A quieter robot, a more open partner
A randomized study found that people paired with a robot using fewer personal details in small talk disclosed more about themselves overall than people paired with a version that shared more. The difference appeared in the turn-by-turn conversation and in participants' reflections after the task. The work is an arXiv preprint, version 1, dated 28 August 2026, with no journal publication reported in the supplied record.
The experiment paired each participant with a 7-DOF Franka Emika Panda manipulator, a robotic arm equipped with speech recognition and a GPT-4o conversation-generation pipeline. Its physical behavior followed the same fixed, pre-programmed motion sequence for every participant. During a simulated factory quality-control task, the robot sorted PVC pipes by color while the participant inspected and sorted PVC fittings. The low-disclosure version used minimal personal information; the high-disclosure version used additional personal details. Researchers manually checked the transcripts and coded utterances for type and disclosure.
More robot talk did not mean more human talk
The manipulation worked as intended. The high-disclosure robot produced an average of 745.40 words, compared with 610.44 for the low-disclosure robot. Its no-disclosure rate was 0.54, versus 0.83 for low disclosure. Robot-turn counts, user-turn counts and user word counts did not differ significantly.
On the main turn-level measure, users in the high-disclosure condition disclosed less overall than users in the low-disclosure condition. The paper reports a statistically significant condition estimate, but no confidence interval or standardized effect size for it. In other words, the direction of the difference was clear in the study's model, but its size is harder to judge from the reported results.
Prompting mattered within the conversation. Users disclosed more on turns directly prompted by the robot than on unprompted turns, and that association was stronger in the high-disclosure condition. The finding does not show that prompting increased overall conversational engagement; it shows a difference in disclosure on the turns examined.
The survey results were consistent, with exceptions
Post-study surveys pointed in the same broad direction. Low-disclosure users reported higher disclosure scores than high-disclosure users, stronger perceived teaming and higher rapport-coordination scores. Prior robot-teaming experience moderated the disclosure difference, with more experienced participants reporting less disclosure in the high-disclosure condition. Experience also moderated the rapport-coordination pattern.
Connection and boredom were more mixed. There was no overall condition effect on rapport-connection or boredom, but prior experience interacted with condition for both outcomes. Experienced robot-teamers reported lower connection in high disclosure, while greater prior robot experience was associated with more boredom in that condition. These subgroup findings were exploratory, and subgroup estimates and confidence intervals were not reported.
A narrow test with clear limits
The study included 50 participants, with 25 randomly assigned to each condition; the overall mean age was 23.66 years (SD = 8.29). The participant pool was drawn largely from university students and employees. Because the task was simulated, the findings may not carry over to broader populations or real industrial settings.
It also tested one non-humanoid manipulator, one fixed set of disclosure details and one collaborative task. The authors call for research across robot forms, task timing and longitudinal settings. Until then, the results are best read as evidence about this robot, this task and this participant pool, not as a rule for every robot that works with people.
Overall, the result is a caution against assuming that a robot becomes a better teammate simply by sharing more personal details. In this experiment, lower robot disclosure was associated with more human disclosure and stronger ratings of teaming and rapport-coordination, while connection and boredom depended on prior experience. The study was funded by National Science Foundation award #2141335.
Paper data and sources
Original title: From Small Talk to Rapport: Exploring Robot Self-Disclosure in Collaborative Tasks
Authors: Kaitlynn Taylor Pineda, Anvii Mishra, Brian Chien et al.
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-28
DOI: Not available
Original paper · Full text