same process. We chose these two because they are the digital natives that every traditional bank now competes with. Their digital experiences set the benchmark, and part of our hypothesis was that their reviews would reflect a superior experience, which makes them a useful baseline against which to read the incumbents.

Firstly, the data validated our assumptions that the majority of US digital banking customers were dissatisfied with their AI-driven experiences. Yet, beyond that, the sample turned up patterns we hadn’t expected:

• consumer dissatisfaction can be strong enough to invoke regulators
• decades-long relationships can be cut short due to a single unresolved incident
• automated assistants do their worst work with estate cases and expat lockouts
• small businesses in particular rate their AI-assisted banking activities most poorly.

Following, we detail five key findings from our research, along with five surprising insights that surfaced in the process.

Across the 57 banks with Trustpilot profiles, large retail incumbents clustered between 1.1 and 1.9 stars—three full stars below Revolut (4.7 stars, 391K reviews) and Monzo (4.6 stars, 67K reviews). Meanwhile, in the dated sample, incumbents averaged 2.11 stars and were falling.

Contrast that with credit unions and digital-first players, who treat the review channel as part of the product and were the top outliers: Navy Federal (4.5 stars, 47K reviews) and SoFi (4.0 stars, 10.8K reviews).

Incumbents’ average dated-review rating fell from 2.39 stars (2023) > 2.29 stars (2024) > 2.11 stars (2025–26), while their one-star share climbed from 60% to 68%. Over that same period, challengers moved in the opposite direction, rising from 3.34 to 4.49 stars.

Note the share of one-star dated reviews (below)—indicating that anger against incumbents deepened, while challenger detractors thinned out.

When we reclassified every negative (≤2 star) dated review against a consistent taxonomy, the complaint mix not only rotated but also grew. (Interestingly, only two themes decreased: branch and staff problems—from 19.4% to 17.0%—and complaints about hold times and unreachable call centers—from 15.3% to 11.3% of negative reviews.)

The largest complaint surface in 2025–26 was the digital channel itself. From app login issues to lockouts, verification loops to bot walls, automated systems “do everything possible not to connect you to a human,” as one reviewer put it.

When we combined “app/online failures” + “AI/bot complaints” + “no-human-available” themes, digital interaction failures appeared in 38% of negative reviews in 2025–26, a slight increase from 34% in 2023.

Even so, this understates a bigger problem: a further 16.7% of negative reviews described customers being forced offline entirely—told to visit a branch, mail a form, or send a fax to resolve something the digital channel started.

Explicit blame of AI, chatbots, or automated systems tripled from 0.8% in 2023 to 2.4% in 2025–26. Though a small share, this gives customers a new grievance: “no human available.”

“Long wait-time” appeared in 2% of Revolut and Monzo negative reviews vs. 11% of incumbents. Remaining complaints focused on risk operations, including fraud handling and account freezes. Meanwhile, customers praised instant execution, app speed, travel and FX handling, and budgeting control.

This is where things begin to get very interesting: 15.8% of all negative incumbent reviews (a total of 310) indicated that an escalatory action was taken due to their dissatisfaction with the bank. This reportedly* included filing CFPB complaints, contacting state attorneys general, opening cases with the Better Business Bureau, and invoking class action (“I will pursue legal remedies.”)

That being said, the sample supports two readings: first, regulator mentions hold stable across all three periods, which is evidence of settled behavior rather than an outlier response. For a meaningful minority of customers, the CFPB is now a standard service step, listed in reviews the way an order number would have been years ago. Second, one escalation category did in fact fall: “acquisition and loan-sale fallout,” from 5.8% to 1.4%. This demonstrates that the record responds to real-world events, and the 2023 integrations (Union Bank into U.S. Bank, Bank of the West into BMO Harris, TCF into Huntington, and Flagstar) flooded the channel with conversion complaints that have since been worked through. And so, we can conclude that the record is a functioning sensor. What it senses in 2025–26 is service design rather than merger indigestion.

Tenure statements open 6.5% of the negative reviews (127 total): “Customer for 25 years,” “Banked here since 1983,” “40 years, multiple homes and vehicles.”

Banks should read these twice. The authors are narrating the end of the most profitable relationships the bank has—one that required minimal input of time and investment, and still compounded over time.

A further 5.1% of negative reviews announced a switch in progress, such as “taking my business to a credit union,” alongside customers closing accounts and moving money. Combined with the tenure citations, roughly 1 in 9 negative reviews described measurable, in-flight attrition rather than mere mud-slinging or venting.

Such negative ratings, taken en masse, produce a damaging narrative that other consumers can see and act upon, alongside the AI assistants that provide recommendations on “the best bank” for them.

