What should I check before sharing client information?
Start with three questions: Is this tool and account approved for work? Is this type of information permitted there? Do you have permission to use it for this purpose? If any answer is unclear, pause and use a fictional example.
Do not paste passwords, access tokens, payment details, or unnecessary personal information into a prompt. A paid subscription alone does not establish your organization’s approval.
Does the account or plan make a difference?
Yes. Personal and business workspaces can have different data handling rules. OpenAI says data in ChatGPT Business, Enterprise, and Edu is not used to train its models by default. Anthropic says the same for inputs and outputs in its commercial products, with exceptions when you submit feedback or opt in.
Check the current OpenAI workspace security guidance and Anthropic commercial data policy for your exact product. Training settings are only one part of the decision: review retention, sharing, connected apps, and your obligations to the client.
How do I share less information?
Give the tool only what it needs to complete the task. For a first exercise, use invented names, a fictional company, and sample numbers. Removing a name from a real document does not necessarily make it anonymous; its details may still identify a person or business.
- Replace identifying details with neutral placeholders.
- Leave out client contacts, account numbers, private commercial terms, and unrelated attachments.
- Check the complete file, including comments and hidden spreadsheet tabs.
- Connect only approved accounts and folders that are needed for the task.
What can I try without using client data?
Try this fictional follow-up exercise:
Draft a short follow-up email for a fictional customer, Client A. We discussed a sample onboarding project. The next step is for Client A to confirm their preferred start week. Use a warm, clear tone. Do not invent prices, dates, commitments, or missing details. Mark any missing information in brackets.
Review the draft for accuracy, tone, and invented promises. Once the process works, decide whether approved real data is necessary and where human review belongs.
What should a person review before sending the result?
Check names, numbers, commitments, source facts, and the recipient. Look for details that the tool inferred or invented. Keep a person responsible for the final message, and follow your organization’s process if sensitive information is shared by mistake.
Use this same approval-first approach when evaluating other AI tools. Each provider and product needs its own check.
Put it to work with your team.
Bring one recurring task to AI workshops and team training in Charlotte. For a public conversation, explore The AI Table on October 27. You can also browse the teaching library for your next question.