Intermediate Prompting

Advanced Prompt Structuring

Basic prompts can take you far, but these advanced structuring techniques will help you tackle more complex challenges with AI assistants.

Chain-of-Thought Prompting

Chain-of-thought prompting guides the AI through a logical reasoning process, similar to showing your work in a math problem. This technique is particularly effective for complex problems that benefit from step-by-step analysis.

How it works:

Break down complex reasoning into sequential steps
Ask the AI to “think through” each step explicitly
Build toward a conclusion based on the preceding analysis

Example:

Instead of: “What’s the best investment strategy for my situation?”
Try: “To determine the best investment strategy for my situation, think through this problem step by step: First, analyze my risk tolerance based on age and financial goals. Next, consider appropriate asset allocation given market conditions. Then, evaluate specific investment vehicles that match this allocation. Finally, conclude with a recommended strategy and implementation plan.”

When to use it:

Mathematical or logical problems
Complex decision-making scenarios
Multi-factor analyses
Troubleshooting technical issues

Role-Based Prompting

While the beginner guide introduced the Expert Selection block, intermediate prompting takes this further by creating more nuanced roles, combining multiple expertise areas, or establishing specific relationships between the AI and the task.

Single Specialized Role:

“Act as a growth marketing specialist who has worked exclusively with SaaS startups for the past decade and has a strong background in customer acquisition cost optimization.”

Multiple Combined Roles:

“Approach this question as both a pediatric nutritionist and a child psychologist who specializes in eating behaviors. Analyze how both nutritional needs and psychological factors affect children’s eating habits.”

Role with Relationship Context:

“As a personal financial advisor who has been working with me for five years and is familiar with my conservative investment approach and goal of early retirement, evaluate this potential investment opportunity.”

When to use it:

When you need interdisciplinary perspectives
For specialized technical knowledge
When simulating stakeholder viewpoints
To create consistent tone and expertise across multiple interactions

Strategic Context Management

Information Hierarchies

How you organize information in your prompt significantly impacts the quality of responses. Strategic arrangement creates a framework that helps the AI process your request more effectively.

From General to Specific:

Start with broad context, then narrow to specific details:
“I’m developing a marketing strategy for an e-commerce business (general) that sells sustainable home goods (more specific) with a focus on plastic-free kitchen products (most specific).”

Order of Relevance:

Present the most important context first:
“My primary goal is increasing email sign-ups (most important). I’m considering redesigning our website landing page (relevant context). Our audience is primarily environmentally-conscious millennials (additional context).”

Procedural Information Using Lists:

Use numbered lists for steps or processes:
“Analyze our customer acquisition funnel following these steps:

1. Examine traffic sources and entry points
2. Evaluate landing page conversion rates
3. Analyze shopping cart abandonment
4. Review checkout completion rates
5. Assess post-purchase engagement”

State Management

In multi-turn conversations, maintaining context becomes crucial. These techniques help manage the conversation state more effectively:

Reference Previous Exchanges:

“In our previous discussion, we identified three key market segments. Now, I’d like to focus on developing specific messaging for the second segment—busy professionals with limited time.”

Create Working Memory:

“For this conversation, let’s define our key metrics as follows: CAC = Customer Acquisition Cost, LTV = Lifetime Value, and Conversion Rate = percentage of visitors who make a purchase. Please use these definitions consistently throughout your analysis.”

Context Anchoring:

“Keep in mind throughout this analysis that our primary constraint is a limited marketing budget of $5,000 per month, and our target ROI is 3x spend.”

Output Control Techniques

Once you’ve mastered input structuring, the next step is controlling what you get back. These techniques help you shape AI responses with precision, ensuring you receive exactly the type of content you need in the format that works best for your use case.

Precision Over Creativity:

For technical or factual outputs, explicitly request precision over creativity. This helps the AI focus on accuracy rather than stylistic flourishes.

Output Length Control:

Set specific parameters for response length: “Provide a 2-3 sentence summary” or “Limit your analysis to 500 words maximum.” This prevents overly verbose or truncated responses.

Sequential Depth Control:

Guide the AI to build progressively detailed responses with instructions like “First, provide a one-paragraph summary. Then, expand on the three most important points in detail.”

Format Specification

While the beginner guide mentioned the Desired Output block, intermediate prompting takes format control to a new level of precision.

Structured Data Formats:

“Provide your analysis in JSON format with the following structure: { “marketSegment”: “string”, “recommendedChannels”: [“string”], “estimatedCAC”: “number”, “potentialROI”: “number”, “implementationTimeline”: “string”, “risks”: [“string”] }”

Template with Placeholders:

“Please fill in this email template for our product launch:

Subject: [Compelling Subject Line]

Hi [First Name],

[Opening sentence that creates urgency]

We’re excited to announce [product name] is now available. It helps [target audience] to [key benefit] without [common pain point].

