Information Order Effects in Candidate Evaluation: A Preliminary Comparison of Human and AI Decision-Making

This study investigates whether the order in which identical information about a job candidate is presented can influence evaluative judgments. A hypothetical candidate was described using six identical characteristics: three positive and three negative. Human participants were exposed to one of three versions of the candidate profile: positive characteristics presented first (Condition A), negative characteristics presented first (Condition B), or characteristics presented in a mixed order (Condition C). Each participant evaluated the candidate's suitability for hiring, confidence in their decision, and overall candidate quality.A total of 126 responses were collected (n = 42 per condition). Descriptively, the proportion of participants recommending the candidate for hiring was 61.9% (A), 66.7% (B), and 83.3% (C); mean overall ratings were 6.29, 6.43, and 7.48 respectively. Inferential analysis of the raw individual-level data showed a significant omnibus difference in overall candidate ratings across conditions (one-way ANOVA, F = 3.43, p = .036, η² = .053; confirmed by Kruskal–Wallis, H = 6.50, p = .039), but the omnibus difference in hiring proportions did not reach conventional significance (χ² = 5.13, df = 2, p = .077), and confidence ratings showed no meaningful difference across conditions (F = 0.31, p = .731). Pairwise post-hoc comparisons for ratings (A vs. C, B vs. C) were nominally significant but did not survive Bonferroni correction for multiple comparisons, indicating the effect should be treated as suggestive rather than confirmed.A parallel exploratory experiment was conducted using ten AI systems (latest versions available as of August 31, 2026), with each system independently evaluating all three versions once. Most AI systems consistently rejected the candidate regardless of information order; only one model (Claude) changed its hiring decision across conditions. Because each AI model contributed only a single, non-independent observation per condition, no formal inferential test was applied to the AI data. Overall, the findings provide preliminary, partially statistically supported evidence that information order can influence human evaluative judgments — specifically overall ratings more than binary hiring decisions or confidence — while the tested AI systems appeared comparatively more stable, though this comparison remains exploratory given the limited AI sample and lack of repeated trials.