Training effects concentrated in portrait images (+39.8 percentage points, significant), with directional non-significant negative estimates for full-body and posed-group images¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? The estimated training effect was largest for portrait images, a statistically significant 39.8 percentage point increase in accuracy (95% CI: [27.5, 52.0]). → Negar Kamali 2026 q3 i? Full-body and posed-group images showed directional non-significant estimated effects of -11.6 and -5.2 percentage points, whose confidence intervals include zero. → Negar Kamali 2026 (2)

Evidence¶

Negar Kamali 2026¶

Negar Kamali, Candice Rockell Gerstner, Jessica Hullman, Matthew Groh. (2026). Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images. https://arxiv.org/abs/2606.28510 doi:10.48550/arxiv.2606.28510

q3 · i? · causal · r2

Exploratory pre-registered secondary analysis estimating the BA×Post interaction within each image pose category of the counterbalanced experiment. Portraits had one of the lowest baselines (69.3%), and the article reports the largest gain there.

"The estimated training effect was largest for portrait images, corresponding to a statistically significant 39.8 percentage point increase in accuracy (95% CI: [27.5, 52.0])."

Negar Kamali 2026 (2)¶

Negar Kamali, Candice Rockell Gerstner, Jessica Hullman, Matthew Groh. (2026). Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images. https://arxiv.org/abs/2606.28510

q3 · i? · causal · r2

Same exploratory pose-category analysis. The article treats the negative directional estimates as exploratory: their confidence intervals include zero, so they are not interpreted as evidence that training harmed performance on these categories. Candid images showed a non-significant 8.7-point gain.

"Full-body images and posed group images exhibited directional non-significant estimated effects of -11.6 percentage points (95% CI: [-24.2, 1.0]) and -5.2 percentage points (95% CI: [-18.6, 8.2]), respectively."

Discussion¶

Related Claims¶

• A 30-minute generative AI literacy training increased intelligence analysts' overall accuracy at distinguishing real from AI-generated images by 9 percentage points — related
• The training gain was driven by a 14.2-percentage-point improvement in accuracy on real images, while the gain on AI-generated images was not statistically significant — related
• Training improved discrimination sensitivity (d′) with only a small, non-significant criterion shift toward labeling images as AI-generated — related
• Training gains were distributed broadly across the 97 stimulus images rather than concentrated in a small subset, though stimulus-level effects were heterogeneous — reports the opposite
• Training improved accuracy in every digital-forensics experience group and most generative-AI familiarity groups, with the largest estimated gains among analysts with no forensics experience and frequent generative-AI users — related