Survey: Nearly half of Americans are comfortable with AI approving or denying their mortgage

Nearly half of American consumers are comfortable with artificial intelligence approving or denying their mortgage - the single most consequential yes-or-no in personal finance. That finding, from a new survey of 1,000 U.S. consumers by mortgage marketplace Lower, is one of the clearest measures yet of how quickly AI has moved from answering questions to making decisions.

The comfort is not limited to the final verdict. According to the survey, 67.3% of consumers are comfortable with AI recommending how much they should borrow, 64.7% are comfortable with it determining whether they qualify at all, and 54.3% are comfortable with AI reviewing their financial documents. More than seven in ten would let AI recommend a mortgage type. The approval decision itself - 49.2% comfortable - is the only task tested where comfort falls below half, and it is barely below.

## Comfort Follows Experience

The sharpest divide in the data is exposure. Among consumers who have already used AI for a mortgage or loan question, 72.1% are comfortable with AI approving or denying an application. Among those who have not, the number is 41.1%. Current mortgage holders are more comfortable than consumers who have never had one and do not plan to. "Once AI has answered the question and saved someone real time, they'll trust it with more," said Gino Fronti, VP of product for LOAI at Lower. "People have seen it work."

But comfort is not consent. Asked what role AI should play overall, the most common answer, chosen by 38%, was that AI can make recommendations while a human makes the final decision. Another 18.8% would accept AI making some decisions independently as long as major decisions still get human review. Only 5.2% were comfortable with AI making most decisions independently, and just 4.3% would hand over the entire process. When an unexpected problem threatens closing, 82.8% want a human involved: 54.4% would turn to a human loan officer, 28.4% want a loan officer and AI working together, and only 7% would rely on an AI assistant alone.

The survey's authors read this as a mandate for a hybrid model - AI at the front of the process, humans at the moments that matter. The single most requested feature for making consumers comfortable was access to a human whenever they want one, at 49.1%, followed by human review of important AI recommendations, at 38.1%.

## The Law Is Not Moving as Fast as the Sentiment

Lenders have an obvious incentive to move quickly. Mortgage origination is slow, document-heavy and expensive, and AI promises to compress underwriting from weeks to hours while cutting fixed costs. If consumers are genuinely half-comfortable with algorithmic underwriting, the last soft barrier to adoption is already lower than the industry expected.

The harder barriers are legal. The Equal Credit Opportunity Act still requires lenders to give applicants specific reasons for an adverse decision, and it prohibits discrimination on the basis of protected characteristics. Both requirements are awkward fits for machine learning models whose internal logic resists plain-language explanation. The risk with AI underwriting is not that it is new; it is that models trained on decades of lending history can quietly reproduce the disparities baked into that history, at scale, behind an aura of mathematical objectivity.

Regulators have spent the last several years circling this problem without settling it, and the consumer comfort documented in this survey all but guarantees the technology will arrive at scale before the rules do. That makes the survey's human-in-the-loop preference more than a consumer quirk. It is, functionally, the last line of defense against failure modes that fair-lending law has not yet been redesigned to catch.

## What This Means For You

**If you are applying for a mortgage:** Assume AI will be involved in reviewing your file whether or not anyone tells you, and act accordingly. You are entitled to specific reasons if you are denied, so request them in writing. If a decision seems borderline, ask for human review - lenders that deploy AI responsibly have a process for exactly that, and the survey suggests most borrowers want one.

**If you work in lending or fintech:** The survey is permission to build, with a condition attached. The 49.1% of consumers who want on-demand human access are effectively writing your product spec. Institutions that treat human-in-the-loop as a cost to be minimized will be the ones explaining their model to regulators after the first discrimination complaint; the ones that design for explainability and human escalation up front will move fastest through a market this survey says is ready.

**If you make policy:** Read the 49.2% number as a deadline. Comfort is here; the rules are not. The window for requiring auditable decision trails, meaningful adverse-action notices and bias testing for underwriting models closes with every quarter that adoption outpaces oversight. Consumer sentiment will not wait for the statute book, and the statute book should not wait for a scandal.