**The strongest relationships are all in one cluster: how much people use AI, how much time they say it saves, and how productive and satisfied they feel.** I computed Pearson correlations across the 300 survey rows, after cleaning text artifacts like "2 hrs" and "3.2 hrs" out of the numeric columns. The last cleaned run (step 5) had 277–293 usable pairs per correlation because some values are missing. **Strong relationships (r ≥ 0.65)** - **AI usage hours per day and time saved per week: r ≈ 0.78.** This is the strongest pair in the dataset. - **Productivity score and time saved per week: r ≈ 0.75.** - **AI usage hours per day and productivity score: r ≈ 0.66.** - **Productivity score and satisfaction score: r ≈ 0.65.** **Moderate relationships (r ≈ 0.3–0.55)** - **Accuracy rating and satisfaction: r ≈ 0.53.** Satisfaction tracks perceived accuracy. - **Satisfaction and time saved: r ≈ 0.49.** - **AI usage hours and satisfaction: r ≈ 0.42.** - **Age and work/study hours per day: r ≈ 0.30.** This is the only demographic link of note: older respondents report slightly longer days. **Essentially unrelated (|r| < 0.1)** - **Monthly AI cost** is near zero against productivity (0.01), satisfaction (0.01), time saved (0.05) and usage (0.04). Paying more for AI shows no association with better outcomes here. - **Monthly income** is near zero against productivity, satisfaction and AI cost. - **Accuracy rating** is nearly unrelated to productivity (0.07) and time saved (0.05). It relates to satisfaction but not to measured output. - **Age** has weak negative links to time saved (−0.12) and satisfaction (−0.12). **What this means** - Heavier daily use goes with more reported time saved and higher productivity, and productivity goes with higher satisfaction. Satisfaction appears to follow from perceived productivity and accuracy. - The data is observational self-report, so these correlations do not show that more usage causes the gains. Time saved and productivity are both self-rated, so some of the overlap may be shared response bias. - I only measured correlations between numeric columns. I did not test associations among the categorical columns (AI tool, purpose, user type, education level, would-recommend), which would need grouped comparisons. - Several numeric fields (usage hours, time saved, work/study hours, age, cost, income) are stored as text, some with unit suffixes. They need cleaning before any analysis.