What is Prompt Engineering, and Why is It Important?
Prompt engineering is the art and science of crafting proper instructions for Generative Artificial Intelligence (GenAI) tools to produce desired outcomes. Think of prompting like a dialogue with a highly capable digital assistant. For it to meet your requests, you must give specific and direct instructions and guidelines. A deep understanding of AI’s limitations is key to using these tools accurately and effectively, ensuring the best outcomes.
Using an LLM to create first draft content, edit, summarize, and outline are excellent ways to leverage GenAI technology. However, upon first use of these platforms, people report varying levels of satisfaction with the results. Most of the time, a less-than-stellar output is a direct result of poorly designed prompts. GenAI can help you work more efficiently by generating insights, ideas, and content. However, the quality of these outputs is directly related to the quality of the prompt.
This blog will highlight 10 tips for creating effective prompts that can be applied regardless of the job at hand. We will also provide an example of prompts with a few of our best practices applied.
Ten Ways to Make Your Prompts More Effective
- Give Clear and Specific Instructions
Imagine that you are teaching a new skill to a group of students. You must provide as much information and context as possible to get a positive result. Look at leveraging an LLM in the same way. Aim to keep it simple: stick to essentials and avoid misunderstandings.
- Include Context on Formatting and Length
Do you want a list? A draft or narrative? A summary? Be clear on the structure of your desired output.
If you want a title, an intro paragraph, and three headings with subheadings, include that in your prompt. If you have length or character limitations, include that as well. Although exact character or word counts may vary, providing that instruction will yield a result closer to your ideal.
- Provide Proper Content and Context
Provide an appropriate amount of detail about the context. Ambiguous prompts often lead to irrelevant or vague responses. For example, instead of saying, “Tell me about historical conflicts,” say, “Tell me about the causes of World War II.”
Outline specific topics or information to include in the prompt. In government contracting, instead of a generic prompt like “Write a report on safety protocols,” use “Write a compliance report on the safety protocols of a federal RFP.” Make sure to include details like “ensuring alignment with OSHA standards.”
- Include Instructions on Style and Tone
Specify if you want the text written conversationally or in active voice in your prompt. For government proposals, maintain a formal, precise tone. Specify this in your prompt, e.g., “Write in a formal tone for a government proposal, avoiding creative language.”
- Give the Appropriate Level of Detail
Provide context on how detailed you want your response to be. Are you looking for an overview or an in-depth analysis? Giving instructions will provide a more accurate first draft. For proposal management, you might say, “Provide an in-depth analysis of the evaluation criteria for this RFP.”
- Use Open-Ended or Specific Questions
- Open-Ended Prompts for Creativity: To generate different ideas, ask open-ended questions like, “What are creative ways to improve time management?”
- Specific Questions for Factual Answers: If you need a precise answer, frame your question accordingly, such as, “What year did the Apollo 11 mission land on the moon?”
For government contracting, specific questions might be: “What are the main compliance requirements for a GSA Schedule 70 contract?”
- Incorporate Examples
You can provide the model with concrete examples by referencing part or all of a document. This helps to clarify your expectations and remove ambiguity by focusing on the provided structure. Some examples include:
- Specific process documents
- Sample text to give content on tone, structure, style, etc.
- Templates
- Document sections for summarization
For proposal writing, you could incorporate examples such as, “Use the compliance matrix from our previous successful proposal as a guide.”
- Break Down Complex Requests
Step-by-Step Instructions: Break the prompt into smaller parts for multiple-step tasks. Instead of asking, “Write a report on inflation rates and suggest solutions,” start with, “Explain the current inflation rate in the United States.” Then ask, “Suggest practical solutions to decrease inflation.” Build the narrative as you would in a conversation.
This is a similar approach when tackling an RFP. We wouldn’t ask the model to “write an RFP response.” We may start by asking for an outline, an executive summary, answers to specific questions, etc. For instance, “Provide an outline for a technical volume of a government proposal responding to an IT infrastructure RFP.”
- Focus on What You Want the Model to Do
The general idea here is to avoid focusing on what you want to avoid. The best results are created when we direct our efforts toward our desired outcome. For example, don’t say, “Write a document without passive voice;” instead, say, “Write a document in active voice.”
Negative instructions increase the potential for misunderstanding, thereby increasing the potential for confusion and hallucinations.
- Plan to Iterate
The best results for more complicated asks generally come from an iterative approach. Even with detailed and effective prompts, the model may not provide a perfect answer on the first try. Focus on one task or workflow at a time to build the complete product. Always remember, if something isn’t quite right the first time, you can work with the model to make adjustments.
Additionally, “utility” prompts help to expand on topics and facilitate the editing process. Some examples include:
- Simplifying or summarizing a paragraph
- Reducing complexity
- Addressing structure
- Expanding on a topic
- Changing tone
- Reducing/expanding word count
- Checking compliance with RFP requirements
For instance, you could prompt: “Review this proposal draft to ensure compliance with the solicitation’s evaluation criteria.”
Prompting Example
A basic prompt generally does three things:
- Contains a task or question for the AI to perform
- Frames the context of your request
- Gives any relevant detail or constraints that the AI needs to create the most relevant output
Example:
Create an audit report on the safety of The Beaver Valley Power Station in Beaver County, Pennsylvania. The recipient is the Department of Energy.
In this example, I provided an instruction (create an audit report), the context (facility location), and relevant detail (requesting agency). By adding additional framing and context, I can enhance this prompt and the relevance of the response.
I can add what I would like the focus to be:
Create an audit report on the safety of The Beaver Valley Power Station in Beaver County, Pennsylvania. The recipient is the Department of Energy. The report must focus on safety, quality, downtime, and environmental risk mitigation.
I can further specify by adding context on what I would like my output to look like:
Create an audit report on the safety of The Beaver Valley Power Station in Beaver County, Pennsylvania. The recipient is the Department of Energy. The report must focus on safety, quality, downtime, and environmental risk mitigation. The response must have headings for each focus area and be less than 2,500 words.
For a government contracting audience, consider a similar example:
Create an executive summary for a proposal responding to a Department of Defense RFP for cybersecurity services. The executive summary should highlight our team’s experience, technical capabilities, and compliance with the requirements outlined in the solicitation. The output should be no more than 500 words, in a formal tone suitable for federal contracting.
- This is an example of how I can enhance the output for this specific task. I am adding new elements to my prompt, but I am not introducing new ideas or additional concepts.
Final Thought
By mastering prompt engineering, you can guide AI to analyze complex information, create compelling content, and generate structured outputs. Effective prompt engineering is the key to unlocking the full potential of GenAI, helping you stay competitive and innovative.
Want to learn more?
We would love to connect!
To learn more about prompting or to discuss how VisibleThread is leveraging Generative AI to support government contractors, click here to schedule a chat.
