It reads your file before you ask a thing.

An assistant that starts guessing before it looks at your data is answering about a file it imagined. MySheetAI looks first. The second your orders file loads, it reads all 1,240 rows and sorts out each column: Order ID is a label, Order Date is a date from July to September, Category has six values, Revenue is a number from $9.99 to $6,480.

That first read is why the rest holds up. Knowing Revenue is the number and Category is the breakdown is what lets it answer a revenue-by-category question correctly, suggest questions your file can actually answer, and skip the ones it cannot.

It reads your rows right in your browser to do this. Only a small summary of the numbers ever leaves your device, never the file itself. So starting from what is real never means giving up what is private.