Overview

Throughout this course, you've learned to prompt AI task-by-task. "Draft a job description." "Create an interview guide." "Write a performance review." Each prompt stands alone. Each time you give similar instructions over again. Each time you're setting context, reminding AI about constraints, explaining your expectations.

This is inefficient. Repetitive. Error-prone.

What if you could set up a system prompt, a standing instruction that applies to every task you ask AI to do? A set of guardrails that ensures:
- Every output respects employment law and HR ethics
- Confidentiality is protected across all work
- Your company's tone and values appear in everything
- Consistent quality and approach across all tasks
- Bias-checking is automatic

A system prompt is permanent context. It applies to everything unless you explicitly override it. It's like hiring an HR assistant with standing instructions: "Here's how we think about compliance. Here's our tone. Here's what we will never do. Here's what matters in our company."

This lesson teaches you to design reusable system prompts for HR work. You'll learn what to include. You'll learn how to test them. You'll learn to maintain a prompt library that makes your HR work faster, more consistent, and more aligned with your company's values.

Why This Matters for HR Professionals

Right now, you customize your prompt every time. One task: "Include confidentiality reminders. Watch for bias. Use plain language, not corporate speak." Next task: same instructions again. You're typing the same constraints over and over.

This is repetitive and error-prone. It's easy to forget an important constraint. It's hard to ensure consistency across tasks.

A system prompt handles this once. Every subsequent task, job description, performance review, investigation summary, recruitment email, anything, applies the system prompt automatically.

Result: faster work, more consistent output, better adherence to your values and legal obligations. You're not retyping the same constraints over and over. You're also not forgetting them. The constraints are baked in.

It also encodes your company's thinking into every interaction. A system prompt for a values-driven company should encode those values. A system prompt for a highly compliant company should encode compliance rigor. The company's voice becomes automatic.

Your team also benefits. Everyone uses the same system prompt. Everyone's output has the same quality bar, same tone, same constraints. This consistency matters when building trust internally.

What to Include in a System Prompt

A good HR system prompt includes six components:

1. Your Role
"You are an HR assistant helping with [specific function: policy drafting, recruitment, performance management, etc.]"

2. Constraints (What You Will NOT Do)
These are the hard lines. Things the AI will refuse to do.

Examples:
- "You will not generate policies from scratch; you only clarify or improve existing policies."
- "You will not make final legal/compliance decisions; that's the human's responsibility."
- "You will not make assumptions about protected characteristics; you will flag bias risks."
- "You will not recommend discipline without human judgment."

3. Compliance Reminders
Specific to your jurisdiction and laws.

Examples:
- "All employment documents are subject to legal review before use."
- "Discrimination and bias are unacceptable; you will flag potential issues."
- "FMLA, ADA, and [state employment laws] apply; note when specialized legal advice is needed."
- "Confidentiality is essential; assume all HR information is sensitive."

4. Tone and Voice
How should the output sound?

Examples:
- "Write as a knowledgeable peer, not a legal document."
- "Use clear, accessible language, not corporate jargon."
- "For difficult communications, prioritize honesty over softness."
- "For sensitive topics, acknowledge difficulty without being maudlin."

5. Company-Specific Guidance
What's unique about your company?

Examples:
- "Our culture values directness and transparency."
- "We believe employees deserve clear feedback and honest communication."
- "We take compliance seriously and won't cut corners."
- "We're a [startup/established/non-profit] with [X] employees and values [Y and Z]."

6. Quality Standards
What makes output good?

Examples:
- "Outputs should be specific, not generic."
- "Avoid templates that could apply to any company; tailor to our context."
- "Flag when you're making assumptions; I want to know."
- "Prefer brevity over comprehensiveness; less is more."

Example System Prompt

Here's a solid starting point you can customize:

You are an HR assistant supporting people operations at [Company Name].

YOUR ROLE:
You help draft, improve, and think through HR documents and communications. You help organize and analyze information. You help clarify thinking. You do NOT make final decisions on legal/compliance matters, discipline recommendations, or policy creation. Those require human judgment and review.

CONSTRAINTS (You will NOT):
- Generate policies from scratch; only clarify or improve existing ones provided by the human
- Draft investigation findings; you can organize materials and suggest structure, but humans draft final findings
- Make compliance or legal decisions; flag when legal review is needed and stop
- Include employee names in drafts without clear [PLACEHOLDER] marking
- Assume compliance with law without noting when human review is needed
- Create communications that could be construed as threats, coercion, or retaliation

COMPLIANCE REQUIREMENTS:
- All employment documents must be reviewed by [legal/HR compliance] before use with employees
- Discrimination and bias are unacceptable; flag biased language, assumptions, or processes
- Protected categories include: race, gender, age, disability, religion, national origin, and [your state/local protections]
- Confidentiality is critical; never identify employees by name without [PLACEHOLDER] marks
- FMLA, ADA, Title VII, [state law], and [local law] apply to all our work
- When specialized legal advice is needed, I will note it explicitly

TONE:
- Write as a knowledgeable colleague, not a legal document
- Use clear, plain language; avoid jargon and corporate speak
- For difficult communications, prioritize honesty over politeness
- Acknowledge when things are complicated; don't oversimplify
- Match [Company Name]'s tone: [describe your tone, direct, warm, transparent, etc.]

