Qiyao Peng彭琪瑶

Research

I study persuasion and the design of persuasive messages, campaigns, and tools, with empirical work anchored in health communication. My approach is computational at every level: my current agenda is organized around the role of generative AI in persuasive communication, along three axes — AI as an object of study, as a methodological tool, and as an intervention. At the core of my methodological program is multimodal computational analysis of communication content — combining LLM annotation, natural language processing, computer vision, and interpretable machine learning with experimental design and neurophysiological measures (fMRI, eye-tracking).

Programs of Work

1 · AI as a research object — studying AI-generated media

With collaborators at UC Davis and Northwestern, I examine the visual realism, misinformation potential, and surrealist signatures of photorealistic AI-generated images across platforms — how convincing they are to audiences, how their content circulates online, and what risks they pose to the broader information environment. Relevant outputs: Proceedings of CHI Extended Abstracts, 2025.

2 · Multimodal computational analysis for communication research

I specialize in multimodal computational analysis of communication content — combining LLM annotation, natural language processing, computer vision, and interpretable machine learning to ask new questions about how persuasive messages work and where they break down.

My dissertation, Multimodal Emotional Appeals in Anti-Vaping Videos: A Computational Approach to Fear–Hope Flow and Message Effectiveness, applies multimodal computational analysis across visual, auditory, and linguistic channels — with LLM-assisted annotation at the core — to track how fear and hope move across an anti-vaping clip frame by frame and to test whether this emotional flow predicts psychological reactance and behavioral intentions in young adults who vape. In a parallel project with Dr. Jiaying Liu, I use interpretable machine learning to identify the cognitive, social, and emotional “recipes” that make anti-vaping messages effective (manuscript ready, targeted for American Journal of Preventive Medicine). Another line uses LLM annotation at scale to detect systematic “surreal” signatures of AI-generated images (Computational Communication Research, 2026).

3 · Generative AI as a persuasive communication intervention

With Dr. Jiaying Liu, I run experiments testing whether generative AI can tailor anti-vaping messages across socioeconomic and value-based subgroups — and whether scale-first AI personalization comes at the cost of equitable persuasion. This project received the 2026 AEJMC Mass Communication and Society Division Student Research Award. I am also developing an ongoing anti-vaping AI chatbot intervention with the CHARM Lab. This line of work builds on my earlier study of how audiences perceive AI-generated content (Hong, Peng, & Williams, New Media & Society, 2020) — a foundational question for whether AI-driven communication interventions land with the people they aim to reach.

Related work in persuasion and digital media

Beyond the AI agenda, I also work on news, norms, and emotional engagement in digital media environments — including how news headlines, user comments, and mixed emotional appeals shape norm perceptions and engagement (see Publications below).

Peer-Reviewed Publications

  • 2026
    Liu, X., Lu, Y., Peng, Q., Qian, S., Peng, Y., & Shen, C. Seeing the surreal: Mapping surrealism in photorealistic AI-generated images using large language models. Computational Communication Research, 8(2), 1–48.
  • 2025
    Peng, Q., Lu, Y., Peng, Y., Qian, S., Liu, X., & Shen, C. Crafting synthetic realities: Examining visual realism and misinformation potential of photorealistic AI-generated images. CHI Extended Abstracts (CHI EA '25), 1–12. ACM.
  • 2020
    Hong, J., Peng, Q., & Williams, D. Are you ready for artificial Mozart and Skrillex? An experiment testing expectancy violation theory and AI music. New Media & Society, 23(7), 1920–1935.

Registered Report Accepted in Principle

  • 2026
    Rathje, S., Asimovic, N., …, Peng, Q., …, & Van Bavel, J. J. Testing the causal impact of social media reduction around the globe. Nature. Stage 1 RR Accepted

Manuscripts Under Review

  • 2026
    Worsdale, A., …, Peng, Q., …, & Liu, J. Depressive symptoms alter the predictive value of neural responses to vaping prevention messages in young adults who vape. Submitted to Nicotine & Tobacco Research.
  • 2026
    Peng, Q., Duong, H., Shi, R., & Liu, J. Contradiction disrupts and reinforcement plateaus: The interplay of news and user comments on norm perceptions in the digital age. Submitted to Mass Communication and Society.
  • 2026
    Ismail, I.*, Peng, Q.*, Liu, J., & Oh, V. Y. Specific mixed emotional headlines drive online media engagement over and above positivity and negativity. Submitted to Human Communication Research. *Equal contribution.
  • 2026
    Gonzales, A. L., Wang, L. H., Kim, Y. W., & Peng, Q. Meso-level theorizing the digital divide: A model of institutional capacity for digital equity. Submitted to Communication Research.

