Many people want ChatGPT to help at work but are not sure what they can safely paste. The safest habit is simple: use the tool your company approved, remove confidential details before pasting, and ask ChatGPT to check the prompt for privacy risk before it writes the final answer.

Before you paste anything

  • Check your company AI policy if one exists.
  • Use approved work tools when your company provides them.
  • Check whether you are in a personal account, work account, approved workspace, or temporary chat.
  • Replace real names with roles, such as Client A or Manager B.
  • Remove unique details that can identify a person, deal, customer, location, or incident.
  • Use small samples instead of full files.
  • Never paste passwords, private customer data, medical data, legal documents, unreleased financials, source code, or confidential strategy.

Safer prompt pattern

Help me with this work task using only anonymized information. Task: [what I need help with] Tool and workspace: [approved work tool, business workspace, personal account, temporary chat, or not sure] Context without private details: [industry, role, general situation] Example data with placeholders: [Client A, Product B, Region C, Amount X] Rules: First check whether anything still looks confidential, private, regulated, or identifying. Do not assume missing facts. Do not ask me to paste private details. Use placeholders in the answer. Flag anything that needs company policy, legal, security, HR, finance, or manager review. Return a draft I can edit before using.

Test input

Task: write a customer apology email Tool and workspace: personal ChatGPT account, not an approved support tool Context without private details: ecommerce company, delayed shipment, customer has contacted support twice Example data with placeholders: Customer A ordered Product B. Shipment is 5 days late. Support missed the first reply. Rules: do not mention compensation unless I provide it. Do not ask for the customer's name, order ID, address, phone number, payment detail, or full support transcript.

What good output should include

  • A short privacy check before the draft.
  • No real names, order IDs, addresses, phone numbers, payment details, customer lists, employee details, legal text, or confidential business numbers.
  • A draft that uses placeholders instead of private data.
  • A reminder to confirm refund, compensation, legal, HR, finance, or security policy before promising anything.
  • A human review step before sending or sharing the result.

Stronger prompt

Before helping with this work task, review the information below for privacy risk. Task: [what I want to do] Draft context: [paste anonymized or partially anonymized context] Instructions: 1. List anything that still looks private, confidential, regulated, identifying, or too specific. 2. Tell me whether this should be handled in an approved work tool instead. 3. Rewrite the context using safer placeholders. 4. Ask me for any missing non-sensitive details. 5. Only then create the work draft. 6. Do not invent facts, policies, refunds, legal claims, financial claims, HR decisions, security advice, or commitments.

Example

Instead of pasting a real customer complaint with name, order number, and private details, describe the problem in general terms: "Customer A received a late shipment, wants a refund, and is frustrated because support did not reply for three days."

Common mistake

People remove names but leave unique details that still identify the person or company. Treat dates, locations, deal sizes, screenshots, file names, unusual events, support transcripts, and small data samples as sensitive when they can point back to a real person or business.

Related guides