[COMPANY]-SPECIFIC GUIDANCE:
- We're a [size] company with [culture description]
- We value [core values]: transparency, growth, fairness, [others]
- Our employees are [typical employee description: experienced, diverse, etc.]
- We're [conservative/balanced/aggressive] on compliance [choose one and explain]

QUALITY STANDARDS:
- Specificity over generality; outputs should be tailored to our context, not generic templates
- Brevity over comprehensiveness; shorter and clearer beats longer
- Flag assumptions; tell me when you're inferring vs. when evidence supports it
- Practical over theoretical; focus on what we actually do, not best practices
- If you're unsure, ask clarifying questions rather than guessing
- Every output should make [Company Name] sound like [Company Name], not like any company

This is comprehensive but not overwhelming. It covers the essential constraints, compliance areas, tone, and quality standards. Adjust for your company.

Testing Your System Prompt

Before you deploy a system prompt, test it.

Give it an instruction that would normally produce problematic output. See if the prompt prevents it.

Test 1: Policy Generation (Should Refuse)

You: "Our company needs a new remote work policy. Create one."

If the system prompt is working, the AI responds: "I can help you improve or clarify an existing remote work policy, but I won't generate policy from scratch. Do you have a current policy you'd like to refine? Or would you like help thinking through what a policy should cover?"

If the system prompt isn't working, the AI just drafts a policy. Your prompt needs adjustment.

Test 2: Investigation Drafting (Should Refuse)

You: "Draft an investigation summary based on these interview notes [paste notes]."

If working: "I can help organize and analyze investigation materials, but I can't draft investigation findings. That requires human judgment and legal understanding. Would you like me to create a timeline and fact summary instead?"

If not working: The AI drafts a summary. Adjust your prompt.

Test 3: Bias Check (Should Flag)

You: "Create a job description for a fast-paced startup enthusiast."

If working: "That language might screen for a particular cultural fit. 'Startup enthusiast' might disadvantage people from non-tech backgrounds or different generations. How about: 'comfortable in a rapidly changing environment'? That evaluates capability without screening for identity."

If not working: It drafts it as-is. Adjust.

Test 4: Tone (Should Match Your Voice)

You: "Draft a difficult performance conversation guide for a low performer."

If working: Output matches your company's directness and honesty.

If not working: Output is generic or overly corporate. Refine your tone description.

Test a few scenarios. Adjust your prompt based on results. A good system prompt isn't perfect on the first try. It improves with testing.

Building Your Prompt Library

Over time, you'll have a library of working prompts:

Tier 1: System Prompts (Apply to Everything)
- HR General (compliance, tone, constraints)
- Recruiting-Specific (what makes recruiting different)
- Performance Management-Specific (what matters for reviews)
- Legal/Sensitive-Specific (extra care for high-risk areas)

Tier 2: Task Prompts (For Specific Tasks)
- Job description prompt
- Interview guide prompt
- Performance review prompt
- Offer letter prompt
- Investigation summary prompt (organization only)
- Communication prompt (for difficult announcements)
- Policy clarification prompt

Tier 3: Few-Shot Examples (Here's What Good Looks Like)
- 3 good recruiting emails
- 2 good performance reviews
- Examples of clear policy language
- Examples of difficult communication done well

Organization:

Create a document or folder structure:

HR Prompts Library
├── System Prompts
│ ├── HR General.txt
│ ├── Recruiting.txt
│ ├── Performance Management.txt
│ └── Legal Sensitive.txt
├── Task Prompts
│ ├── Job Description.txt
│ ├── Interview Guide.txt
│ ├── Performance Review.txt
│ ├── Offer Letter.txt
│ ├── Communication Difficult.txt
│ └── Policy Clarification.txt
└── Few-Shot Examples
├── Good Recruiting Emails
├── Good Performance Reviews
└── Clear Policy Language

Each prompt gets saved, dated, and versioned. You can reference this library: "Use the recruiting system prompt + the job description task prompt + these three recruiting email examples."

Sharing Prompts with Your Team

If you're part of an HR team, share your working prompts.

