Figure 1.1: Usage shares among top 10 tasks over time by platform, Claude.ai and 1P API. Share of conversations assigned to the ten most prevalent O*NET tasks, by platform and report version.
Figure 1.2: Work, personal, and coursework usage on Claude.ai in November 2025 and February 2026. Share of conversations identified as work, personal, or coursework related on Claude.ai.
Figure 1.3: Collaboration mode share, Claude.ai. Collaboration mode frequencies across Anthropic Economic Index Reports in Claude.ai.
Figure 1.4: Shifts in the average task value across version and platform. This plot uses the O*NET framework to estimate the dollar value of tasks performed on Claude.ai and the 1P API. Task value is estimated as the average hourly wage paid to workers who do that task.
Table 1.1: Changes in key primitives. This table shows average primitives in Claude.ai compared to the previous Economic Index report. All differences are statistically significant with p<0.001, except Human-only time with p<0.05. See the Appendix for definitions of these primitives.
Figure 1.5: Geographic convergence. This figure shows Lorenz curves of the Anthropic Usage Index for US states (left panel) and countries (right panel).
Figure 2.1: Model choice and occupational domains. This plot shows how much more or less paid Claude.ai users select the Opus class of models depending on which occupational domain the task is in. In this sample, 51% of overall usage is Opus, and 55% of Computer and Mathematical usage (+4.4pp) is Opus.
Figure 2.2: Model choice and occupation. This plot shows how often users select the Opus class of models depending on which occupation the task is associated with. Each point is an occupation (x-axis) and its Opus share (y-axis). The left panel shows Claude.ai users, the right panel shows 1P API users.
Table 2.1: Differences between high and low tenure users. This table shows average characteristics for high and low tenure users. We define high-tenure users as those who signed up for Claude at least six months before our data pull.
Figure 2.3: How tenure correlates with years of education and personal use. This figure shows two binned scatterplots. The left panel shows human education years vs. days since signup. The right panel shows the percentage of personal use conversations vs. days since signup.
Figure 2.4: The association between experience and success. This plot shows the results from regressing our binary success measure on an indicator for high tenure, with increasingly stringent controls. The coefficient is given in percentage points, and the whiskers give 95 percent confidence intervals. “Task FEs” indicates fixed effects for O*NET task and request cluster. Full controls adds model, use case, and country fixed effects.
