Police officers must get consent before they search someone unless there is probable cause or evidence of a crime. In practice, officers might use explicit terms like “consent” and “search” or the more ambiguous request, “You don’t mind if I take a look?”
Police body-worn cameras document interactions between officers and the public, and they capture exactly how consent was gained.
A new study by UC Davis and Stanford SPARQ applied AI tools and techniques to New York Police Department, or NYPD, body-worn camera footage to learn whether officers followed legal rules for stops and searches. The study, part of independent monitoring of NYPD search practices, found that officers frequently used ambiguous phrasing in requests for consent and that their language often depended on a person’s race.
“The prevalence of implicit requests and ambiguous phrasing in consent searches raises questions about whether individuals understand they are being asked to consent, that they can refuse, and what they are consenting to,” said Rob Voigt, the study’s lead author and an assistant professor of linguistics in the College of Letters and Science at UC Davis.
Stop-and-frisk and a New York District Court ruling
The police search policy “stop-and-frisk,” in which officers temporarily detain people to ask questions and physically check for weapons does not rely on probably cause. The policy first emerged in the 1950s and expanded in New York City throughout the 2000s.
In 2013, the U.S. District Court for the Southern District of New York found that NYPD stop-and-frisk practices violated the Constitution’s Fourth Amendment, which requires stops be based on reasonable suspicion. The court determined that the practice also violated the Fourteenth Amendment because stops were based largely on race and were a form of racial profiling.
The court ordered the NYPD to have body-worn cameras to capture what officers say when they make a stop. This court-approved study led from UC Davis used AI and language-based processing methods to analyze over 6,553 videos from body-worn cameras recorded from 2022 to 2024.
The analysis found a low amount of explicit language related to consent in the camera footage. The results suggest that at least 75% of interactions that involve a search do not contain the word “consent” and at least 40% do not contain the word “search” in any video.
The analysis also showed differences in the language officers used depending on a person’s race or ethnicity. Officers used a more ambiguous “[do you] mind” request for consent in 20.6% of interactions with people who are Black but in 15.3% of interactions for people who are Hispanic, white or some other race or ethnicity.
Interactions involving a search with people who are Black or Hispanic also included commands 20% more frequently than with people who are white.
Improving review of body-worn camera footage
This study led from UC Davis is the second court-approved study of body-worn camera footage to evaluate whether NYPD officers are complying with the 2013 court ruling. The first was conducted by the City University of New York’s Institute for State and Local Governance. Its May 2025 report documented significant rates of unconstitutional stops and encounter underreporting by hand-coding of footage from police encounters.
This UC Davis study’s AI-approach to analyzing body-worn camera footage could be extended to analyze footage more comprehensively, and in other police departments around the country. The study’s AI-powered predictive models can identify policy violations, such as incomplete documentation or unconstitutional stops, at high accuracy and at scale.
The research team has spent several years refining this approach in California, where two police departments have shared full access to 1.3 million videos representing more than 300,000 hours of footage for analysis.
“Compared to random audits of footage, this approach would maximize supervisors’ valuable time while reducing the time and resources required to achieve the same level of oversight,” said Voigt.
The full report was published on April 29 by NYPD Monitor and by Stanford SPARQ, a behavioral science research center, where Voigt is a faculty affiliate. Co-authors include Nicholas P. Camp of the University of Michigan, Dan Sutton of Stanford Law School and Jennifer L. Eberhardt of the Stanford School of Business and faculty co-director of Stanford SPARQ.
You may also like these stories
Deportations Meet the Demographic Cliff
The sudden drop in the number of immigrants in the U.S. comes at a time when fertility rates among the native-born population are also falling. Research in economics suggests that these two trends could tip a precarious economy into decline.
Research Suggests Nationwide Racial Disparities in Media Reporting on Gun Violence
Mass shootings in white-majority neighborhoods received roughly twice the news coverage of mass shootings in neighborhoods where a majority of residents were people of color, while coverage of police-involved shootings was disproportionately high in majority-minority communities, according to new research in linguistics.