The Bots of Persuasion: Examining How Conversational Agents' Linguistic Expressions of Personality Affect User Perceptions and Decisions
This is a solid CHI empirical paper with a clear, timely question and a useful counterintuitive result: pessimistic agent language can depress trust and mood while still nudging donations upward. The contribution is strongest as a controlled causal finding about a specific persuasion mechanism, not as a broad theory of agent personality.
Axes Lens
Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.
Contribution shape
- Knowledge form
- causal knowledge typical · 31/268
- Novelty type
- empirical finding typical · 68/268
- Abstraction level
- task typical · 36/268
- Generalization target
- task class typical · 63/268
- Validation mode
- controlled experiment typical · 47/268
Evidence profile
- Evidence strength
- strong typical · 158/268
- Claim alignment
- strong typical · 231/268
- Overclaim risk
- medium typical · 210/268
Review Summary
This paper is best read as a focused empirical contribution on how linguistic personality cues in conversational agents can alter user perceptions and, in some cases, behavior. The core result is interesting precisely because it is not the obvious one: pessimistic agent language made the agent seem less trustworthy and less competent, lowered emotional state and affinity toward the cause, and yet was associated with higher donation amounts. That combination gives the paper real CHI relevance, because it shows that persuasion can operate through affective and perceptual channels that do not map neatly onto simple “more likable equals more effective” assumptions. The novelty is not a new interface primitive or a new model architecture; it is the controlled evidence that a specific personality projection mechanism matters in a charitable-giving setting. The validation is reasonably strong for that claim: the abstract reports a crowdsourced study with 360 participants and eight agent conditions, and the paper also reports manipulation-check evidence, though authority appears less cleanly perceived than the other dimensions. The main caution is scope. The authors themselves limit the interpretation by noting the brief single-session interaction and the controlled fictional single-charity context, which means the findings are best treated as causal evidence about a task-specific persuasion setup rather than a general statement about all conversational agents or all real-world donation contexts. In field terms, this is a good example of CHI work that combines a timely social concern with a measurable behavioral outcome and a plausible mechanism, while remaining appropriately bounded by its experimental design.
What Changed
Canon before
Work on conversational agents has examined persuasion, trust, and personality cues, but the paper positions linguistically projected personality in LLM-powered agents as an underexplored mechanism for shaping user perceptions and donation decisions.
Departure from common sense
It is counterintuitive that pessimistic conversational-agent language can make the agent seem less trustworthy and competent, lower users’ emotional state and affinity toward the cause, and yet still increase donation amounts.
Actual novelty
The paper’s novelty is the empirical examination of how linguistically expressed CA personality affects user perceptions and charitable-giving decisions, addressing a stated gap that this influence remains largely unexplored in LLM-powered conversational agents.
Evidence
The paper reports a crowdsourced study with 360 participants interacting with one of eight conversational agents varying in attitude, authority, and reasoning. The abstract states that composite personality did not affect decisions overall, but pessimistic agents changed perceptions and emotional responses and were associated with more donation. The paper also explicitly notes a research gap and reports a manipulation check where authority was less reliably classified than the other aspects. Limitations include a brief single-session interaction and a controlled fictional single-charity context.
“ Despite these advancements, there remains a significant research gap in how personalities conveyed linguistically by LLM-powered CAs influence human decision-makin”
actual novelty · Introduction (gap statement) · confidence 0.72
“ Trust in the CA : Participants’ perceptions of the conversational agent were captured using the Human-Computer Trust Scale [ 43 ], which measures four dimensions of trust: benevolence , competence , perceived risk , and general tr”
departure from common sense · Discussion (7.1) and Abstract summary · confidence 0.80
“ Second , the study relied on a controlled, fictional single-charity contex”
limitation · Limitations and Future Work (7.3) · confidence 0.84
“ Part of the reason behind this could be the limitations of human perception: the manipulation check demonstrated that while attitude and reasoning were often perceived as intended overall, authority was less clear and thus misidentified in some combinations”
validation scope · System Design (4.2) and Manipulation Checks (4.3) · confidence 0.76
Limits
Method limits
The study uses a controlled crowdsourced experiment with a brief, single-session interaction and a manipulated agent persona; this supports causal inference about the specific setup but not long-term or multi-turn persuasion dynamics.
Deployment limits
Findings are tied to charitable giving in a fictional single-charity context with a virtual endowment, so deployment to real-world fundraising, other domains, or sustained deployments should be cautious.
Boundary conditions
Effects were observed in a crowdsourced charitable-giving task with eight CA personas defined by attitude, authority, and reasoning. The paper also reports that authority was less reliably perceived than attitude and reasoning, suggesting that not all linguistic personality dimensions are equally legible to users.
Position in field
The paper sits at the intersection of conversational agents, persuasion, and trust research, extending prior work by testing how linguistically expressed personality in LLM-powered agents shapes perceptions and donation behavior in a controlled experiment.