Pick a depth. Each prompt opens in your AI pre-loaded with the lesson. Click a row to preview the prompt.
Jumping straight to a formula on a dataset you do not understand is how you compute a confident wrong answer. Maybe the 'Amount' column is in cents, not dollars, so your revenue is off by 100x. Maybe 'Status' has values you never noticed - 'refunded', 'cancelled' - that you should exclude. Spending two minutes having AI describe your columns first is the cheapest insurance in analytics: it surfaces the field meanings, likely types, and weird values before they poison your answer. Crucially, you do this WITHOUT pasting your entire sensitive dataset - a small sample is enough for AI to read the shape, and it keeps customer names and private numbers out of a chat log.
The move: copy your header row plus about 10 sample rows - not all 500 - and paste them into ChatGPT or Claude with a framing prompt. Here is the actual prompt to use: 'Here are the header and first 10 rows of a dataset I am about to analyze. For each column, tell me: (1) what the field probably means, (2) its likely data type - date, number, text, category, and (3) anything odd or worth checking before I trust it. Here is the data:' then paste. Suppose you paste a sales sample with columns Order Date, Product, Amount, Status, Region. A good AI reply reads like this: 'Order Date - a date, format looks like MM/DD/YYYY. Product - text, product names. Amount - a number, but values range 4.99 to 89.00 so this is likely dollars, not cents; confirm your currency. Status - a category with values completed, refunded, pending - you will probably want to EXCLUDE refunded and pending when totaling real revenue. Region - a category, 4 values seen: North, South, East, West.' Notice what that just saved you: you now know to filter out refunded orders before summing, something you would have missed. On privacy - if your data has customer emails or names, either delete those columns from your sample before pasting, or replace real values with fake ones (Customer A, Customer B). The AI only needs the structure, not the identities. This is a habit worth keeping: describe before you compute.