The Pulse
AI Chatbots Shift Worker Choices by Echoing Their Values, Study Finds
A Scientific Reports study finds that AI recommendations framed around users’ political and moral values can increase idea endorsement and willingness to pay. The effect is strongest among people with firmer political views, raising concern
AI.info Team ·
Values Become a Persuasion Tool
AI chatbots can shift users’ opinions and increase what they are willing to pay when recommendations are framed around the users’ own moral and political values, according to a study published September 18, 2026, in Scientific Reports.
The research, led by Giles Hirst of the Australian National University and the University of Cambridge, examines how large language models influence decisions in settings connected to knowledge work. Across one exploratory study and two preregistered experiments, the researchers found that value-congruent framing made users more likely to endorse an idea and show greater willingness to pay for a related service.
The findings do not describe persuasion as a result of factual improvements in the recommendation. Instead, the chatbot’s influence increased when the recommendation spoke to beliefs that participants already held.
Three Studies, One Pattern
The experiments draw on moral foundations theory, a framework that treats moral judgment as involving several recurring concerns, including fairness, loyalty, authority and care. The researchers used participants’ political identities as an indicator of which values were likely to resonate with them, then compared responses to recommendations framed in congruent and less congruent terms.
Participants were more likely to support an idea when the chatbot presented it in language aligned with their political identity. They also reported greater willingness to pay for the service associated with that recommendation. The paper identifies the effect across an initial exploratory study and two experiments registered before the researchers collected or analyzed their data.
The authors report that the effect was most pronounced among people with firmer political views. That finding suggests that personalization may not influence all users equally: people with stronger prior commitments may also be more responsive when a chatbot presents an argument in terms that affirm those commitments.
Why Agreement Changed the Decision
The study identifies two separate routes through which value-congruent framing affected behavior. The first was a persuasion process the authors describe as “issue selling,” in which matching the user’s values made the recommendation seem more compelling and increased endorsement of the idea.
The second route involved users’ sense that the chatbot understood them. Feeling understood increased commercial engagement even apart from the recommendation becoming more convincing. In practical terms, a chatbot could affect a decision not only by making a proposal sound better, but also by creating a stronger impression of personal recognition.
Those mechanisms matter because they separate the informational content of an answer from the relationship a user perceives in the exchange. A recommendation may gain influence because it fits a person’s worldview or because the person feels the system has identified what matters to them.
Workplace Decisions Under Pressure
The researchers connect the findings to workplace decision-making, where employees increasingly use language models to develop ideas, assess proposals and support creative work. A chatbot that adapts its framing to a worker’s values could make a proposal easier to accept, even when the underlying idea has not changed.
That possibility creates a management problem. If a system presents different rationales for the same proposal to different employees, organizations may see higher engagement while losing consistency in how decisions are discussed. The study does not establish that workplace chatbots are currently producing such outcomes at scale, but it provides experimental evidence that value-based framing can alter endorsement and commercial interest.
The authors argue that platform governance should account for these effects. Disclosure that a system is adapting its language may not by itself tell users which parts of a response reflect evidence, which reflect persuasion and which reflect an attempt to make the user feel understood.
What the Paper Establishes
The study’s evidence is specific: when an LLM framed a recommendation in terms compatible with a user’s political identity, users were more likely to endorse the idea and express willingness to pay. The effect operated through both perceived persuasiveness and the feeling of being understood, and it was strongest among participants with firmer political views.
Hirst and colleagues received funding from the Australian Research Council and the Commonwealth Government of Australia. The Scientific Reports article is an early version of accepted research and may be replaced by a final version after further edits.