Recap: Prompt Fundamentals
The anatomy of an effective prompt
Every powerful prompt has four layers working together:
1. Role — Who should the AI be? You are a senior UX designer with 10 years of experience building mobile apps.
2. Task — What exactly should it do? Be specific: not 'help me' but 'write a 200-word explanation of why onboarding flows need a skip button.'
3. Context — What background does the AI need? Share the audience, the platform, the constraints, the stakes.
4. Format — How should the output look? Bullet list, numbered steps, JSON, table, email, code block — always specify.
When any layer is missing, the model fills the gap with assumptions — and those assumptions are rarely what you wanted.
Zero-shot vs few-shot — when to use each
Zero-shot works for clear, unambiguous tasks: Summarize this article in 3 bullet points. The model knows exactly what a summary looks like.
Few-shot becomes essential when you need a very specific style, format, or judgment the model cannot infer from the instruction alone. You provide 2–5 examples of input→output pairs before your actual request.
Example few-shot setup:
```
Classify the sentiment of each tweet as Positive, Negative, or Neutral.
Tweet: 'This new feature saved me hours!' → Positive
Tweet: 'The app crashed again.' → Negative
Tweet: 'Updated to v3.1.' → Neutral
Tweet: 'Finally got my order after three weeks.' → ?
```
The more unusual your desired output, the more examples you need.
The three biggest beginner mistakes
Mistake 1: Vagueness. 'Write something about marketing' gives the AI infinite degrees of freedom. The narrower your constraint, the better the output. Instead: 'Write a 150-word Instagram caption for a Ramadan sale campaign targeting Saudi mothers aged 25–40.'
Mistake 2: No format instruction. Without a format, the AI defaults to paragraphs. Always end your prompt with the structure you need: Format as a numbered list, Return JSON with keys: title, summary, tags, Write in the style of a WhatsApp message.
Mistake 3: Giving up after one attempt. The first output is a draft, not the final answer. Follow up: Make it shorter, Sound more casual, Add a call to action at the end, Translate the key points to Arabic. Iteration is the real skill.
Building your prompt library
Advanced prompt engineers don't write from scratch every time. They maintain a personal library of prompts that work. Start yours today:
1. Create a template folder — Google Doc, Notion page, or 404Fault bookmarks.
2. Save every prompt that produced a great result — include the output so you remember why it worked.
3. Categorize by use case — Content, Analysis, Code, Research, Summarization.
4. Tag by AI tool — some prompts work better on Claude, others on ChatGPT or Gemini.
By the end of this path, you'll have 14 tested, working prompts you built yourself — one from each lesson.
Key Takeaways
- Every prompt needs Role, Task, Context, and Format to perform at its best.
- Use few-shot examples whenever the desired output style is non-standard.
- Iteration — refining the output with follow-up instructions — is the core skill.
- Build a personal prompt library from day one; it compounds over time.
The 4-layer prompt challenge
Take any task you did manually this week and write a 4-layer prompt for it. Include: Role (who the AI is), Task (what to do, specific), Context (background info), Format (exact structure). Compare the output to your first instinctive prompt.