Prompting

Prompting is how you give instructions to AI agents. Clear instructions deliver consistent results - just like training your best employee, but at scale.

What

Prompting = Giving Instructions to AI Agents

Instead of clicking buttons or filling forms, you write instructions in plain English. The agent reads your instructions and follows them.

Think of it like:

• Telling a new employee how to handle customer emails
• Writing procedures for your team to follow
• Giving directions to complete a specific task

Example:

• Bad: “Handle this email”
• Good: “Read this customer email and respond professionally. Address their question and include next steps.”

Why it matters: Better instructions = better results = competitive advantage. Your prompt quality determines how much time you save and how reliably agents perform.

Purpose

Prompting solves the communication gap between humans and AI:

Consistency

Problem: Different people handle tasks differently Solution: Clear prompts ensure same quality every time Advantage: Deliver enterprise-level consistency regardless of team size

Accuracy

Problem: AI agents guess what you want Solution: Specific instructions eliminate guesswork Advantage: Reduce errors that damage customer trust

Efficiency

Problem: Explaining tasks repeatedly Solution: Write instructions once, use everywhere Advantage: Scale operations without adding training overhead

Basics

Good prompts have 3 parts:

1. Context: What situation is this?

   • “You handle customer service for a consulting firm”
   • “You process invoices for a manufacturing company”

2. Task: What should the agent do?

   • “Read emails and categorize them”
   • “Extract key information from documents”

3. Format: How should the result look?

   • “Respond in a friendly, professional tone”
   • “List information as: Name, Amount, Date”

Simple example:

Context: You handle customer emails for a bookstore
Task: Read this email and determine if it's a question, complaint, or compliment
Format: Answer with just one word: Question, Complaint, or Compliment

Templates

Copy these proven templates. Replace bracketed sections with your details.

Email

Handle customer service for [COMPANY TYPE].

Read emails and respond professionally:
- Address specific question
- Use [friendly/professional] tone
- Include clear next steps
- Escalate to [MANAGER] if [CONDITION]

If you can't resolve, explain what you've done and next steps.

Test Example: Input: “My order is late. Where is it?” Output: “I’ve checked your order status. It shipped yesterday and arrives tomorrow by 5 PM. Here’s tracking: [link]. If not delivered by 6 PM tomorrow, reply and I’ll expedite a replacement.”

Data

Extract from this [DOCUMENT TYPE]:

Field 1: [what to extract]
Field 2: [what to extract]
Amount: $[number]
Date: [date format you want]

Rules:
- Flag if amount over $[LIMIT]
- Mark "NEEDS REVIEW" if any field unclear
- Use "N/A" if information not present

Test Example: Input: Invoice from “ABC Corp for $2,500 consulting work, due March 15” Output: “Company: ABC Corp, Amount: $2,500, Due: March 15, Services: Consulting, Flag: No”

Leads

Qualify this lead for [BUSINESS TYPE]:

HIGH: Has budget + timeline under 6 months + decision authority
MEDIUM: 2 of above criteria
LOW: 1 of above criteria
REJECT: Clear mismatch for our services

Format:
Priority: [HIGH/MEDIUM/LOW/REJECT]
Reason: [1 sentence why]
Action: [what happens next]

Customize

1. Copy template that matches your biggest time-waster
2. Replace [BRACKETS] with your specific details
3. Test with 3 real examples from your business
4. Refine if results aren’t perfect

Fixes

Problem: Agent responses too long Fix: Add “Keep responses under [X] words”

Problem: Agent misses key info Fix: Add specific field to extract: “Always include: [field name]”

Problem: Agent makes wrong decisions Fix: Add clear rule: “If [condition], then [action]”

Problem: Unclear edge cases Fix: Add “If uncertain, mark as ‘NEEDS REVIEW‘“

Test

Pick one template, customize for your business, test with real data from today.

Next

Explore related concepts:

• Tokens - Understand how prompts affect costs
• Models - See how agents produce quality results
• JSON - Advanced prompting for structured data

Master prompting, master agents.