Guide to Mastering AI Prompting

An in-depth guide to AI prompt engineering, covering techniques to craft effective prompts, optimize outputs, and enhance productivity with large language models.

Introduction

Artificial Intelligence (AI) refers to the simulation of human intelligence by machines. It enables machines to process large amounts of data, recognize patterns, and make predictions. One important subset of AI is Machine Learning (ML), which analyzes large datasets to identify trends that help machines predict future outcomes.

Among AI technologies, Large Language Models (LLMs)—like ChatGPT, Gemini, and Claude—excel at understanding and generating human-like text. Trained on extensive data such as books, articles, and websites, LLMs use pattern recognition to predict and complete text based on a user's input. Despite their capabilities, LLMs depend on effective prompting to deliver useful and accurate outputs.

What is Prompt Engineering?

Prompt engineering is the practice of writing, refining, and optimizing prompts to control how AI models respond. By carefully structuring prompts, users can guide AI to produce desired results more efficiently, saving time and resources (especially token limits on platforms like ChatGPT). Similar to improving search engine queries, better prompts yield higher-quality outputs.

Prompt engineering also involves monitoring and iterating on prompts, as AI models can generate variable and unpredictable results depending on how the task is phrased.

Key Components of Effective Prompts

Mandatory

1. Task

Clearly define the goal of your prompt using action verbs such as "generate," "analyze," or "summarize." There are two types of tasks: simple tasks and multi-tasks. The following are examples of both:

• Simple Task Example: "Generate a list of popular science fiction authors."

• Multi-Task Example: "Analyze the text, summarize it, and categorize the main arguments."

Important

2. Context

Providing context helps the AI focus on relevant information. Technically you could give endless amounts of context. But you have to limit the endless possibilities. Example:

• Poor context: "I’m a 70kg male."

• Better context: "I’m a 70kg male aiming to gain 5kg of muscle in 3 months with 2 gym sessions per week." Use these three guiding questions to nail your context

  • Who is the prompt for? Helps identify the key audience or subject.

  • What specific outcome do you expect? Ensures the goal is clear and measurable.

  • What constraints or conditions apply? Focuses on the limitations or parameters that will shape the answer, like time, resources, or tools available.

3. Exemplars/Few-Shot Prompting

In few-shot prompting, you provide (an) example(s) within your query to guide the AI. These examples demonstrate the desired format, tone, or structure, helping the model better understand what you expect. Example:

• Task & Context: "Rewrite my resume bullet points using this format: 'I accomplished X by Y, resulting in Z.'"

• Exemplar: "I lowered hospital mortality by 10% by educating nurses on new protocols, saving 200 lives annually."

Few-shot prompting can improve accuracy for complex tasks by offering the model a pattern to follow, but it consumes more tokens and requires careful selection of examples. Examples aren’t necessary but they greatly improve the quality of the result.

Optional, but Useful

4. Persona

Define a character or perspective for the AI to adopt. This can help ensure responses match the tone, style, or expertise required for your task.

Oftentimes this is helpful because you’re narrowing down the possibility of predictions the LLM can take. When you ask a general prompt without making the AI adopt a persona it will get one of the most likely responses based on all the information on the internet.

If you’re looking to write a section on your book, for example, the answer the AI would give without a persona may seem completely off to the rest of your book. Suggesting different styles or authors the AI could take on and/or feeding it your own writing samples could make the writing fit better within the context of the whole book.

5. Format

Being explicit about the format of the response you want can make AI outputs much more useful. Tip: Visualize the exact format you want the end result to be.
Instead of writing, “Summarize this article,” you could say, “Summarize this article into 5 bullet points, with each point no longer than 15 words.” This way, the AI knows exactly how to structure the response, saving you time and effort.
Specific formats you could tell the LLM to follow includes checklists, word counts, bullet points, tables, step-by-step instructions, dialogue, headings/subheadings, code, Q&A format, email format, etc.
**Note: Formatting and exemplars may be interchangeable in some cases. If you give examples which follow a specific format it may not be necessary to explicitly mention the format you want the response to be in.

6. Tone

Tone can be useful in order for the LLM to convey the feeling you want the writing to give off. If you can’t find a word that describes the feeling you want to convey, you can do one of two things

1. Describe the feeling directly in the prompt, or

2. Ask the LLM for a list of 5 descriptive words based on the feeling description and then choose two of the best ones.

   1. For example: “I’m writing an email and I don’t want to sound too stuck up or cringy. Give me a list of 5 tone keywords I can include in a prompt for Gemini that describes this feeling.”

The Perfect Prompt

Using all of the knowledge in this article, we can build the perfect prompt by combining all of the building blocks aforementioned. The purpose of this prompt is to write an email to our boss sharing some positive news received. The building blocks are as follows.

Prompt Building Blocks

• Persona: You are a senior product marketing manager at Apple.

• Context: You have just unveiled the latest Apple product in collaboration with Tesla, the Apple Car, and received 12,000 pre-orders, which is 200% higher than target.

• Task: Write an email to your boss, Tim Cookie, sharing this positive news.

• Format: The email should have structured sections, including clearly defined content for each part.

• Exemplar: Include a tl;dr (too long, didn’t read) section, project background (explaining why the product came into existence), business results (quantifiable metrics), and a thank you section for the product and engineering teams.

• Tone: Use clear, concise, and confident language with a friendly tone.

Combining Building Blocks

You are a senior product marketing manager at Apple. You have just unveiled the latest Apple product in collaboration with Tesla, the Apple Car, and received 12,000 pre-orders, which is 200% higher than target.

Write an email to your boss, Tim Cookie, sharing this positive news. The email should have structured sections, including a tl;dr (too long, didn’t read) section, project background (explaining why the product came into existence), business results (with quantifiable metrics), and a thank you section for the product and engineering teams.

Use clear, concise, and confident language with a friendly tone.

Other Prompting Tips

1. Iterative Prompting

Treat prompting as a collaborative process. If the AI's initial response doesn't meet your expectations, provide additional instructions or follow-up questions.
Example:

• Initial Prompt: "Explain photosynthesis."

• Follow-up: "Simplify the explanation for a 10-year-old."

• Follow-up: "Break the process into three simple steps."

2. Avoiding Leading Prompts

Be careful not to bias the AI with suggestive phrasing. Instead of asking, "Why is solar energy better than wind energy?" frame the question neutrally: "What are the advantages and disadvantages of solar and wind energy?"

3. Bold Changes

If you want to proofread the LLM’s changes, ask them to bold all changes made. This will make it easy to see what has changed and what has remained the same in the new version.

Common Challenges and Considerations

1. Token Limits

AI models process text in units called tokens. For platforms with free or limited access, you must ensure your prompts are efficient to avoid exceeding token limits.

2. AI Hallucinations

AI hallucinations occur when the model generates false or misleading information. This can result from vague prompts. For research purposes, it's advisable to use search-integrated tools like ChatGPT Search, Perplexity, or Google Search for reliable, fact-based information.

3. Prompt Libraries

Maintaining a prompt library can save time by allowing you to reuse and refine successful prompts for recurring tasks.

Conclusion

Mastering prompt engineering empowers users to harness the full potential of AI tools like ChatGPT, Gemini, and others. By clearly defining tasks, providing context, including examples, and specifying output formats, you can guide AI to produce more accurate, efficient, and creative responses. With practice and iteration, you’ll develop prompts that maximize both productivity and the quality of AI-generated content.