Even if you've adopted AI and your team uses Copilot, Claude, or another tool, that doesn’t mean what you get out will be perfect. The outputs can feel generic or inaccurate and require too much tweaking to be useful. One problem could be that your prompts aren’t structured well enough to give you the product you need. Knowing how to give AI better prompts means getting better results.
What Is Prompt Engineering?
Prompt engineering sounds technical. It's not. It's the skill of asking AI the right question with the right context to get the right answer.
It’s similar to how someone might give instructions to a new team member. Unclear instructions could be, “Follow up with this client.” What’s better is, "Draft a friendly follow-up email to a contractor who missed our meeting, keep it under 150 words, and use a casual but professional tone." Instructions work best when they’re specific.
Here's an example:
Vague prompt: "Write a customer email."
Precise prompt: "Write a brief follow-up email to a long-time client who hasn't purchased in six months. Acknowledge their past business, offer a 10% discount on their favorite product category, and include a direct link to book a call. Keep it under 100 words and conversational."
Why This Matters and Five Techniques
If workers using AI tools save 30 to 60 minutes per day on routine tasks, that adds up. For a 10-person team, that's around 10 hours daily. But it works best with the highest ROI if the AI output is actually useful. Here's how to make sure it is:
1. Be Specific
Vague requests produce vague results. Specific requests produce the specific output you want.
Instead of prompting, “Write an email to a vendor," try: "Write a professional but friendly email to our office supply vendor asking to renegotiate our contract terms. We've been loyal for three years and want better pricing on bulk orders. Keep it under 400 words and include a request to schedule a call next week."
The AI now knows tone, history, goal, length, and action. It delivers better results.
2. Provide Context
Give AI the bigger picture. Just like a human assistant, this influences the response. Who is the audience? What's the situation? What’s the tone? What do you want the recipient to feel or do?
Example: "I'm an HR manager at a Portland-area tech company. Draft an onboarding email for a new hire who'll work hybrid. We're casual but professional. I want them to feel welcomed and know what to expect on day one. Include a link to our onboarding portal and mention lunch plans with the team."
If your company is AI-forward, you might have a workflow that will handle new employee onboarding, too.
3. Define the Output Format
What do you want? Do you want a list? A paragraph? A script? A table? If you know what you’re expecting, tell the AI.
"Give me five bullet points on why businesses need cybersecurity insurance" is better than "Tell me about cybersecurity insurance." Format constraints save you editing time.
4. Use Examples (Few-Shot Prompting)
Users can also provide a sample of the tone or style you want.
"Here's an example of our company's writing style. Using this tone, draft a blog intro about architectural trends in 2026."
5. Iterate & Refine
The first output is seldom the final result. As you get more used to prompting, the results will come faster and more easily. Treat AI like a new team member. Give the AI feedback and adjust.
Example: "I like the first section, but make it 20% shorter and add a call-to-action at the end."
Iteration can train the AI while giving you practice in how to make your prompts.
Common Mistakes to Avoid
- Being too vague: "Write something about marketing" wastes time.
- Overloading one prompt: Ask for three things at once; get three mediocre answers.
- Not reviewing output: AI is impressive but not infallible. Always fact-check.
- Sharing sensitive data with free tools: Your customer data, employee records, and proprietary processes belong in enterprise tools like Copilot Business, not free ChatGPT.
- Treating first output as final: iteration is going to get you what you want.
- Ignoring tone and audience: "Professional email" means something different to a law firm than a startup.
Getting Started: Three Steps
- Pick one task you repeat weekly: something low-stakes, like a routine email or meeting agenda.
- Write a prompt using the five techniques above. Be specific, include context, and define format.
- Test it two to three times, then scale. Once you nail it, document your best prompts so your team can reuse them and learn together.
That's it. AI is easy to adopt when you start small. Start with one strong use case and build from there.
Key Takeaways
Prompt engineering isn't magic, nor is it complicated. It's a skill you can pick up with practice. The people who spend an hour this week writing clear prompts will gain hours back next week and into the future. The early investment helps product later on. It’s also an investment in your time.
FAQ
Q: How do I protect my business data when using AI?
A: Use enterprise-grade tools. Microsoft 365 Copilot Business keeps your data in your tenant and never uses it to train models. Hatz AI is another option with strong privacy. Avoid free consumer tools for sensitive business information.
Q: What is prompt engineering?
A: Prompt engineering is writing better instructions for an existing AI. For most small businesses, prompt engineering is the answer.
Q: Can I use the same prompt for different AI tools?
A: Yes, a well-written prompt works across ChatGPT, Claude, and Copilot. But each model has strengths. Your Copilot prompts might work slightly better in Copilot because it's integrated with your Microsoft 365 data.