The counterintuitive rule the community learned the hard way: do not ask AI to edit your data directly. Ask it to write a small script that edits the data. A script runs the same way every time and you can read exactly what it will do, while a model editing cells will quietly hallucinate and reassure you that it did not.
So the safe pattern is:
A prompt to start from:
I have a spreadsheet of contacts. Write me a script I can run in Google Sheets that standardizes phone numbers, fixes name capitalization, and flags likely duplicates in a new column. Ask me to confirm before it changes or deletes anything.
A few things AI can clean directly, cell by cell, where each row stands on its own: fill-in passes with the Gemini formula in Google Sheets, image-to-table extraction with a tool like Rows, and light fixes on a small sheet uploaded to ChatGPT. These are real people's details, so lean on the script approach above as your default: your data never leaves your own sheet. When you do upload a sheet to a consumer AI tool, remove anything sensitive first, and never paste in donor or financial information. When a list gets large, work in chunks of about twenty rows and save after each pass, because big jobs make the model drop rows or fill them with placeholders.
More detail is on the community wiki.
Discussions are coming soon. Every skill, app, and guide will carry its own thread where shluchim compare notes on what worked.