Emma Lurie

"We stand on the brink of an AI advertising age that will differ from social media advertising in important ways," Lurie wrote with Sorelle Friedler and Danaé Metaxa in a Brookings commentary on October 26, 2026. "The conversational tone of chatbots is persuasive," they added, and chatbot memory will let AI companies build detailed profiles of users and slip ads into ordinary answers.

The commentary summarized a study Lurie led with Friedler, Metaxa, and Ro Encarnación. Starting in February 2026, the team created 91 ChatGPT accounts that used IP proxies and location cues to signal Black, Hispanic, and White users across three income levels. They collected more than 3,000 ads from 186 advertisers in response to 335 prompts. Ads began appearing 14 days after an account was created, and accounts from lower-income zip codes were more likely to receive them, while the study found no evidence that race affected who saw ads.

Some prompts were obvious fits for ads, such as a question about the best streaming service for sports. Others drew odder matches: a request for a step-by-step pad Thai recipe turned up an ad for DoorDash. About a month into the study, every account stopped being shown ads, which the authors suggest may mean OpenAI detected them. The paper appears at AIES 2026 with a public, searchable archive of every ad collected, and the authors argue for mandatory transparency rules and researcher access so audits do not depend on fake accounts. "AI advertising may be about to rerun this playbook, and we would benefit from learning from the past and skipping to the resolution," they wrote, pointing to how outside investigations of Facebook's ads ended in a settlement and third-party oversight.

Lurie traces her interest in search to the 2016 election, when relatives and acquaintances kept forwarding her political emails she considered plainly untrue. "That made me wonder who started the emails in the first place and why did the recipients think they were true," she told the UC Berkeley School of Information in 2021. She describes her field as "online civic infrastructure," meaning how voters find information and the role platforms play in steering them to it.

A week before California's 2020 election, she heard a rumor that Google favored sides on Proposition 22, about app-based drivers, and Proposition 24, about consumer data privacy. She began saving Google results pages. "After 2016, it seemed like the right thing to do," she said. The project, built with undergraduate data science students, found no evidence of intentional bias, but results sometimes included summaries drawn from the California secretary of state's neutral election pages. "This is concerning from my point of view because this takes results from election authorities and can make them seem not neutral, which can make our democratic institutions weaker," she said. "With low levels of information, is it more or less likely that an election can be bought? This area is very understudied."

Lurie earned a bachelor's degree in computer science and Chinese language and culture at Wellesley College in Wellesley, Massachusetts. Working with Eni Mustafaraj, she published audits of how Google's results pages shape judgments of news credibility (2018) and of Google's Top Stories algorithm (2019). She also studied Google's Reviewed Claims fact-check labels, which news publishers criticized as "highly partisan" and "blatantly wrong," and co-wrote a 2020 FAccT paper making the case for voter-centered audits of search engines during elections.

She completed a PhD in information science at UC Berkeley in 2025 with a dissertation titled "Searching for Accountability: Contested Harms, Algorithmic Audits, and the Governance of Sociotechnical Systems," and a JD at Stanford Law School the same year. She joined Penn as a postdoctoral researcher in August 2025, working with Metaxa and Friedler. Her other 2026 papers compare federal AI transparency rules in the United States and, with Encarnación, show how ChatGPT's answers and cited sources shift between the chat interface and the API when web search is switched on.