**The most surprising finding: paying for AI buys nothing measurable.** Monthly AI spend is essentially uncorrelated with every outcome in the survey, and paying users report slightly *lower* productivity than free users. **Spend vs. outcomes (correlations, ~288 users with a usable cost value):** - Cost vs. productivity: r ≈ 0.013 - Cost vs. satisfaction: r ≈ 0.009 - Cost vs. accuracy rating: r ≈ 0.008 - Cost vs. time saved per week: r ≈ 0.055 - Cost vs. income: r ≈ -0.05, so higher earners do not spend more either **Free vs. paid users (step-3 table):** | Group | Users | Productivity | Satisfaction | Accuracy | Time saved (hrs/wk) | Usage (hrs/day) | |---|---|---|---|---|---|---| | Free (cost = 0) | 145 | 5.41 | 6.26 | 3.40 | 5.80 | 3.03 | | Paid (cost > 0) | 128 | 4.99 | 5.98 | 3.41 | 5.82 | 3.04 | Paid users report lower productivity and satisfaction, and no gain in accuracy or time saved. They use AI for the same number of hours. **Why this is counterintuitive:** other relationships in the data behave as you would expect. - Daily usage hours track time saved strongly (r ≈ 0.76) and productivity (r ≈ 0.64). - Productivity and satisfaction move together (r ≈ 0.65). - Accuracy rating and satisfaction are linked (r ≈ 0.53). So **how much** people use AI matters a great deal, but **how much they pay** does not. The decision "is a paid subscription justified?" gets no support from this data. A second, smaller oddity: accuracy rating is almost unrelated to productivity (r ≈ 0.07), even though it tracks satisfaction. People like accurate tools, but accuracy does not show up in their self-reported output. **Caveats:** - This is observational survey data with 300 rows. - 12 rows have no usable cost value and fall outside the cost correlations. The free-vs-paid table also counts 27 users as "Unknown" cost, so cost is missing for more rows than that. - The 0.42-point productivity gap is a raw difference without a significance test. - Productivity and the other scores are self-reported, so measurement noise could mask a real effect of spend. Even so, a flat relationship this close to zero across four different outcomes is a clear signal.