How to Write Effective Prompts – Tips from OpenAI’s Prompt Engineering Guide (Part 1)

Knowing how to write effective prompts is essential when working with AI models like ChatGPT. By crafting clear and detailed instructions, you can significanly improve your chances to get more accurate and useful responses from ChatGPT.

And when it comes to learning to craft effective prompts, who’s better to learn from than from OpenAI itself—the closest we can get to the figurative horse’s mouth?

The 6 tactics we’ll cover in this article, drawn directly from OpenAI’s Prompt Engineering Guide, provide expert advice on how to craft better prompts. We’ll walk you through each of them, complete with practical examples and tips, so that you can start using them right away.

Tactic 1: Include Details in Your Query to Get More Relevant Answers

One of the most important things you can do is to ensure your prompts are clearly and specifically written. Include the necessary details in your prompts, add more context to help the AI understand exactly what you’re asking for, and write your prompt as precisely as possible. This will save the AI from guessing what you want, which can lead to inaccurate or irrelevant answers.

Let’s look at an example from the guide:

• Vague Prompt: “How do I add numbers in Excel?”
• Detailed Prompt: “How do I add up a row of dollar amounts in Excel? I want this to be done automatically across the sheet, with the totals displayed in a column labeled ‘Total’.”

The detailed prompt provides context, specifies the task, and makes it clear how the result should look.

Tactic 2: Ask the Model to Adopt a Persona

Assigning a persona means asking ChatGPT to act in a specific role when attending to your task. This can shape the tone and style of its response to better meet your needs. Additionally, it helps the AI focus on relevant information (out of its vast pool of knowledge) that aligns with the assigned role.

You can ask the AI to act as a professional advisor, a friendly teacher, or any other role that suits your context. An example prompt for this is:

Imagine you are a stand-up comedian and a cat owner of 7 years. Give me a cat joke that even non-cat lovers can relate to.

The AI will respond in a way that reflects the persona of a stand-up comedian and cat owner, delivering a joke that’s relatable even to non-cat lovers, as you requested.

Tactic 3: Use Delimiters to Clearly Indicate Distinct Parts of the Input

A delimiter is a character or sequence of characters used to clearly separate or mark distinct sections of text, such as triple quotation marks, XML tags, and labels (see below). Using delimiters in your prompt helps prevent ChatGPT from mixing things up. It is especially useful in complex tasks where different sections of text need to be handled differently.

Some examples of delimiters are:

• Triple quotation marks

Summarize the following article on typical cats' behaviors at night:

""" Insert the article """

• XML tags

Summarize the following article on typical cats' behaviors at night:

<article> Insert the article </article>

• Labels

Below is an article on cats' behaviors at night. Check if the title given to it fits the article. If not, suggest five alternatives:

Article: Insert the article

Title: Insert the title

Tactic 4: Specify the Steps Required to Complete a Task

Certain tasks are best handled in a sequential manner. Breaking down such tasks into clear, step-by-step instructions and explicitly writing the steps out helps guide the AI through complex processes, ensuring ChatGPT doesn’t miss important parts.

An example of this is:

You are a customer service chatbot that will automatically answer customers' queries from Facebook Messenger. Apply the following steps when you respond:
Step 1 - Summarize the customer's input in one or two sentences.
Step 2 - Express empathy.
...

Tactic 5: Provide Examples

This tactic is often referred to as “multiple-shot prompting” or “few-shot prompting” in many guides on how to write effective prompts.

Providing examples helps ChatGPT understand the style or format you expect. By showing the AI what kind of response you’re looking for, you will reduce the likelihood of receiving an output that doesn’t meet your expectations.

This approach is particularly useful when you want ChatGPT to replicate a specific style, tone, or manner when responding.

For example:

Refer to Steve Jobs’ Stanford Commencement Address in 2005, where he said, “Your time is limited, so don’t waste it living someone else’s life. Don’t be trapped by dogma—which is living with the results of other people’s thinking.” Model this speech's tone and inspirational style, and craft a motivational speech on embracing failure as a learning opportunity.

Tactic 6: Specify the Desired Length of the Output

This is often the most neglected specification in our prompts. Being clear about how long you want the response to be can save you from rounds of revision. You can specify the length in words (e.g., “1000 words”), sentences (e.g., “5 sentences”), paragraphs (e.g., “3 paragraphs”), or even bullet points (e.g., “9 bullet points”).

An example prompt is:

Summarize this article on the typical behaviors of cats at night in less than 800 words.

Conclusion

6 tactics from OpenAI’s Prompt Engineering Guide have been reviewed in our discussion on how to write effective prompts. These tactics fall under the “Write Clearer Instructions” strategy of OpenAI’s guide. There are a total of six strategies in that guide, and we will soon have a blog post that covers all of them in detail.

You might find that many of these tactics seem familiar because they have been revealed in other posts on our website. This post, for example: “What is Prompt Engineering and How to Improve Your ChatGPT Prompts”.

This repetition isn’t by chance. All these resources are “singing the same tune” simply because these tactics have been tested and proven to work effectively.

Try out these tactics for writing effective prompts in your next interaction with ChatGPT. Let us know in the comments how they worked for you!