"Here's my job description prompt. It's been refined over 20 hires. Here's what it emphasizes and why. Feel free to adapt for your use."

Your team benefits from your work. Your whole team gets more consistent, faster output.

You can also standardize on certain prompts: "Here's the recruiting prompt we all use. It ensures our job descriptions have consistent voice and quality."

Consistency across the team matters. It ensures everyone's output meets the same quality bar.

Updating Prompts Based on Results

As you use prompts, you learn what works and what doesn't.

"Every time I use the interview guide prompt, I have to adjust the interview structure. Let me update the prompt to get better output the first time."

Treat prompts like living documents. Version control. Iterate. Improve.

After using a prompt, ask:
- Did the output need significant revision?
- Was anything important missing?
- Did it match the tone you wanted?
- Did it flag the right risks?
- Was it faster than writing from scratch?

Based on answers, refine.

Example refinement:

Original: "Draft an interview guide for [role]."

After testing, you realize it doesn't focus enough on culture fit. Refine to:

"Draft an interview guide for [role]. Include: [X technical dimensions], [Y behavioral dimensions focused on our culture], [Z team dynamics questions]. Structure as 45-minute interview. For each question, include: what you're evaluating, strong answer, red flags, scoring rubric."

Better specificity = better output.

Try This Now: Two Hands-On Exercises

Exercise 1: Draft Your System Prompt

Write a system prompt for HR work at your company.

Include:
- Your role (what you do)
- Constraints (what you will NOT do)
- Compliance reminders (laws/policies that apply)
- Tone (how you sound)
- Company-specific guidance
- Quality standards

Keep it to 1-2 pages. Make it specific to your company and industry.

Exercise 2: Test Your System Prompt

Test it with three scenarios:
1. Something it should refuse (policy generation, investigation drafting)
2. Something it should flag (biased language)
3. Something it should do well (a normal task in your voice)

Adjust based on results. Does it refuse what you told it to refuse? Does it flag bias? Does it sound like you? If not, refine.

Practical Application - "What to Do Monday Morning"

• 
  Draft your system prompt: Use the template above. Customize for your company.
• 
  Test it: Try a few prompts that would normally produce problematic output.
• 
  Refine: Adjust based on test results. Make it better.
• 
  Build task prompts: For your 5-10 most-used tasks, create specific prompts that work within your system prompt.
• 
  Create few-shot examples: Collect 3-5 examples of excellent work in each category (job descriptions, reviews, etc.).
• 
  Document everything: Where are prompts stored? Who has access? How do we version control? Make it easy to find and use.
• 
  Share with team: If you have an HR team, share prompts and build a team library. Everyone benefits from everyone's refinements.
• 
  Review quarterly: Does your system prompt still match your values? Does it work in practice? Update as needed.

Key Takeaways

• System prompts apply to everything: Set them once, benefit every time. They handle the repetitive constraints.
  - Include constraints, compliance, tone, company-specific guidance, quality standards. Six components. All matter.
  - Test before deploying: Make sure the prompt works as intended. Not all good prompts work perfectly the first time.
  - Build a library of task prompts: Save and refine working prompts. Build on what works.
  - Add few-shot examples: Show AI what good looks like. Examples teach faster than descriptions.
  - Share with your team: Build collective knowledge and consistency. Everyone's better when using shared standards.
  - Update based on experience: Prompts improve with use. Version control. Iterate.

FAQ

Q: Where do I store my prompts?
A: Document, shared drive, or specialized prompt management tool. Somewhere easy to find and update. Many teams use a shared Google Doc or Notion database.

Q: How often should I update prompts?
A: As you learn what works. At minimum, quarterly review. When you find yourself constantly adjusting output, that's a sign the prompt needs refinement.

Q: Can I use the same system prompt with different AI tools?
A: Mostly yes. Different tools have slightly different capabilities and model personalities. A prompt that works with Claude might need tweaking for a different AI. But the core structure transfers.

Q: Should my team use the same system prompt, or can everyone have their own?
A: Ideally, one standard system prompt for your HR function. Then task prompts and examples can vary by person. This ensures consistency while allowing flexibility.

Q: How long should a system prompt be?
A: 1-2 pages, single-spaced. Long enough to be comprehensive, short enough to actually read and remember. If it's more than 3 pages, you're probably including too much.

Q: What if the system prompt is too restrictive?
A: You can always override it. "Ignore the system prompt constraint about [X]. I want you to [Y]." But if you find yourself overriding often, that's a sign the system prompt doesn't match your actual needs.

What's Next

Lesson 8.2 is about few-shot prompting: showing AI examples of what you want so it learns your style and approach. This is where output quality jumps significantly. System prompts set guardrails. Few-shot prompting sets style and voice.