**The most counterintuitive finding is that paying for AI buys nothing measurable.** Monthly AI spend is essentially uncorrelated with every outcome the survey records. **Correlation of Monthly_AI_Cost with each outcome** (steps 2 and 5, n = 288 users with a parseable cost): | Outcome | Correlation with cost | |---|---| | Productivity | 0.013 | | Satisfaction | 0.009 | | Accuracy rating | 0.008 | | Time saved per week | 0.055 | All four are effectively zero. **Free users did slightly better than paid users** (step 3): - **Productivity:** free users averaged 5.41 versus 4.99 for paid users. - **Satisfaction:** free users averaged 6.26 versus 5.98 for paid users. - **Accuracy:** the two groups were identical, at 3.40 versus 3.41. - **Time saved and usage:** both were identical, at about 5.8 hours saved per week and about 3.0 hours of use per day. - **Group sizes:** 145 free users and 128 paid users. A further 27 users have no usable cost. Of the 288 users with a cost, 154 pay nothing (step 2). **Why this is surprising.** Hours of use behaves as you would expect. Daily usage correlates strongly with time saved (0.76) and productivity (0.64), and weakly with satisfaction (0.40) (steps 2 and 5). Productivity and satisfaction also move together (0.65), and accuracy and satisfaction (0.53). So the outcomes respond to how much people use AI, but not to how much they pay for it. Heavier use pays off; a pricier subscription doesn't. **Caveats.** - This is observational survey data, so it shows association, not causation. - It does not account for differences in which tools or purposes paid and free users choose, or for what those tools offer. I did not break the result down by tool. - Many numeric fields are stored as text, and I had to parse them. About 12 users have an unparseable cost and about 29 have unparseable usage hours. For anyone deciding whether a paid subscription is justified, this dataset gives no evidence that it is.