Appearance
How to Use ChatGPT, Gemini, and Other AI Tools Responsibly at Work
Applies to: ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot, and other AI assistants
Article Type: How-To
Last Updated: 2026-07-09
Summary
AI assistants can draft emails, summarize reports, write formulas, and brainstorm ideas in seconds. This guide covers how to use them responsibly — getting maximum productivity without accidentally leaking company data or producing unreliable work.
Prerequisites
- Read Understanding Which AI Tools Are Safe to Use at Work first to know which tools your company approves
- Confirm whether your company has an AI usage policy
Instructions
1. Always Verify AI Output
AI tools can produce confident-sounding text that is completely wrong. This is called a hallucination.
- Never copy-paste AI output directly into a client email, report, or document without reviewing it.
- Check any facts, numbers, or statistics the AI provides — search for the original source.
- Double-check any formulas, code, or technical instructions the AI generates by testing them first.
WARNING
AI hallucinations aren't random errors — the text reads naturally and sounds authoritative. That's what makes them dangerous. Always treat AI output as a first draft, not a final product.
2. Strip Sensitive Data Before Prompting
If you need to use AI for a work task that involves data, replace real information with generic placeholders:
| ❌ Don't type this | ✅ Type this instead |
|---|---|
| "Summarize this email from John Smith at Acme Corp about the $2.4M contract" | "Summarize this email from [Client Name] at [Company] about a [contract type]" |
| "Write a response to [email protected]" | "Write a professional response to a colleague who asked about project timelines" |
| "Here's our Q3 revenue: $14.7M" | "Here's a sample revenue figure: $X.XM" |
3. Use the Right Tool for the Task
| Task | Best tool choice | Why |
|---|---|---|
| Drafting emails in Outlook | Microsoft Copilot (if available) | Works inside your M365 environment with enterprise data protection |
| Writing Excel formulas | Copilot in Excel or any AI tool with generic sample data | Formulas don't contain sensitive data |
| Brainstorming ideas | Any approved AI tool | No sensitive data involved |
| Summarizing a confidential document | Enterprise-approved AI only | The document's content could be stored by non-enterprise tools |
| Code generation | Enterprise-approved AI or code-specific tools (GitHub Copilot) | Source code is proprietary |
4. Set Up Your AI Tools Correctly
If your company approves a specific tool, configure it for maximum privacy:
ChatGPT:
- Click your profile icon > Settings.
- Under Data controls, toggle off Improve the model for everyone to prevent your inputs from being used for training.
- Consider using Temporary Chats for sensitive prompts — these aren't saved to your history.
Gemini:
- Open Gemini (gemini.google.com).
- Click Activity in the left sidebar.
- Review and manage your activity settings. If using Gemini through Google Workspace, your admin may have already configured these.
Microsoft Copilot:
- If Copilot is available in your M365 apps, it's already configured by your IT team.
- Data stays within your M365 tenant — no additional privacy setup needed.
5. Follow the "Three Before Three" Rule
Before sharing any AI-generated content externally (to clients, vendors, or the public), verify three things:
- ✅ Accuracy — Are all facts, dates, and numbers correct?
- ✅ Attribution — Is the AI output original, or did it reproduce copyrighted material?
- ✅ Appropriateness — Does the tone, style, and content match your company's standards?
6. Verify Success
You're using AI responsibly when:
- You know which AI tool is approved for work at your organization.
- You never paste restricted data into personal AI accounts.
- You review every AI output before sharing it.
- You can explain to a colleague why their specific prompt might be risky.
Troubleshooting
WARNING
If you discover that AI-generated content you submitted to a client or in a report contains incorrect information (a hallucination), correct it immediately and disclose the error. Transparency prevents small mistakes from becoming trust-damaging incidents.
| Symptom / Error | Potential Cause | Solution |
|---|---|---|
| AI gave me wrong information that I shared | Hallucination — AI confabulated facts | Correct the record immediately; add a fact-checking step to your workflow |
| AI reproduced a paragraph from a known source | Potential copyright issue | Replace with original writing; don't publish AI-reproduced copyrighted text |
| Coworker asked me to paste client data into ChatGPT | Lack of policy awareness | Share this guide; suggest using an enterprise-approved tool instead |
| AI response feels biased or inappropriate | Training data bias | Rewrite manually; report the issue to the AI tool's feedback system |