Special-education support/adaptation shows the strongest demand–AI-use alignment, with a 16.32-percentage-point increase in predicted use across a two-point relative-demand contrast¶

Claim · All claims
Evidence · 1 study · associational r2 · q2 · i2 medium
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i2 Predicted special-education AI use rises from 34.0% to 50.3% comparing tasks one point below versus one point above the teacher's mean demand (OR = 1.442). → Holster 2026

Evidence¶

Holster 2026¶

Holster, J. (2026). Where Artificial Intelligence Enters Teacher Work. https://www.oecd.org/en/data/datasets/talis-2024-database.html doi:10.35542/osf.io/a39rp_v1

q2 · i2 · associational · r2

Standardized probability contrast from the primary pooled allocation model (Figure 1; Table 2); the task slope was OR = 1.442, 95% CI [1.373, 1.514], p < .001, the largest of the three tasks.

"Special-education support/adaptation AI use increased from 34.0% to 50.3%, a 16.32-point difference (95% CI [14.14, 18.50])."

Discussion¶

Related Claims¶

• Assessment/marking shows an inverse demand–AI-use association in the primary system-average analysis, with predicted use falling 3.47 percentage points across a two-point relative-demand contrast — related
• Overall demand–AI-use alignment varies modestly across education systems, but task-specific association profiles show clear residual heterogeneity — related
• Teachers experiencing higher average demand across lesson planning, assessment/marking, and special-education adaptation report higher odds of AI adoption — related
• The prespecified task–demand mapping produces the strongest within-teacher alignment of all six possible one-to-one assignments, supporting task-specific rather than arbitrary demand alignment — related
• Among AI-using teachers, tasks experienced as relatively more demanding are more likely to be AI-supported overall — a broader claim this one bears on
• The within-teacher demand–AI-use association differs sharply across tasks, with only a modest positive planning gradient — a broader claim this one bears on