內容目錄
Crafting effective prompts isn’t just about asking a question—it’s about communicating intent clearly.
Over the past two years, AI researchers and practitioners have converged on several structured frameworks that make prompt writing more systematic. Among these, Google’s five-part framework has become one of the most referenced models, shaping how creators, developers, and teams interact with large language models (LLMs).
How Google Developed Its Five-Part Framework
Google’s How to Write a Prompt guide didn’t appear overnight. It evolved from extensive UX research and thousands of real prompts tested across its AI products like Bard and Vertex AI.
By studying how different prompt formats affected quality, clarity, and user satisfaction, Google identified five elements that consistently improved output:
- Persona: Define who the AI should “be.” For example, “Act as a marketing strategist…”
- Aim: Start with a clear action verb that expresses the goal.
- Recipients: Clarify who the response is for (e.g., customers, students, engineers).
- Theme: Set stylistic parameters such as tone, format, or length.
- Structure: Specify how you want the answer presented—bullets, outline, table, or code.
These principles were refined through iterative A/B testing. The result is a lightweight yet flexible framework that users—from students to enterprise teams—can easily apply across creative, analytical, and technical contexts.
Other Notable Prompt-Writing Frameworks
While Google’s approach has become widely adopted, other frameworks have emerged from both the prompt-engineering community and related disciplines. They share similar goals—clarity, completeness, and repeatability—but differ slightly in focus and terminology.
| Framework | Components | Emphasis |
|---|---|---|
| ACE | Action, Context, Example | Focuses on clarity through examples — instructs the model, provides context, and shows sample outputs. |
| CLEAR | Context, Language, Examples, Ask, Review | Encourages iterative refinement — adds a “Review” step to improve prompts based on early responses. |
| MARGE | Model, Audience, Role, Goal, Example | Starts by naming the model being used and aligns the prompt to its capabilities and target audience. |
| Four-P | Persona, Purpose, Parameters, Presentation | Closely mirrors Google’s structure but expands “Theme” into “Parameters,” covering tone, format, and constraints. |
| ABCD (Learning Objectives) | Audience, Behavior, Condition, Degree | Originally from instructional design; defines clear performance criteria and measurable goals for learning tasks. |
A Closer Look at Each
Google’s How to Write a Prompt guide didn’t appear overnight. It evolved from extensive UX research and thousands of real prompts tested across its AI products like Bard and Vertex AI.
By studying how different prompt formats affected quality, clarity, and user satisfaction, Google identified five elements that consistently improved output:
- ACE focuses on clarity through examples. It tells the model what to do, gives context, and shows a sample output.
- CLEAR adds a review loop—after seeing the first response, the user refines the prompt for better accuracy.
- MARGE emphasizes alignment between model choice and audience. Starting with “Model” ensures you tailor expectations to capabilities.
- Four-P mirrors Google’s structure but combines style, tone, and length under Parameters for simplicity.
- ABCD, though originally from education, helps frame measurable learning outcomes that LLMs can follow for training or tutoring content.
Choosing the Right Framework
Different frameworks suit different goals. The “best” one depends on your workflow and what type of content you’re generating.
| Goal / Use Case | Recommended Framework | Why It Works |
|---|---|---|
| General content creation | Google’s Five-Part or Four-P | Both include role, goal, tone, and format—ideal for blogs, copywriting, or summaries. |
| Tasks requiring examples | ACE or MARGE | These frameworks emphasize showing sample outputs, reducing ambiguity and rework. |
| Iterative or collaborative work | CLEAR | Its built-in “Review” step encourages refinement based on feedback loops. |
| Educational or training materials | ABCD | Aligns naturally with learning objectives and measurable outcomes. |
| Technical or code-related tasks | MARGE or Google’s Framework | Starting with “Model” ensures the prompt fits the tool’s strengths and output format. |
Key Takeaways
No single framework fits every situation, but they all share the same DNA: clarity, context, and control.
A good prompt framework helps you think before you type—who the AI is, what it’s supposed to do, and how you want the answer presented. The result is less trial and error, more consistent outputs, and a faster path from idea to usable result.
As AI tools evolve, expect these frameworks to keep evolving too. But the core principle will remain timeless: the clearer you communicate, the smarter your AI becomes.
