Six Strategies for Crafting Effective Prompts

1. Write Clear Instructions

Tactics:

1. Include Details in QueriesProvide specific details or context to ensure relevant responses, avoiding ambiguous requests that force the model to guess intent.
2. Assign a Role to the ModelUse system messages to specify the role (e.g., travel planner, technical writer) for the model’s responses.
3. Use Delimiters to Separate Input SectionsEmploy delimiters (e.g., ```, ---, ###) to clearly distinguish between different parts of the input (e.g., problem, background, format requirements).Example:
4. Specify Output Format/StructureClearly state the desired output format (e.g., table, code block, bullet list) to reduce post-processing efforts.Hands-on Example:Scenario: Ask GPT to generate a comparison table for AI painting tools.Reference Prompt:
5. Provide Examples or TemplatesGuide the model’s output style with concrete examples, especially for creative tasks (e.g., copywriting, storytelling).System Prompt Example:
6. Emphasize Key ConstraintsClarify limitations (e.g., word count, forbidden terms, timeframes) to prevent off-target outputs.Contrast Example:❌ “Write an article about AGI.”✅ “Write an 800-word introductory article on AGI, avoiding technical jargon and including 3 everyday-life analogies.”

2. Provide Reference Text

Upload or paste reference materials (e.g., brand style guides, historical copy, academic papers) to align the model’s output with specific styles or rules.

Use Cases:

• Enterprises: Have the model write social media copy matching brand tone.
• Students: Adjust paper citations to follow reference formats.
• Creators: Mimic a writer’s narrative style for storytelling.Prompt Tip:“Please write a 种草 (product recommendation) note about AI painting tools based on the attached Little Red Book Blockbuster Copy Guide, including 3 pain-point scenarios and 2 comparative usage diagrams.”

3. Split Complex Tasks into Simpler Subtasks

Adopt a phased approach for multi-level requirements to avoid output 偏差 (deviations) from task overload.

Splitting Example:

Original Requirement: Develop an AI resume optimization tool.Subtasks:

1. Analyze resume content (extract education, work experience, etc.).
2. Match target job keywords (provide industry term libraries).
3. Optimize language (convert declarative sentences to quantified achievements).
4. Generate format suggestions (adapt to different company templates).Tool Prompt:“Current step: Step 2. Please label skill keywords in the user’s resume that need enhancement based on the attached Internet Operations Job JD Keyword Library, and provide replacement suggestions.”

4. Give GPT Time to “Think”

Use Chain of Thought (CoT) prompting to guide step-by-step reasoning, especially for logical or computational tasks.

Operation Method:

1. Require the model to list problem-solving steps first.
2. Explain each step in detail.
3. Provide the final answer.Example:User Question: A bookshelf has 3 layers, each layer holding twice as many books as the one below. The bottom layer has 5 books. How many books are there in total?Model Thinking Process:“Step 1: 5 books on the bottom layer.Step 2: Middle layer = 5 × 2 = 10 books.Step 3: Top layer = 10 × 2 = 20 books.Step 4: Total = 5 + 10 + 20 = 35 books.”

5. Use External Tools

Integrate plugins or APIs to extend capabilities for tasks beyond the model’s native scope (e.g., real-time data queries, file conversion, image generation).

Common Tool Types:

Prompt Template:“Please use [WeatherAPI] to query the weather in Shanghai on July 7, 2025, and reply in the format: ‘Date + Temperature + Weather Condition + Travel Advice’.“

6. Test Changes Systematically

Optimize prompts using controlled experiments, recording how different parameters impact outputs to build a personalized prompt library.

Test Dimensions:

1. Role Setting: Compare itineraries from “travel planner” vs. “local friend” roles.
2. Detail Density: Test recommendation effectiveness with 3 vs. 10 attractions.
3. Format Requirements: Compare readability of list, paragraph, and table formats.
4. Tone Style: Try formal, humorous, and literary tones for copy conversion rates. Test Record Example: