A well-crafted prompt is essential for effectively using ChatGPT. By being clear, specific, and structured in your queries, and by providing the right amount of context and detail, you can guide the AI to provide the most useful and accurate responses.
You can approach this topic on 3 Levels:
On a general level: What is Prompt Engineering?
Good prompt engineering requires an understanding of the AI model’s capabilities, the context of the task, and the needs of the user…
On a specific level: What makes a great ChatGPT Prompt?
How to craft effective prompts is what you’ll learn in this article you are reading right now.
On a self-replicating level: How to use ChatGPT to generate prompts?
You can also instruct any Language Model (LLM) to create prompts for you. And you can structure your instruction by referencing these guidelines for crafting effective prompts. We have prepared such a prompt here on how to phrase such a request.
Let’s dive into the 7 key elements to craft effective ChatGPT prompts.
1. Clarity and Specificity
A good prompt is clear and specific.
Vague or overly broad prompts can lead to responses that are similarly unfocused.
A good prompt clearly articulates the question or task. It eliminates ambiguity and makes it evident what type of response is expected.
For instance, asking "Tell me about climate change" might yield a general overview, whereas a prompt like "Explain the impact of climate change on Arctic ecosystems in the last decade" will guide the generative AI to provide a more targeted and detailed response.
- Be Clear: Your prompt should clearly state what you're asking or looking for. Avoid ambiguity that could lead the AI in an unintended direction.
- Be Specific: Detailed prompts help narrow down the AI’s focus.
2. Avoiding Ambiguity
Ambiguity in prompts lead to varied and sometimes unintended responses.
It's important to phrase prompts in a way that minimizes potential misunderstandings.
For instance, the prompt "How do you start a business?" can be interpreted in many ways – from legal steps to philosophical approaches to entrepreneurship. Specifying "What are the legal steps to start a small business in California?" would yield a more focused response.
- Be Clear: Your prompt should clearly state what you're asking or looking for. Avoid ambiguity that could lead the AI in an unintended direction.
- Be Specific: Detailed prompts help narrow down the AI’s focus.
3. Contextual Information
Including relevant context within the prompt significantly alters the response. Language models rely on the prompt to understand not just the question but the scope and depth of the answer required.
For example, "As a beginner, how should I start learning Python programming?" includes the context of being a beginner, which tailors the response towards foundational concepts and resources suitable for beginners.
Providing relevant background information within the prompt will guide the AI to generate more accurate and contextually appropriate responses.
- Background Details: Include relevant background information to frame your question. This helps the AI understand the scope and depth of the response needed.
- Skill Level Appropriateness: If you’re asking for explanations or instructions, mention your level of understanding or expertise, e.g., beginner, intermediate, or expert.
4. Using Keywords and Phrases
Leveraging keywords and phrases can lead to “better” results. Certain keywords or phrases act as strong indicators for the model to take a particular direction or adopt a specific tone.
For example, adding "in a professional tone," or "using simple language," to a prompt shapes the style of the response. Using technical jargon and lingo in a prompt like "in a professional tone," or "," will guide the AI to a certain answer more likely (because it is semantics and vectors after all).
Using examples in the prompt will also guide the AI model in understanding the expected format or approach for the response. For instance, when asking for a poem, specifying a style or giving an example line can be helpful.
- Guide the Tone and Style: If you require the response in a particular style or tone, include that in your prompt. For example, “Explain quantum mechanics in simple terms for a non-scientist.”
- Technical Jargon: Use technical terms if you seek a professional-level response, but avoid them if you prefer a layman’s explanation.
5. Structured Querying for Complex Questions
Try using structured querying. Especially in technical fields, structuring the prompt in a logical, step-by-step manner can aid in receiving comprehensive and well-organized responses.
This is the infamous "let's think step by step" prompt fragment.
This is akin to breaking down a complex problem into smaller, more manageable questions. It also helps GPT to “think” and move along its semantic vectors.
Complex reasoning capabilities in LLMs can be elicited by prompting the model to “think step by step” (Wei et al., 2022; Kojima et al., 2022) and can be further improved by instructing them to reflect on their outputs (Madaan et al., 2023; Chen et al., 2023).
- Break Down Complex Questions: For complex topics, break down your query into smaller, more manageable questions.
- Step-by-Step Approach: Ask the AI to follow a logical sequence or steps, especially for technical or analytical topics.
6. Desired Output Structure
Incorporating the desired structure will help you gain better results. If the desired output is a list, essay, summary, or another specific format, it helps to mention this in the prompt.
For instance, "List the top 10 most populated cities in the world" clearly indicates that a list is expected, not a descriptive paragraph.
Providing examples helps here as well.
- Specify the Format: If you want the response in a specific format (e.g., list, essay, bullet points), state this in your prompt.
- Use Examples: Providing an example of the desired format can be helpful.
7. Balancing Brevity with Detail
While being concise is generally good, overly brief prompts might miss essential details. A balance is needed where the prompt is neither too verbose nor too skimpy.
- Not Too Brief: Ensure your prompt has enough detail to guide the AI accurately.
- Not Too Long: Avoid overly lengthy prompts that could confuse the main request.
8. Adapting to AI’s Capabilities and Limitations
Understanding the strengths and limitations of the AI model is vital. For instance, GPT models are excellent at generating human-like text but have limitations in handling real-time data or providing highly specialized expertise in niche domains.
- Understand Limitations: Be aware that AI may not have the latest data or deep expertise in very niche areas.
- Realistic Expectations: Tailor your prompts to align with the capabilities of the AI.
Further Considerations
Iterative Process: If the initial response isn’t what you expected, refine your prompt based on the AI’s output and try again.
Reflect on Responses: Use the AI’s responses to refine your approach to asking questions.
