Core Concepts

Prompts

Best practices for writing effective AI prompts.

Five Elements

ElementPurpose
RoleSet an expert identity
InstructionsList specific actions to perform
ContextProvide background or references
InputData or content to process
OutputShow format examples or templates

Core Principles

  1. Put key instructions first — Models weight the beginning more heavily.
  2. Show, don't tell — One example beats a paragraph of explanation.
  3. Be specific — "Under 75 words" not "keep it short".
  4. Use simple verbs — "Summarize in 3 points" is clearer than "analyze".

Quick Wins

  • Add a role → More professional tone
  • Give an output example → Better formatting
  • Use numbered steps → Less fluff
  • End with "No extra explanation" → Skip the pleasantries

Common Mistakes

ProblemFix
Key requirements buriedMove to top, use a list
Vague verbs like "analyze"Be specific: "one bullet per metric"
Incomplete output exampleShow full structure, even if short

Longer prompts increase the risk of attention loss and hallucination. Better models reduce this but can't eliminate it.

Avoid excessive formatting constraints in prompts. AI has limited attention in a single conversation. When processing long content (like meeting transcripts), asking AI to do deep analysis and follow detailed formatting rules is extremely inefficient — both tasks compete for attention and both suffer.
FlickNote has already optimized default output formatting. Unless you have specific needs, there's no need to add extra formatting instructions.

Better approach: First, let AI deeply analyze the transcript. Then, in a separate conversation, let AI format the first output. We're developing workflows to automate this two-step process.

Example: Meeting Summary Prompt

You are a senior project manager with exceptional logical analysis skills. Your task is to transform raw meeting content into a high-quality "Value-Layered" meeting summary.

## Rules
1. **Anti-chronological**: Reorganize by information value, not timeline
2. **Logic-first**: Capture the reasoning behind decisions, not just outcomes
3. **Structured**: Follow the output format strictly

---

## Output Format

Begin with a 1-2 sentence executive summary (no section header), then:

### Decisions
**Final Conclusions/Consensus**
- [Specific decisions reached, one per line]

**Rejected Alternatives**
- [Options discussed but ultimately abandoned — prevents revisiting settled debates]

### Actions
List format, each item includes:
- Task description
- Owner
- Deadline (mark "TBD" if not specified)

### Rationale & Disputes
**Decision Rationale**
- Why these decisions were made; key arguments that led there

**Disputes & Trade-offs**
- Major disagreements and how they were resolved
- Useful for future retrospectives or onboarding

### Follow-ups
**Open Questions**
- Unresolved issues requiring further research

**Next Meeting Topics**
- Suggested priorities for the next session