How Social Influence Shapes Decisions with GenAI

When people make subjective decisions with GenAI in groups, they must evaluate both the GenAI’s recommendation and one another’s judgments. I study how they decide whether a teammate’s agreement or disagreement offers useful information, exerts social influence, or both.

My role Lead researcher and experience designer

I lead the program from question formulation and study design through data collection, analysis, interpretation, and communication.

Research approach Controlled experiments and mixed methods

I combine controlled experiments, behavioral measures, surveys, and qualitative analysis to connect changes in judgment with the reasoning behind them.

Current evidence Established

Engineers expected stronger verification, despite similar GenAI use and perceived value across disciplines.

In progress

Testing how teammate agreement and disagreement change decisions, confidence, and information seeking.

Research Questions

Foundational study

Disciplinary Differences in AI Verification

How do engineers and designers differ in their use, perceptions, and verification of GenAI systems?

Manuscript under review Disciplinary Differences in AI Verification: Engineers vs. Designers 2026

The question this opened

When individual and group decisions produce different outcomes, how does peer pressure change whose judgment guides the group’s response to GenAI?

Main study · Current phase

Social Influence in GenAI-Assisted Decision-Making

How does a teammate’s agreement or disagreement change the subjective decisions people make with GenAI?

Method

Mixed-method survey

A survey of 117 upper-level engineering and design students compared GenAI use, trust, confidence, and verification expectations.

Research framing
Study development
Data collection
Surveys
Qual + quant analysis
Interpretation

Measurement Reliability

Survey measure design

Built the survey using a combination of validated measures and study-specific questions tailored to the constructs and research questions.

Qualitative coding reliability

Two researchers independently coded responses using a structured codebook, resolved discrepancies, and recoded until reaching at least 80% agreement.

Controlled within-subject experiment

Graduate students (50+) first make subjective business decisions with ChatGPT recommendations independently, then revisit them after discussing with a teammate.

Research framing
Study design
Instrument development
10 pilot studies
Protocol refinement
Confederate calibration
Recruitment
Standardized participant training
Controlled experiment
Data collection
Interviews and surveys
Qual + quant analysis
Interpretation

Study Rigor & Measurement Reliability

Iterative piloting & refinement

Conducted 10 pilot sessions before the main study, using each round to refine the experimental flow, materials, timing, instructions, and interaction protocol.

Confederate calibration

Developed and iteratively calibrated a scripted participant (confederate) protocol to standardize behavior, responses, timing, and interpersonal presentation across experimental sessions.

Participant training & standardization

Created and delivered standardized training so participants understood the task, decision process, and experimental procedure before beginning the study.

Instrument development

Combined validated survey measures with study-specific items designed to capture the constructs central to the research questions.

Qualitative coding reliability Planned

Developed a structured codebook with definitions, decision rules, and examples. Multiple researchers will code the qualitative data independently, check agreement, resolve disagreements, refine the codebook when needed, and recode until they reach an acceptable level of agreement.

Findings

Verification expectations differed by discipline

Engineers and designers used and valued GenAI similarly, but engineers expected stronger verification. This distinction showed that individual GenAI use could not explain what happens when people use and evaluate GenAI together.

Peer-pressure effects are being tested

The study measures changes in decisions, confidence, and intention to seek additional information after discussion. Data collection and analysis are ongoing; findings will be added when the evidence is ready.

Research Notes

Dated reflections on consequential changes in the research—not a forced weekly timeline

Method Decision · [Month Year]

[A decision that changed how you study the problem]

[Explain the decision, what prompted it, the alternatives you considered, and how it changed the research. This can be several paragraphs when you are ready.]

study designmethodsdecision-making

Emerging Finding · [Month Year]

[A pattern, tension, or unexpected observation]

[Describe what you observed, why it matters, and how cautiously it should be interpreted at this stage of the project.]

social influenceGenAI trustgroup dynamics

Reflection · [Month Year]

[Something that changed your thinking]

[Share what you previously assumed, what challenged that assumption, and what you are thinking about differently now.]

reflectionresearch process

These are draft placeholders. Notes can be added, removed, or reordered whenever the research evolves.