[2-3 sentences about key features]

[Call to action with specific next step]

[Closing line]

[Signature]”

Section Headers and Structure:

“Structure your competitive analysis with exactly these headers and in this order:
1. Executive Summary (2-3 sentences)
2. Methodology (how you approached the analysis)
3. Market Position Comparison (with subheadings for each competitor)
4. SWOT Analysis (in table format)
5. Key Differentiators (bullet points)
6. Strategic Recommendations (prioritized list)”

Calibrated Language

Fine-tuning the language level, technical depth, and tone of responses is crucial for tailored communication.

Technical Depth Control:

“Explain machine learning overfitting at a technical level appropriate for a data science intern with basic statistics knowledge but no machine learning experience.”

Reading Level Specification:

“Provide an explanation of climate change impacts that would be understood by a bright 10-year-old child.”

Tone Calibration:

“Write this rejection email using a professional, empathetic tone that leaves the door open for future opportunities while clearly communicating our decision.”

Domain-Specific Language:

“When discussing the marketing strategy, use standard industry terminology from growth marketing and customer acquisition, particularly regarding attribution models and conversion optimization.”

Error Recovery and Refinement

Even with well-crafted prompts, you may not always get ideal results on the first try. Intermediate prompters excel at iteratively improving responses through targeted feedback and systematic refinement techniques.

Ineffective Refinement

“This isn’t what I wanted. Please try again.”

Effective Refinement

“Your analysis focuses too much on market trends and not enough on competitive positioning. Please revise to include a detailed comparison with our top three competitors, especially regarding their pricing models and customer acquisition strategies.”

Specific Refinement Directives:

Instead of vague feedback, pinpoint exactly what needs improvement: “The tone is too formal. Please rewrite using a more conversational voice while maintaining all the technical information.”

Component Isolation:

When a prompt yields mixed results, identify which parts worked and which didn’t: “The technical analysis portion is excellent, but the implementation recommendations lack specificity. Keep the technical section as is, and expand only the recommendations with concrete action steps.”

Incremental Improvement:

Approach complex refinements in stages rather than attempting to fix everything at once: “First, let’s focus on improving the data analysis methodology. Once that’s refined, we’ll revisit the conclusions and recommendations.”

Alternative Approaches:

When a particular prompting strategy isn’t working, try a fundamentally different approach: “Instead of analyzing this as a marketing problem, let’s approach it from a user experience perspective. Reframe the analysis to focus on customer journey touchpoints rather than conversion metrics.”

Variant Testing Example:

“I’ll try requesting this information in three different ways. Please tell me which approach gives you the clearest understanding of what I need:

Variant 1: “Analyze the financial implications of our expansion into European markets.”

Variant 2: “As a CFO with experience in international business, evaluate the ROI potential, regulatory challenges, and tax implications of our planned expansion into France, Germany, and Spain in Q3 2023.”

Variant 3: “Create a financial analysis with these specific sections: 1) Initial capital requirements for European expansion, 2) Expected timeline to break-even by country, 3) Tax optimization strategies, 4) Currency hedging recommendations, and 5) Regulatory compliance costs.”

Feedback Integration

Learning to refine prompts based on previous responses is key to advancing your prompting skills.

Specific Feedback:

“Your previous response was too focused on theoretical frameworks rather than practical implementation. Please revise to include specific, actionable steps rather than general principles.”

Reference Examples:

“Your previous explanation was too complex. I need something closer to this example: [insert example]. Notice how it uses simple analogies and visual descriptions.”

Contrast Pairs:

“Your writing style should be less like an academic paper (formal, passive voice, complex sentences) and more like a blog post (conversational, active voice, concise explanations with examples).”

Prompt Debugging

When you don’t get the results you want, systematic troubleshooting helps identify and fix the issue.

Pattern Documentation:

Keep a record of which prompt structures work best for specific types of tasks, creating a personal library of effective patterns.

Practical Applications

Intermediate prompting techniques truly shine when applied to specific use cases. By tailoring your approach to different contexts, you’ll achieve more consistent, high-quality results for various professional tasks.

Generic Approach

“Write me some code to analyze customer data.”

Specialized Approach

“Create a Python function that segments customer data based on purchase frequency, average order value, and recency of last purchase using the RFM modeling approach.”

The examples below showcase how to structure prompts for specific professional use cases. Each demonstrates how tailoring your prompting approach to the task at hand yields superior results compared to generic requests.

