US Home-Pricing and Tenant-Screening Algorithms Draw Discrimination Scrutiny

For several years now, artificial intelligence has been pricing American homes in a fraction of a second and deciding who gets the keys to a rental unit. The same technology that was supposed to level the playing field for buyers and renters against sellers and property managers is, in practice, increasingly acting as an invisible filter that sorts people by income, race and address.

Valuation in Seconds

Automated valuation models, known in the industry by the acronym AVM, analyze hundreds of thousands of transactions and produce an estimated home value within seconds. Zillow's Zestimate is the best-known example, covering 116 million homes across the United States, with a median error of about 2 percent for properties currently listed for sale.

For properties not currently on the market, the error rises to around 7 percent, which in real dollar terms can mean a difference of tens of thousands of dollars. Zillow learned the scale of this problem the hard way: its own home-buying program, Zillow Offers, systematically overpaid for properties based on flawed model forecasts, resulting in losses in the hundreds of millions of dollars and the program's shutdown.

Who Screens the Tenant

The other side of the same technology is tenant-screening algorithms, which property managers use to decide who gets to rent an apartment. SafeRent settled in 2024 for $2.3 million after its applicant-scoring system was found to systematically lower the scores of Black and Latino renters as well as holders of housing subsidies.

Studies cited in the analysis show that algorithms of this kind reject as many as 10 percent of housing voucher holders, since they require income up to three times the rent without accounting for public subsidies that actually cover part of the payment. As a result, people eligible for housing assistance lose access to apartments even though, on paper, they can afford the rent.

Price-Fixing by Algorithm

A separate case involves RealPage, a provider of rent-setting software, which paid $50 million in a class-action settlement. The allegation was that the company's algorithm, by pooling data from many competing property managers, effectively allowed them to coordinate rental prices instead of competing for tenants, pushing rents above what normal market competition would produce.

The RealPage case has led several US states and cities to restrict algorithmic rent-setting software, treating it as a mechanism that enables price collusion even when no direct conversation between competitors ever takes place.

The Black Box

The analysis points to a fundamental asymmetry that has accompanied this technology since the real estate market's earliest days. Historically, sellers, brokers and banks held information the average buyer or renter couldn't access. AI was supposed to close that gap, giving ordinary people the same analytical tools as professionals.

In practice, the effect is often the reverse. People with solid incomes and good addresses benefit from the full transparency algorithms provide, while for poorer people or those from less desirable neighborhoods, the same systems act as a filter that blocks access to listings. Personalization, which was meant to match offers to users' needs, in practice also means that some opportunities are simply never shown to them at all.

What Comes Next for Oversight

The demand raised most often in this context is for algorithms to be transparent and testable for discrimination, and for people to have a guaranteed right to appeal an automated decision to a human being. In the United States, regulation is moving in that direction piecemeal, state by state and city by city, without a unified federal framework for valuation and screening algorithms.

For Poland's real estate market, cases like SafeRent and RealPage remain a distant scenario, as algorithmic valuation and tenant screening are only just emerging there. Still, the EU's AI Act and national data protection law already provide a legal basis for challenging automated credit or hiring decisions, and the same mechanisms could eventually extend to algorithms operating in the rental and home-sale markets.

A home remains a different kind of good from a typical consumer product, since it often determines access to a good school, a job or health care. A decision about who gets one, made inside an opaque algorithm, therefore provokes stronger resistance than other consumer applications of artificial intelligence.