The two sample tables above show 15 rows each. They cover students and professionals, about ten countries, and most of the AI tools and purposes (coding, studying, writing, research, customer support, content creation and data analysis). The second sample adds Monthly_Income. I haven't quoted individual incomes here because they're person-identifying. **What the rows look like** - **Grain:** one row per surveyed user, keyed by User_ID. Each row has demographics, the AI tool and purpose, daily usage hours, a free-text task list, monthly AI cost, and outcome scores (productivity, accuracy, satisfaction, time saved, would-recommend). - **Typical row:** a 20–40-year-old using ChatGPT, Copilot, Claude, Gemini, Perplexity or Meta AI. Satisfaction is mostly 4–9, accuracy is mostly 2–4, and productivity is around 5. - **Overall scores** (step 3): - Productivity ranges from 1.6 to 10.0 and averages about 5.23. - Satisfaction averages about 6.15. - Accuracy averages about 3.43. **Data quality issues the sample exposes** - **Duplicate IDs:** there are 300 rows but only 288 distinct User_IDs (step 1), so some users are repeated. - **Missing values:** Productivity_Score is filled in 284 of 300 rows, Accuracy_Rating in 285, and Satisfaction_Score in 293. Blank AI_Tool, AI_Purpose, Profession and Gender cells also show up in the samples. - **Mixed formats in text-typed columns:** - Age appears as "16 yrs". - Usage hours appear as "1.6 hrs". - Cost appears as "$20". - Time-saved and work-hours values sometimes carry "hrs". - **Inconsistent labels:** - Gender appears as Female, F, FEMALE and male. - Country appears as france, india and canada in lowercase. - AI tool appears as ChatGPT, Chat GPT, chatgpt, Microsoft Copilot and copilot. - Education appears as Undergraduate and undergrad. - These variants explain the high distinct counts (12 genders, 17 education levels, 17 tools). - **Implausible or mixed content:** some rows pair a 16-year-old with a Master's degree, or a 17-year-old with a PhD. Some professions and tasks are in German, French or Portuguese. Because of these issues, normalize the label columns and strip the units before any aggregation or comparison.