Experiences with bereavement, estate, and probate management, dealing with elderly or otherwise vulnerable customers, or customers who get locked out of their assets while abroad, may be the most rare for customers, but they should also be the most routine for banks.

And while the negative reviews about life moments made up the smallest share of the sample, they also tended to be the longest, most detailed, and most legally threatening overall. What’s more, they were effectively absent from challenger bank records (partly due to a younger customer base, and also its partly digital-first estate workflows).

One widow describes a two-year, 17-call effort to close a $5 account. Forty reviews describe estates and bereavement, the same story told 40 times. Death certificates are mailed, faxed, and “lost.” Beneficiary departments can’t be called, only awaited. Funds are released only after the CFPB or a lawyer is engaged. Estate handling is a pure process problem, with no credit risk and no fraud ambiguity. It’s also where the gap between automated front doors and back-office reality is widest. A bot can’t process a death, and the humans behind it have been made unreachable.

Expat lockout, on the other hand, is the same architecture failing in a different way. When security models assume that a US phone number and a domestic IP are the only legitimate signifiers, they treat every American abroad as an attacker. Eight reviews describe multi-month lockouts while overseas—relatively few in number, but sizable in impact.

The negative reviews from business owners averaged 1.03 stars—the lowest of any segment we could isolate. True, they totaled just 128 reviews, 6.5% of the sample. But frozen or closed accounts and held funds appeared in 35% of them, against roughly 6% of consumer negatives.

At stake are payroll runs, contractor payments, and a $30,000 wire fraud on a business account with reimbursement denied. Several reviews described companies that were unable to pay staff for weeks while a compliance review continued without explanation, timeline, or a named human taking ownership of the process.

Interestingly, one of the challengers—Revolut Business—drew the same frozen-funds complaints when its own risk operations backfired. This makes “small business” the segment where our comparisons least-flatter the challenger banks. But the difference is in how cases get resolved.

The way that banks handle reviews makes a difference: their review pages are themselves a kind of customer service channel. Some banks go the extra mile to address customer concerns, while others hardly bother.

When banks reply, two patterns emerge in opposite directions:

A templated “we’re sorry, please call us” reply sends the customer back into the queue they’ve already been complaining about, thereby reinforcing the dead end (though politely, this time). CIT Bank posted 67 such replies at a 1.6-star rating.

In contrast, Navy Federal’s named-human replies, Fifth Third’s branch-level engagement, and the challengers’ case-thread routing all do the one thing that the boilerplate response can’t: they move the resolution to a place where it can actually be resolved.

The cut-and-dry verdict: Yes, based on the sample data, AI integration among incumbent banks has influenced customer reviews for the worse, not the better. Specifically, explicit blame of AI, bots, and automation, counted together with “no human available” complaints, roughly quadrupled from 0.8% of negative reviews in 2023 to 3.4% in 2024—a much smaller magnitude than our hypotheses at first supposed. What’s more, the share plateaued at around 3% in 2025–2026, even as AI uptake increased. And, taken a step further, when we remove the “no human available” complaints, explicit AI blame decreases again.

This study makes several conclusions clear:

• Customers who are angry at a bank in 2026 are still mostly dissatisfied with money mechanics (e.g., fees, holds, frozen funds, denied disputes), rather than the role that AI plays in delivering such outcomes.
• The depth of reviews in any given period of time, as well the timing of chatbot deployment, have a recognizable influence on negativity: So, for example, Q3-2023 registered only 40 reviews, and Q1-2026 registered a small, late-arriving sample—both quarters demonstrated a lower-than-average share of negative reviews—while the level-up across 2023–2024 coincides with mass chatbot adoption in the industry.

Furthermore, three secondary signals show AI-era service design reshaping not just what but how customers experience those failures:

• Hold-time complaints fell (15.3% to 11.3%) as bots absorbed first contact. The queue got shorter because fewer people reached it.
• Digital dead ends rose to 38% of negative reviews, and “forced offline” resolution (branch, mail, fax) rose to 1 in 6 customer reviews. The human escape hatch was being removed faster than the automation could carry the load.
• The tone changed. In 2023, complaints read as impatience. By 2025–26 they read as abandonment: 40-year customers described verification loops with no exit; disabled accounts locked out abroad; executors faxing death certificates into the void.

Ultimately, customers were shouting at the machine because it was the last thing that would ever answer them back:

All of these insights point in the same general direction: customers aren’t asking you to build new machine intelligence. And they’re not necessarily averse to AI assistance. They simply want better processes and outcomes.

The common thread is resolution. AI that only talks is the problem. AI that can block a card, raise a limit, dispute a payment, and close most matters on the customer’s behalf is the fix, which is what Backbase’s agentic assistants are built to do. These steps can help you to deliver more of what your customers want and need from their AI-assisted banking experiences: