Do You Look Trustworthy? Not to Everyone
Research explores to what degree perceived traits are in the eye of the beholder.
- By
- June 23, 2025
- CBR - Behavioral Science
When it comes to assessing who is trustwothy, there are patterns we tend to follow. People typically see certain facial features and expressions—such as smiles, particularly on women—as more trustworthy.
However, there’s considerable variation between individuals’ perceptions, finds research by Chicago Booth postdoctoral scholar Daniel Albohn, University of Illinois’s Stefan Uddenberg, and Booth’s Alexander Todorov. Instead of relying on group averages, they used machine learning to develop a model that translates individual responses into personalized facial representations. This allows researchers to understand, for example, what type of face comes to mind when a person is asked to imagine a “trustworthy” individual, and how it differs from another person’s ideal.
“When we make complex judgments, our personal characteristics and biases play a bigger role in the decision-making process than the actual thing we’re judging,” says Albohn.
The researchers conducted four experiments, each designed to investigate different aspects of human judgment. In the first three, they examined how people perceive traits such as femininity and masculinity, trustworthiness, attractiveness, and familiarity. Participants rated faces on these dimensions, and their ratings were used to create the AI models that in turn produced photorealistic representations of faces.
When participants rated new faces generated by the model, their assessments largely aligned with those of other participants for more physically evident traits such as femininity and masculinity.
The task at hand matters too
Using AI to visualize how experiment participants saw trustworthiness in different contexts, the researchers find that context matters. Participants tended to agree on what a trustworthy person looked like for a specific task, but that shared decision didn’t apply across tasks. For example, a trustworthy mechanic didn’t look the same as someone trusted to care for a child.
However, there was more variation for complex traits such as trustworthiness, familiarity, and attractiveness. For example, in their data, one person’s model of a trustworthy person was a middle-aged, smiling White woman with big eyes, while another’s was an older, relatively stoic-looking Asian woman. Importantly, because all the faces that participants initially saw and rated were neutral, any emotionality that came through from the models was entirely driven by each individual’s preconceptions about how important facial emotion was for a particular trait.
In the final experiment, which explored how context influences judgments of trustworthiness, participants were asked to discern whom they would trust to babysit their child, fix their car, or manage their money. The researchers created AI-generated images on the basis of these judgments and had a new group of participants assess the images. While people’s ideas of trustworthiness varied depending on the task, their overall patterns of judgment remained consistent. For instance, the person one individual deemed most trustworthy to fix a car was often selected by others for the same task. Importantly, the task itself—whether it was childcare or financial management—was the biggest factor in determining how trustworthy someone appeared. A trustworthy babysitter looks quite different from a trustworthy financial adviser.
This work complements previous research by Albohn, Joel E. Martinez at the market research firm City Square Associates, and Todorov, which finds that people tend to have their own idiosyncratic preferences for a wide range of complex attributes. Moreover, these preferences tend to be unmalleable and cannot be predicted by simple demographics, like race or gender. “Our research highlights the complexity of these preferences,” says Albohn.
The researchers say that their method has promising applications. “Imagine being able to model the type of outfit a specific consumer would likely purchase or design a room that appeals specifically to a certain group of individuals,” Albohn says, speculating about how the technology could be used beyond faces. More broadly, understanding individual differences in perception could potentially improve everything from jury decisions to hiring practices, making us more aware of the subtle biases that guide our choices.
- Daniel Albohn, Joel E. Martinez, and Alexander Todorov, “Determinants of Shared and Idiosyncratic Contributions to Judgments of Faces,” Journal of Experimental Psychology: Human Perception and Performance, November 2024.
- Daniel Albohn, Stefan Uddenberg, and Alexander Todorov, “Individualized Models of Social Judgments and Context-Dependent Representations,” Scientific Reports, February 2025.
More from Chicago Booth Review
More from Chicago Booth
Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.