Gains from verified AI materials are concentrated in the bottom quintile, which accounts for roughly three-quarters of the average effect¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Quantile difference-in-differences shows a gain of 11.0 marks at the 10th percentile (95% interval 5.0 to 13.0), 5.0 marks at the 20th, and estimates near zero from the 30th percentile upward. → Dang 2026 q2 i? Quintile decomposition attributes roughly three-quarters of the 2.34-mark average to the bottom quintile, whose difference-in-differences is 8.9 marks. → Dang 2026 (2)

Evidence¶

Dang 2026¶

Dang, C. T., & Nguyen, A. (2026). Verified, not generated: expert-verified AI study materials and the distribution of learning gains in a university course. https://arxiv.org/abs/2610.07097 doi:10.48550/arxiv.2610.07097

q2 · i? · causal · r2

Quantile difference-in-differences at deciles with 95% intervals from 1,000 student-level bootstrap resamples, reported in the distributional results and Fig. 2b. Only the 10th-percentile estimate is bounded away from zero.

"The gain is 11.0 marks at the 10th percentile (95% interval 5.0 to 13.0) and 5.0 marks at the 20th (−1.0 to 10.0), and estimates from the 30th percentile upward are close to zero."

Dang 2026 (2)¶

Dang, C. T., & Nguyen, A. (2026). Verified, not generated: expert-verified AI study materials and the distribution of learning gains in a university course. https://arxiv.org/abs/2610.07097. doi:10.48550/arxiv.2610.07097

q2 · i? · causal · r2

Quintile decomposition of the average effect in the same distributional analysis, supporting hypothesis H2 that gains concentrate in the lower tail and attenuate toward the upper tail.

"Decomposing the average by quintile of each cohort’s distribution attributes roughly three-quarters of the 2.34 marks to the bottom quintile, whose difference-in-differences is 8.9 marks, and almost none to the top three."

Discussion¶

Related Claims¶

• The average effect of verified AI materials attenuates when the lowest pre-intervention marks are removed, while the threshold effect at the upper-second boundary is robust — related
• Access to expert-verified AI study materials is associated with a 2.34-mark difference-in-differences advantage on a 50-mark examination component, reflecting protection from a decline rather than a rise — a broader claim this one bears on
• Chatbot treatment effects on final grades were statistically significant between the 35th and 75th quantiles of the course grade distribution and at the 95th — related
• School-based intelligent environment support shows an inverted U-shaped quantile pattern, stronger at lower-to-middle competence levels — related
• ChatGPT exposure effects vary across the school performance distribution, with positive estimates for lower-performing schools and negative for top-performing schools, but this pattern is not robust — related
• The share of marks below the upper-second classification boundary fell by 24.7 percentage points, with effects significant at every threshold from 23 to 31 marks and at none above — related
• Mark dispersion narrowed in the treated half alone, with the Part 2 standard deviation falling from 7.09 to 5.03 while Part 1's was unchanged — related