Skip to main content

What is Prompt Engineering?

Key Takeaways​

Prompt engineering is the systematic design and optimization of input text to guide Foundation Models (FMs) toward accurate, deterministic, and context-aligned responses. Instead of relying on naive, open-ended queries (e.g., "Summarize what is AWS"), enterprise prompts use Four Core Structural Blocks:

+---------------------------------------------------------------------------------------+
|                    THE FOUR CORE PROMPT ENGINEERING BLOCKS                            |
+---------------------------------------------------------------------------------------+
|  1. INSTRUCTIONS     ---> The explicit task and behavioral guidelines                 |
|  2. CONTEXT          ---> External background information and target audience persona |
|  3. INPUT DATA       ---> The raw text, document, or data to be processed             |
|  4. OUTPUT INDICATOR ---> The exact format, length, or structural constraint          |
+---------------------------------------------------------------------------------------+

To narrow model boundaries further, Negative Prompting explicitly defines what the model must exclude or avoid, eliminating irrelevant details, hallucinations, and unwanted technical jargon.

Main Discussion​

The Four Pillars of Prompt Construction​

When designing prompts for Amazon Bedrock or any LLM, organizing input into four distinct components minimizes ambiguity and improves consistency:

+----------------------------------------------------------------------------------------------------+
|                               THE FOUR CORE PROMPT COMPONENTS                                      |
+--------------------------+------------------------------------+------------------------------------+
| Component                | Definition                         | Concrete Exam Example              |
+--------------------------+------------------------------------+------------------------------------+
| 1. Instructions          | Specifies the exact task to perform| "Write a clear, concise summary    |
|                          | and behavioral rules.              | focusing on core cloud services."  |
+--------------------------+------------------------------------+------------------------------------+
| 2. Context               | Background situation, domain role, | "You are teaching a beginner-level |
|                          | or target audience persona.        | cloud foundations class."          |
+--------------------------+------------------------------------+------------------------------------+
| 3. Input Data            | The raw content, article text, or  | "<article> AWS provides on-demand  |
|                          | data records being analyzed.       | compute and storage... </article>" |
+--------------------------+------------------------------------+------------------------------------+
| 4. Output Indicator      | Explicit schema, word count, tone, | "Provide a 2-3 sentence summary in |
|                          | or structural formatting output.   | a bulleted list format."           |
+--------------------------+------------------------------------+------------------------------------+

Effective Prompt Execution=Instructions+Context+Input Data+Output Indicator\text{Effective Prompt Execution} = \text{Instructions} + \text{Context} + \text{Input Data} + \text{Output Indicator}

+-----------------------------------------------------------------------------------+
|                                        |                                          |
|                                        v                                          |
|                       [ Foundation Model / Bedrock ]                              |
|                                        |                                          |
|                                        v                                          |
|                       [ High-Precision Completion ]                               |
+-----------------------------------------------------------------------------------+

Negative Prompting: Establishing Clear Negative Constraints​

Negative Prompting tells the model what not to do. While positive instructions define the target output, negative constraints prevent common generation pitfalls:

+-----------------------------------------------------------------------------------+
| POSITIVE INSTRUCTIONS VS. NEGATIVE PROMPTING BOUNDARIES                           |
+------------------------------------+----------------------------------------------+
| Positive Guidance (What to include)| Negative Constraints (What to exclude)       |
+------------------------------------+----------------------------------------------+
| * Explain core benefits of AWS S3  | * Do NOT use complex technical jargon        |
| * Focus on storage durability      | * Avoid mentioning pricing numbers or quotas |
| * Write for beginner learners      | * Exclude speculative future capabilities    |
+------------------------------------+----------------------------------------------+

  • Reduces Unwanted Content: Directly suppresses off-topic remarks, speculative answers, or unverified claims.
  • Preserves Topic Focus: Keeps multi-turn conversations anchored strictly to the intended business scope.
  • Improves Output Readability: Prevents the LLM from leaking overly complex implementation code when addressing non-technical users.

Exam Guide​

Exam Tips​

  • Identify the Four Prompt Elements: The exam frequently presents prompt snippets and asks you to identify the missing structural component (e.g., if a prompt provides a task, context, and data but forgets length/schema rules, it lacks the Output Indicator).
  • Negative Prompting Purpose: When a scenario asks how to stop a model from outputting sensitive internal jargon, code snippets, or overly verbose commentary without retraining, the answer is Negative Prompting.
  • Prompt Optimization Order: Always apply structured prompt engineering (instructions, context, formatting boundaries, and negative prompts) before reaching for more complex and costly methods like fine-tuning.

Practice Test​

Question 1​

A marketing manager writes the following prompt for Amazon Bedrock: "Summarize our quarterly release notes into two paragraphs for non-technical retail customers. Do not include internal bug tracker IDs or code syntax." Which two prompt engineering techniques are being used in this prompt? (Select TWO.)

  • A. Negative Prompting
  • B. Model Distillation
  • C. Output Indicator
  • D. Supervised Fine-Tuning
  • E. Continued Pre-training
Correct Answer
  • A. Negative Prompting
  • C. Output Indicator
    • Explanation: Specifying "Do not include internal bug tracker IDs or code syntax" is an explicit example of Negative Prompting. Specifying "into two paragraphs for non-technical retail customers" defines the target structure and length, serving as an Output Indicator paired with context.

Question 2​

An AI developer provides a foundation model with an extensive text article and asks: "Extract all action items." The model returns a lengthy conversational paragraph with extra commentary rather than a clean list. Which core prompt engineering block should the developer add to enforce a concise bulleted list?

  • A. Input Data
  • B. Output Indicator
  • C. Embeddings Vector
  • D. Contextual Grounding Check
Correct Answer
  • B. Output Indicator
    • Explanation: An Output Indicator specifies the exact structure, format, length, or schema (e.g., "Format your response as a numbered bullet list containing only action items") required from the model.