From the book · Unleash Your Business With AI
Chapter 21: What never to paste
The 2 prompts from Chapter 21, ready to copy, with the chapter's mistakes to avoid and three tasks for this week.
Sort your own files in one sitting
I run a [type of business, e.g. "hardware wholesale business"] in India. I want to sort the kinds of files and information my team handles into three groups before we use AI tools. Red: never goes into an AI tool (personal data of customers, staff or suppliers; bank, card and UPI details; passwords and OTPs; ID numbers; confidential contracts; unreleased financials). Amber: may be used only after names, numbers and IDs are removed or replaced with codes. Green: public or non-sensitive, fine to use. Below is a list of the kinds of files and information we handle. These are descriptions only, not the data itself: [list them, one per line, e.g. "monthly sales export from Tally", "WhatsApp order chats", "supplier rate agreement", "product catalogue", "salary sheet"] Task: 1. Put each item under Red, Amber or Green, with one line of reasoning. 2. For each Amber item, list which columns or details must be removed or coded first. 3. Put anything you are unsure about under "Owner to decide". When in doubt, choose the stricter colour. Do not ask me to paste any of the actual files. Do not tell me what the law requires; I will ask my lawyer. Format: one short table per colour.
What you should get back: three short tables and a list of doubtful items for you to settle.
Check: the AI does not know your contracts or your customers. If it puts something green that you would not want a competitor to read, move it to red. Your judgement wins.
That's the first one
Open the rest, free
Enter your email once and every prompt page and guide on this site opens.
- Clean a file before you paste it
- Mistakes to avoid
- Do this week
Clean a file before you paste it
I have already removed personal data from the text below. Check it again for anything I missed. List any remaining item that could identify a person: names, phone numbers, addresses, email IDs, GST or PAN numbers, bank or UPI details, and order IDs linked to a person. For each item found, show only its first two characters followed by asterisks, and the line where it appears. Do not repeat any sensitive item in full. Do not guess who a person might be. If you find nothing, say "Nothing found" and remind me to scan it once myself. Text: [paste the cleaned text]
What you should get back: a short list of line numbers with masked items, or "Nothing found".
Check: fix each item in your file, not in the chat. If the list is long, your first cleaning was not good enough. Delete that chat, clean the file again and start over.
Mistakes to avoid
- Trusting the paid plan to make red data safe. Red means never, on any plan.
- Cleaning the original file. Work on a copy. One wrong delete in the original Tally export costs you a day.
- Forgetting the remarks column. Names and numbers sit in free-text fields, not only in the columns labelled for them.
- Keeping the code list in the chat. The lookup sheet that matches Customer A to a real name stays on your computer.
- Treating the AI's personal-data check as the first check. You clean first. The prompt only catches what you missed.
- Reading a date in a book and calling it compliance. The DPDP timeline has an open conflict. Your lawyer confirms the date.
Do this week
- Print the traffic-light table and put it near every desk that uses AI. Walk your team through it in ten minutes.
- Take one real export (last month's sales), clean a copy using the five steps, and time yourself. Then show your accountant how you did it.
- Send your lawyer the five questions from this chapter and ask for a written reply on the DPDP date that applies to you.