Danaé Metaxa

"Online platforms like social media sites, search engines and LLM chatbots are used more and more each year," Metaxa told Penn Engineering in 2026. "At the same time, companies have become less willing to work with researchers and provide open data."

Their lab's answer to that problem in search was a field experiment. Led by Stephanie T. Wang and Jeffrey Gleason, with Yakov Bart and Christo Wilson of Northeastern University, the team gave 1,100 people a browser extension for ten days and randomly assigned them to Google as it was, Google with AI Overviews removed, or Google's AI Mode. Removing AI Overviews raised click-through to outside sites by 8.8 percentage points. AI Mode cut it by 18.8 points, with drops for news sites, Reddit and Wikipedia, and it lowered people's trust in and satisfaction with Google. The preprint, released in August 2026, is under review at PNAS Nexus.

Advertising is the other half of the lab's work. "We are the product, and our attention is getting bought and sold by different companies," Metaxa said in a 2025 interview with the Connected Learning Alliance. That year they won a $670,000 NSF CAREER award for audits of targeted advertising, including Admix, a tool that lets users choose, alone or together, which kinds of ads they are willing to see.

When OpenAI began showing ads in ChatGPT, Metaxa, Emma Lurie, Ro Encarnación and Sorelle Friedler set up 91 test accounts in February 2026, signaling different racial groups and income levels, and collected more than 3,000 ads from 186 advertisers across 335 prompts. Accounts began receiving ads 14 days after creation, and lower-income accounts were more likely to get them regardless of race. The early ads leaned heavily toward consumer goods, pointed to an advertiser more often than a specific product, and sat clearly apart from the answer text. The paper appeared at AIES 2026 with a public archive of every ad collected.

In 2026 Metaxa co-wrote "Auditing AI," an open-access book in the MIT Press Essential Knowledge series. One of its examples is SABRE, American Airlines' reservation system, which rival airlines showed was favoring American's flights in its search results. "Anyone who's used an online search system and seen a little dropdown in the corner indicating results are ordered by 'Top Recommended' instead of 'Increasing Price' or another objective measure is interacting with a system like SABRE, where the developers of the system are biasing what you see," Metaxa said.

The lab applies the same method to chatbots. A FAccT 2026 study compared thousands of answers from tools such as ChatGPT and Claude on abortion law with a state-by-state legal ground truth. Accuracy averaged about 78% across all 50 states but fell to 44% in North Dakota, against 99% in Vermont.

Metaxa earned bachelor's degrees in computer science, with honors, and in science, technology and society at Brown University in 2015. At Stanford they completed a master's degree in 2018 and a PhD in computer science in 2021, advised by James Landay and Jeff Hancock. An early paper tracked Google results for candidates in the 2018 US elections, and with Hancock and Joon Sung Park they wrote a piece for The Guardian asking whether Google search results were politically biased.

After a postdoctoral year at Stanford's Center for Philanthropy and Civil Society with Nathaniel Persily, Metaxa joined Penn in 2022, where they co-founded the Penn HCI group and advise the Center on Media, Technology, and Democracy.

They have also advised the New York City Commission on Human Rights on a social media platform's AI filters that discriminated against LGBT content, and in 2025 they and Sorelle Friedler received a $200,000 Ford Foundation grant for technical support on civil rights audits of AI.