Manuscripts in Preparation

  • 2026
    Liu, J., Peng, Q., Malik, M., Wang, Y., Norton, E., Markey, C., Ye, T., & Sweet, L. H. Identifying optimal cognitive, social, and emotional profiles of anti-vaping messages for young adult vapers: Insights from interpretable machine learning analysis. Manuscript ready; targeted for American Journal of Preventive Medicine.
  • 2026
    Ye, T., …, Peng, Q., …, & Liu, J. Occipitoparietal response to vape packaging with food-cues indexes visuospatial salience and predicts future vaping frequency in young adults. Manuscript ready.
  • 2026
    Peng, Q. Mapping the discourse network of cervical cancer screening on Reddit: A BERTopic and ANTMN approach. Manuscript ready.
  • 2026
    Peng, Q. What do we measure when we measure emotion? A four-level framework for communication research. Manuscript ready; targeted for Annals of the International Communication Association.
  • 2026
    Peng, Q., & Liu, J. Generative AI for scalable message tailoring: Differential persuasive effects across socioeconomic and value-based subgroups. Manuscript in preparation.
  • 2026
    Zhao, S., Yu, H., Wang, Y., Peng, Q., Ye, T., Sweet, L., & Liu, J. A neural signature of vaping and smoking cues. Manuscript in preparation; targeted for PNAS.
  • 2026
    Peng, Q., …, & Liu, J. Looking without thinking, feeling without looking: Discrete emotions drive visual attention and persuasion resistance through independent pathways in anti-smoking messages. Manuscript in preparation; targeted for Media Psychology.
  • 2026
    Peng, Q., & Lu, Y. When modalities disagree: A computational framework for measuring cross-modal emotion divergence. Manuscript in preparation; targeted for Communication Methods and Measures.
  • 2026
    Gonzales, A. L., Kim, Y. W., Wang, L. H., & Peng, Q. The future of digital equity. Manuscript in preparation; targeted for New Media & Society.
  • 2026
    Peng, Q., & Nabi, R. Empowering young adults against early-onset cancer: Balancing fear, hope, and psychological reactance in health messaging. Data collection in progress.

Selected Conference Presentations

  • 2026Jul
    Two presentations at the Summer 2026 NIH Tobacco Regulatory Science (TRS) Meeting, Washington, D.C. — poster on generative AI for anti-vaping message tailoring across socioeconomic and value-based subgroups, and oral presentation on identifying optimal cognitive, social, and emotional profiles of anti-vaping messages via interpretable machine learning.
  • 2026Jun
    Four papers presented at the 76th ICA Annual Conference, Cape Town, South Africa — on mixed-emotional headlines, anti-vaping ML profiling, mapping surrealism in AI images, and the third-person effect in social-appeal messages.
  • 2026May
    Generative AI for scalable message tailoring. Comm Horizons, UC Davis.
  • 2026Mar
    Optimal profiles of anti-vaping messages via interpretable ML. Annual Meeting of SRNT, Baltimore.
  • 2024Nov
    Congruent and incongruent norms: The impact of news and user comments on norm perceptions in the digital age. NCA (Health Communication Division), New Orleans.
  • 2024Sep
    Misinformation potential of AI-generated images (poster). Trust and Safety Conference, Stanford University.
  • 2024Jun
    A first analysis of the misinformation potential of AI-generated images. 74th ICA Annual Conference, Gold Coast, Australia.
  • 2019Jun
    Are you ready for artificial Mozart and Skrillex? ICA (Communication and Technology Division), Washington, D.C.

For the complete list, see my CV.