AI-Assisted Negotiation Training: A Quasi-Experimental Examination of Buyers’ and Sellers’ Perceived Negotiation Performance in a Used-Car Sales Simulation

Background: Evidence on artificial intelligence (AI) as a preparatory learning aid for novice human negotiators remains limited in consumer negotiation simulations. Existing work emphasizes automated agents rather than students preparing to negotiate. Against this backdrop, the present study examined changes in students’ perceived negotiation performance, differences in those changes between buyer and seller roles, and the purposes for which students used AI during a simulated used-car sale. The outcome was perceived performance, not economic results. Methodology: A single-group, repeated-measures quasi-experiment involved 74 undergraduate business students enrolled in a negotiation course at a university in Bangkok, Thailand. Thirty-seven students served as buyers and 37 as sellers. Participants completed two negotiations outside regular class hours within a 24–48-hour period, retaining their assigned roles and using the same scenario. The first round was conducted without AI assistance. Before the second round, students spent approximately five minutes preparing with generative AI tools of their choice for information analysis, strategy planning, role-play, or feedback. Each negotiation lasted approximately five to seven minutes. Perceived negotiation performance was assessed using five-item self-report scales in each condition. Paired-samples t-tests and a mixed-design analysis of variance (ANOVA) examined differences between conditions and across roles. Reported purposes of AI use were summarized descriptively. Results: Students reported higher perceived negotiation performance after AI-assisted preparation (M = 4.45, SD = 0.42) than after the preceding non-AI round (M = 3.78, SD = 0.58). The mean difference was 0.67 points, t(73) = −9.23, p < .001, with a large standardized difference (|d| = 1.07). The mixed-design ANOVA identified a significant condition effect, F(1, 72) = 83.97, p < .001, partial η² = .54, but no significant condition-by-role interaction. Buyers' mean scores increased by 0.68 points and sellers' by 0.66 points. Information analysis was the most frequently reported purpose of AI use (91.9%), followed by strategy planning (60.8%), role-play practice (51.4%), and feedback (40.5%). These descriptive patterns do not establish the effectiveness of individual AI functions. Discussions: The findings suggest that AI may provide useful analytical and strategic support during experiential negotiation preparation. Similar patterns of change across buyer and seller roles indicate that the observed increase in perceived performance was not statistically distinguishable between the two groups; the groups do not demonstrate equal bargaining power or equivalent economic outcomes. Students' greater use of information analysis and planning than of interactive practice also points to a possible area for instructional development. However, the fixed order of conditions and repetition of the same scenario mean that practice and familiarity may have contributed to the higher ratings in the second round. Conclusions: AI-assisted preparation was associated with substantially higher self-reported negotiation performance in this educational simulation. The findings reflect students' perceptions, not independently assessed negotiation quality, agreed prices, fairness, or information asymmetry. Accordingly, they should not be interpreted as a causal estimate of AI's contribution to objective negotiation performance. Future studies should counterbalance condition order, include behavioral measures, and recruit more diverse participants.