Sono Shah

Sono Shah

Political scientist · leading data science & machine learning at Theknow.io

I study how AI and digital platforms shape political opinion and behavior — and build the systems that measure it. Ph.D. in Political Science, UC Riverside.

What I study

I'm a political scientist by training who focuses on computational social science. My research pairs novel behavioral data — web browsing, platform activity linked to survey respondents — with representative surveys and experiments, to measure how technology shapes what people see, believe, and do politically.

Most recently that has meant asking what AI does to public opinion research itself. I conceived and co-authored Pew Research Center's study of whether large language models can stand in for human survey-takers. Because the real American Trends Panel answers already existed, we could score AI personas against the actual people they were built from, across nearly 300 questions. The answer was largely no, and the way it failed matters more than the average error: when a model misrepresents some parts of the public more than others, anything built on its synthetic opinion inherits that skew.

Before that, the same approach measured how AI-generated search summaries change what people actually click on, how Democrats and Republicans behave differently on Twitter, and how negative rhetoric from members of Congress earns disproportionate engagement. Earlier work used survey experiments to study how elite rhetoric and information about political representation move public attitudes and participation.

What I'm working on

I lead data science and machine learning at Theknow.io, a social-listening and analytics platform. The work spans the full lifecycle: the statistical methods, the production systems built on them, and what clients ultimately see.

Much of it comes down to questions I worked on as a researcher, now asked in production: when one source of data can reliably stand in for another, how to put language models to work without giving up rigor, and how to keep AI-generated output accurate and accountable to the people who rely on it. Most of the work is internal, but I'm always happy to talk through it.

Before this

I was Associate Director of Data Labs at Pew Research Center — technical and methodological lead for an eight-person team, with four direct reports and 20+ public-facing research projects shipped. I set the research agenda with the Managing Director, owned the team's research infrastructure, and set our standards for applying LLMs to classify text, audio, and video at scale. Coverage of the work has run in Bloomberg, Forbes, and Vox.

Before Pew I spent seven years at AAPI Data, on public opinion research covering Asian American and Latino communities. I hold a Ph.D. in Political Science from the University of California, Riverside.

Across all of it, I combine computational methods with survey research to answer questions about political behavior that neither approach can address alone — and, increasingly, questions about AI itself.

Research

Peer-reviewed work, working papers, and the public reports I authored or directed at Pew.

2026
Investigating Click Behaviors On Google Search Result Pages That Produce an AI Overview
with Athena Chapekis, Anna Lieb, and Aaron Smith
arXiv preprint
2026
A large-scale corpus of religious radio broadcast transcripts from webstream recordings in the United States
with Samuel Bestvater, Athena Chapekis, Skyler Seets, Anna Lieb, and Aaron Smith
arXiv preprint (Data Descriptor)
2026
Can AI Stand In for Human Survey-Takers? Not Really
Co-author; conceived the study and wrote the proposal, then supervised the lead researcher
Pew Research Center

All research & reports →

Get in touch

Always glad to talk about AI and politics, measurement, applied ML, or research that has to hold up in public.