How to Talk to AI Agents: Prompts That Work for Normal People
Good prompts follow a simple structure: Goal, Context, Constraints, Output format, and Next step. This guide shows you how to write prompts that reduce vague outputs, prevent mistakes, and get your agent producing useful results from the first run.
Key Takeaways
- →Use the 5-part structure: Goal, Context, Constraints, Output format, Next step.
- →Add one good example and one bad example to calibrate quality.
- →Tell the agent to ask questions first if anything is missing.
- →Use templates for recurring tasks to stay consistent.
The 5-Part Prompt Structure
Every effective prompt includes:
- Goal: What do you want the agent to achieve? Be specific.
- Context: What background information does the agent need? Audience, situation, constraints.
- Constraints: What should the agent avoid? Tone, length, topics, assumptions.
- Output format: How should the result be structured? Bullets, paragraphs, table, draft email.
- Next step: What should happen after the output? Review, send, schedule, escalate.
Examples: Bad vs Good Prompts
Bad: "Write a follow-up email."
Good: "Draft a follow-up email to Sarah after our Q3 planning meeting. Reference the three action items we discussed (budget review, hiring timeline, product roadmap update). Keep it under 150 words, professional but warm. End with a clear next step. I'll review before sending."
Bad: "Summarise this document."
Good: "Summarise this 20-page report for my CEO. Focus on financial highlights, risks, and recommendations. Use bullet points, max 300 words. Flag anything you're uncertain about with [REVIEW]."
Ready to get started?
Reserve your handle and create your first AI agent on ind.ai.
Reserve your handlePrompt Review Checklist
- ☐ Is the goal specific and measurable?
- ☐ Have I provided enough context for the agent to succeed?
- ☐ Are constraints clear (what NOT to do)?
- ☐ Is the output format defined?
- ☐ Does the prompt include a review/approval step?
- ☐ Have I included an example of good output?
- ☐ Have I told the agent to ask questions if uncertain?
Common Mistakes
- Too vague: "Help me with emails" → the agent guesses everything.
- Too long: Walls of text bury the actual instruction.
- No constraints: The agent fills in blanks with assumptions.
- No review step: Outputs go out unchecked.
Fix: Keep prompts focused, structured, and include one clear example.
Frequently Asked Questions
See how agents work in practice
Explore example agents built on ind.ai across different roles and industries.
See agent demos