Does AI-assisted labor change how earnings are spent?

Generative AI is rapidly substituting for human cognitive effort in everyday labor. This substitution attenuates the cues that make income feel earned: subjective effort and felt ownership of the work product. Mental-accounting theory suggests that earnings stripped of these cues should be treated more like windfalls than like earned income. In a preregistered between-subjects experiment, participants completed a real-effort copywriting task under Manual, Augmented, or Substituted AI assistance and then made an incentive-compatible investment decision over their earnings. The manipulation produced large monotone shifts in every perceptual measure (subjective effort, psychological ownership, perceived self-contribution, and AI attribution), including the feeling that the reward was earned. None of the preregistered hypotheses was supported: investment allocation and hedonic-utilitarian consumption choice showed no significant condition differences (nor did self-rated risk preference, a secondary outcome), and the preregistered bootstrap indirect effects through effort and ownership all included zero. The behavioral point estimates were directionally consistent with the predictions but far too imprecise to adjudicate them, falling well below the study's minimum detectable effect, so the data bound medium-to-large effects rather than establish absence. A categorical mental-accounting classification of the earnings did shift toward "windfall," but its anchors partly restate the effort manipulation, so this shift should not be read as an independent behavioral finding. We interpret the results as strong evidence that AI assistance changes the felt effort, authorship, and earnedness of income, and as quantitative bounds on, rather than proof against, the downstream spending consequences that mental-accounting theory predicts.