Research
My research connects the production and selection of public information with how people encounter and judge it. Across health, politics, and journalism, I examine who gains visibility and credibility, and whose information needs are met.
These four questions organize studies of political search, health information, credibility, and AI answers. Throughout, I distinguish information availability, opportunities for exposure, and audience responses.
People
Accusations of “fake news” against political and media opponents were associated with affective polarization in my Mass Communication and Society study. I distinguished these accusations from definitions describing false content, and led the conceptual framing, survey design, and analysis of open responses. The study treats credibility labels as acts of political communication, connecting factual judgments with partisan identity and efforts to discredit opponents.
Collaborative studies extend this question to Russian disinformation, musicians’ political expression, and objections to news commentary. They examine the sources, public figures, and participants shaping political discourse. My manuscript with Tanner Smith, under review at Political Communication, examines framing and nationalization in pink slime outlets presented as local news.
Platform and social curation
Query slant was the primary source of differences in Google results across the three political topics examined in my WebSci 2025 study. Authoritative sources appeared across searches. As sole author, I separated query choices from variation associated with users’ political identities, changing how we interpret apparent search bias. A related UMAP Adjunct paper examines search results during the 2020 Black Lives Matter protests. Collaborative work on #MeToo news coverage extends the selection question to news organizations and differences in attention across the political spectrum.
Accounts acting as brokers offered connections to colorectal cancer racial equity information for audiences with few other expert ties. My Journal of Medical Internet Research study connects this network position with the content available to circulate: practical screening information and symptoms were scarce in the posts we coded. Drawing on the public communication environment framework, I led the conceptual framing, study design, and definitions of equity information disseminators and broker accounts. Brokerage was defined by a majority of followers lacking other expert connections within the studied network, a measure of opportunities for exposure.
Information
My JMIR Medical Informatics study recovered YouTube health videos expressed in colloquial terms and preparation brand names that clinical seed terms missed. I led the design and developed an approach combining word embeddings, recommendation networks, and human relevance judgments. Validation of research retrieval showed improved recall with a limited loss of precision. The contribution is a way to examine how vocabulary choices shape the information researchers observe.
Topics generated by a model broadly corresponded with human coding in our AURA demonstration; additional thematic discovery was limited. My collaborators and I developed the Alignment and Usefulness Rating Audit to evaluate correspondence, usefulness to domain experts, and additional discovery separately. It gives researchers a structured way to assess a topic model’s contribution to qualitative analysis.
I also supervise a longitudinal audit of Google AI Overviews, led by students, across English and Spanish queries. It examines answer availability and the stability of responses and sources; the full paper is under review at ICWSM 2027. A separate study examines uncertainty, guidance, and safeguards in AI health answers. These projects connect system performance with the health information available to readers.
Impact
Maladaptive reactions to cannabis health warnings were associated with greater reactance, less accurate beliefs about cannabis harms, and lower intentions to prevent harm in a collaborative American Journal of Preventive Medicine study. I helped analyze adults’ written reactions using topic modeling, human coding, and statistical analysis, guided by the Extended Parallel Process Model.
Current work examines how people interpret uncertain AI health answers and possible next steps, connecting the information systems provide with audience understanding.
From AI answers to public understanding
My next phase examines how answers generated by AI change the relationship between access to information and understanding it. I am connecting audits of what systems provide with research on how people interpret evidence, uncertainty, and guidance.
Collaborate with me
Current collaborations
- Nursing & IDSI
- With John Robert Bautista and Temitope Eyinla, I developed AURA using interviews about generative AI in nursing. The project connects health expertise, qualitative interpretation, and computational measurement.
- IDSI & Computer Science
- Temitope Eyinla and Krishna Karra contribute to our longitudinal audit of Google AI Overviews. The work brings health informatics and computing into communication research, from study design to data collection and validation.
- Journalism, RJI & Statistics
- Research with Joy Jenkins, Damon Kiesow, and Michael Hatley examines networks of sites presented as local news. This collaboration on pink slime journalism connects questions about political visibility with content analysis at scale.