What the evidence says
The available evidence comes from one large-sample study of a recruiting platform's candidate-shortlisting data (58,765 jobs over 27 months), comparing gender fairness across AI-only, human-only, and hybrid human+AI selection, measured with a CDP score. On its own, AI-driven candidate recommendation produces lower gender fairness than human recruiters' manual selection: AI Recommended Top-K CDP is 0.699 versus 0.813 for Human Contacted, a 0.114-point gap (p<0.001), with high confidence on both direction and magnitude given the sample size and significance testing.
Combining human and AI recruiting increases fairness beyond either channel alone. Human+AI Contacted CDP reaches 0.854, a 0.041-point increase over human-only recruiting (0.813) and a 0.155-point increase over AI-only recruiting (0.699), with statistically significant improvements across the 27-month series (p<.05 or lower) and a consistent pattern across candidate subsets (viewed, clicked, contacted). The claims describe this as more than an additive effect: the combined score exceeds what summing the separate human and AI contributions would predict, which the source calls a synergistic interaction. The mechanism for that synergy is not established — the source states directly that "we were not able to qualify exactly how this interaction takes place."
Human review of AI output increases fairness relative to AI recommendations alone, but the reported magnitude of that increase varies by which comparison is used. One claim reports candidates found through manual search after viewing AI recommendations reach a CDP of 0.876, compared to 0.791 for candidates contacted from the AI recommendation list (ΔCDP = 0.085, p<0.001) and 0.813 for candidates contacted through manual search alone (ΔCDP = 0.063, p<0.01). A separate claim reports AI-oversight CDP in the 0.791–0.793 range against an AI-only Recommended baseline of 0.642, an improvement of roughly 0.15 points that still falls short of the human-only 0.813 figure. Both describe the same qualitative pattern — human oversight of AI recommendations increases fairness but does not eliminate the shortfall relative to human-only recruiting — at medium confidence on magnitude, since the source notes the interaction mechanism was not directly tested.
The evidence is limited to a single outcome measure (CDP, a candidate-selection fairness score) and a single protected characteristic (gender). No claim here addresses whether these shortlisting-fairness gaps translate into differences in actual hiring or offer rates, and no claim addresses bias by race, age, disability, or other protected characteristics. All findings originate from one study and one platform's dataset, so there is no independent corroboration to check them against.
Where the evidence splits
The claims report two different CDP values for the AI-only "Recommended" baseline: 0.699 (used in the AI-vs-human comparison and the hybrid-synergy comparison) and 0.642 (used in the human-oversight-mitigation comparison). Both are attributed to the same source and metric, but the claims do not explain the discrepancy — it is unexplained by the sources on this page.
Separately, one claim's headline effect size (Post-AI oversight Contacted CDP 0.876 vs AI-only Recommended Top-K CDP 0.699, a 0.177-point difference) does not match the comparison actually stated in its own verbatim quote, which instead reports 0.876 against 0.791 (ΔCDP = 0.085, p<0.001) and against 0.813 (ΔCDP = 0.063, p<0.01) — figures that do not involve 0.699 at all. The direction of the finding (manual search after AI exposure outperforms AI-only or human-only alone) holds under either version, but the two stated magnitudes for the same comparison are inconsistent, and no claim here explains why.