Fig. 2.

An overview of treatment effects. The first estimate (the inverted triangle) represents the effect of labeling AI headlines when all headlines are perfectly labeled (H₁). The next four estimates (the circles) represent the effect of labeling AI headlines when not all headlines are perfectly labeled (H₂). The next three estimates (the triangles) represent the effect of mislabeling human headlines (H₃). The last estimate (the square) represents the effect of labeling false headlines as false (RQ₂).