Domain-Specific Templates:

Create reusable prompt templates for recurring tasks in your field, such as legal contract analysis, financial report generation, or medical literature reviews.

Workflow Integration:

Design prompts that fit into existing workflows, ensuring AI outputs can be seamlessly incorporated into your team’s processes and tools.

Cross-Functional Communication:

Use prompts to translate complex information between different professional domains, such as converting technical specifications into marketing language, or translating financial data into strategic recommendations.

Code Generation and Analysis

Prompting for code requires specific techniques to get clean, functional results.

Language and Style Specification:

“Generate Python code using functional programming principles rather than object-oriented design. Use type hints and follow PEP 8 style guidelines.”

Documentation Requirements:

“Include docstrings for all functions following Google’s Python Style Guide format, with parameters, return values, and usage examples.”

Testing Considerations:

“After providing the main function, include unit tests that cover the happy path, edge cases, and error handling scenarios.”

Example:

Expert Selection

I want you to act as a senior Python developer with expertise in data processing and API integration.

Context

I’m building an application that needs to fetch weather data from the OpenWeatherMap API and process it for a dashboard.

Instruction

Write a Python function that:

Takes a city name and API key as parameters
Makes a request to the OpenWeatherMap current weather API
Processes the JSON response to extract temperature, humidity, and weather description
Handles common errors like invalid city names or connection issues
Returns the processed data in a consistent format

Negative Prompt

Avoid using third-party libraries beyond requests. Don’t write a complete application, just the core function with proper error handling.

Desired Output

Provide the function with:

Comprehensive docstring following Google style
Type hints for all parameters and return values
Clear error handling with appropriate exceptions
Example usage code
3-4 unit tests showing how to test the function

Content Transformation

Converting content from one format or style to another is a powerful application of intermediate prompting.

Clear Transformation Parameters:

“Transform this technical product description into marketing copy that:
– Converts technical specifications into benefit statements
– Maintains all key product features but expresses them as user outcomes
– Uses a conversational, second-person point of view
– Includes strong calls-to-action
– Is suitable for a landing page”

Element Preservation:

“Rewrite this academic abstract in plain language while preserving:
– All key statistical findings with their significance values
– The core methodology description
– The main implications of the research”

Example Pairs:

“Transform the following customer service email similar to how this example was transformed:

Original: ‘Your request has been received and will be processed in accordance with our standard procedures.’

Transformed: ‘Thanks for reaching out! We’ve got your request and we’re on it. You can expect to hear back from us within 24 hours.’

Now transform this email: ‘Upon investigation, it has been determined that your account status requires additional verification procedures before the requested changes can be implemented.’”

Using Blink Prompt Canvas for Intermediate Techniques

The Blink Prompt Canvas can be adapted for these intermediate techniques:

Enhanced Block Use

Expanded Expert Selection Block:

Define more nuanced roles with specific experience levels, specialized knowledge areas, and combined disciplines.

Structured Context Blocks:

Organize context in hierarchical order using numbered lists or bullet points within the Context block.

Instruction Block with Chain-of-Thought:

Structure your instructions to guide the AI through specific reasoning steps.

Advanced Output Specification:

Use code formatting, tables, or detailed templates within your Desired Output block.

Template Combinations

Create template combinations for specific use cases:

The Technical Debugger:

Combines Expert Selection (technical expert), Context (system details), step-by-step debugging instructions, and formatted output requirements.

The Content Transformer:

Combines Context (original content), Instruction (transformation parameters), examples of “before and after,” and Negative Prompt (what aspects to preserve).

Best Practices for Intermediate Prompters

Documentation and Iteration

Keep a record of your most effective prompts and how they evolved. Note which techniques worked best for specific tasks and why.

Systematic Testing

Compare results from different prompt structures for the same task to identify optimal approaches. Test one variable at a time when refining prompts.

Contextual Limitations

Understand when to use different techniques. Some approaches (like chain-of-thought) may be unnecessary for simple tasks but essential for complex ones.

Feedback Loops

Develop a process of prompt-response-refinement, treating each interaction as an opportunity to improve your prompt engineering skills.

Conclusion

Advancing from beginner to intermediate prompting isn’t just about more complex techniques—it’s about developing a more nuanced understanding of how AI processes and responds to different prompt structures. By mastering these intermediate techniques, you’ll be able to tackle more complex tasks, generate more precise outputs, and create repeatable patterns for consistent results.

The Blink Prompt Canvas provides an ideal framework for implementing these intermediate techniques, allowing you to save and refine your most effective prompt structures while continuing to experiment with new approaches.

Intermediate Prompting