5 Expert Tips for AI Prompt Engineering with Examples

Chances are, you have heard of (or experimented with) generative AI tools such as ChatGPT. But are you truly maximizing the marketing potential of generative AI? 

If you’re ready to leverage generative AI and get a deeper understanding of how it works, this ultimate guide will help you master the art and science of prompt engineering to take your efforts to the next level!

What is generative AI and prompt engineering?

First, let’s define generative AI. Simply put, generative AI is any algorithm that can create new content, be it text, image, video, or audio. 

Prompt engineering is a way of structuring sentences and requests to guide AI systems, specifically large language models (LLMs) like ChatGPT, to provide more targeted and relevant responses. 

It involves crafting effective prompts (i.e. input queries) to help AI models produce the desired output by minimizing any potential misinterpretations and maximizing the accuracy of the generated output.

3 common types of AI prompts

To start off, let’s explore three common types of prompt engineering techniques:

1. Zero-shot prompting

Zero-shot prompting refers to situations where you provide a single input to the model and expect it to understand and execute the task without having seen any similar examples before. In this case, you're hoping the model generalizes from all of its prior training to handle the novel task.

2. Few-shot prompting

Few-shot prompting, on the other hand, means providing the model with a few examples of the task at hand as part of the input. The idea here is to guide the model to the format and nature of the output you desire.

In the example below, you can see we are asking ChatGPT to respond to (ficticious) customer inquiries. We start by giving it two examples of existing inquiries and responses we find appropriate. Then, we ask it to mimic those responses for the new inquiry.

3. Chain of Thought (CoT) prompting

Chain-of-thought prompting, sometimes called "dialogue prompting” or “prompt chaining”, involves breaking down a complex task into a series of simpler tasks. This method relies on maintaining a continuous dialogue with the model where each question builds on the last response.

Let's look at an example where we want to generate a comprehensive marketing plan for a new product. 

This is a complex task that can be broken down into several steps, like target audience identification, channel selection, message creation, and so forth.

CoT prompting example

Here's an example of how a chain-of-thought prompting session could go (responses truncated for brevity):

User: "Describe our new product, a lightweight, waterproof, hiking boot made from recycled materials."

LLM: "Our new product is a high-performance hiking boot, designed with the adventurer in mind. Crafted from…"

User: "Based on that description, who would be our primary target audience?"

LLM: "The primary target audience for this product would be outdoor enthusiasts who enjoy hiking, trekking, or other adventurous activities. They are likely…"

User: "What would be the best marketing channels to reach this target audience?"

LLM: "To reach this audience, consider utilizing outdoor adventure blogs, eco-friendly lifestyle websites, and social media platforms where outdoor enthusiasts…"

User: "Can you draft a promotional message for this product targeted at our audience, to be used on Instagram?"

LLM: "Embark on your next adventure with confidence and care for the planet! 🌍 Our new hiking boots…"

As you can see, each step builds on the previous response, allowing the user to gradually extract more complex and detailed information from the LLM.

In marketing use cases, this is often the most powerful form of prompting, with recent studies showing that CoT prompting can improve the reasoning ability of an AI model by forcing it to “show its work”.

3 reasons why marketers need prompt engineering

By now, you might be thinking, “I’m a marketer. Does AI prompting really apply to me?” To answer that question, here are a few of the biggest reasons why marketers are embracing prompt engineering.

1. Time and cost savings

Prompt engineering can save valuable time and resources for marketing teams by allowing generative AI to create intial draft content more efficiently. With appropriate prompts, AI models can quickly generate copy (for things like ads, promo videos, and blogs), answer customer inquiries, and create plan outlines, which frees up your team to focus on strategy and creative ideas.

Prompt engineering is typically used in the  ideation/brainstorming phase by helping you generate new ideas quickly. While the content won’t necessarily be perfect, it can still provide valuable inspiration in a variety of different tones and styles, which can lay the foundation for your own creativity and unique content.

2. Improved marketing campaigns and engaging content

Effectively engineered prompts can improve marketing campaigns and create engaging content that speaks to your target audience. A well-crafted prompt ensures AI models generate content that aligns with your brand voice, resonates with your audience, and drives the desired action, which can ultimately lead to higher conversions.

Another hidden benefit of AI prompt engineering is that the act of writing out a prompt forces you to articulate what you want, which can not only give you a better understanding of what you’re looking for, but will also make you better at explaining it when you’re collaborating with other, real-life humans. Chain-of-thought prompting is particularly good for this as you’re able to adapt and refine your prompts depending on the outputs, which helps translate your prompts from abstract ideas into something more specific and actionable.

3. A must-have skill for marketers

As AI plays an increasingly prominent role in marketing, mastering prompt engineering is likely going to become a vital skill for professionals in the field. The World Economic Forum predicts that 97 million new jobs will emerge by 2025 due to humans and machines working together, and prompt engineering is one of those roles. This makes prompt engineering a critical and teachable skill for marketing professionals.

And while ‘prompt engineering’ may sound like something brand new or intimidating, it’s really not all that different from what marketers have already been doing for decades with their creative briefs and multiple rounds of feedback sessions. But the biggest difference here is the audience – instead of speaking with a human, you’re interacting with a model based on a large dataset that can adapt its tone and style on the fly. 

Tips for mastering AI prompt engineering

1. Start by understanding the technology

If you want to extract the most value out of AI, it’s important to thoroughly research and understand the capabilities of the underlying AI model you are using. While you don’t need a master’s in computer science or an in-depth background in machine learning (ML), being familiar with the basics of AI and how large language models function will help you use it more effectively.

One of the most popular examples is ChatGPT by OpenAI. If you're using this kind of AI platform, it's essential to be aware of specific factors that can influence its performance, such as:

• Limitations and bias: A large language model only knows what it’s been trained on. That’s true for any AI model, the difference is the amount of data being used is what makes an LLM. Still, even when a large dataset is being used, there are limiations on timeframes (i.e. ChatGPT has limited knowledge of anything that occurred after 2021). 

OpenAI has been very candid about the biased nature of its LLM and it’s important to keep that in mind with any AI model – especially if you’re using it to generate content. Once again, just because a large dataset was used, it’s still only knows the data it was trained on. 

• Hallucinations: A LLM such as ChatGPT can throw together a response that reads and sounds accurate yet actually contains false infomration. Other times the reponse may be very clearly false or even nonsensical. In either case, this is called a hallucination and anyone using generative AI for anything other than experimentation or play should be aware of this.
• Context Window: The context window refers to the maximum number of input tokens or words that an AI model can understand within a given request. Be mindful of your tool's context window capacity to ensure the relevant information is not left out and that the model processes your instructions effectively.

• Token Limits: Token limits dictate the maximum number of tokens or words an AI model can generate as a response. Understanding the token limits of your model will help you shape your prompts accordingly, enabling you to manage the length and complexity of the generated content.

• Model Parameters (if using the API): If you're working with AI platforms through their APIs, it's important to understand the available model parameters. These parameters can enable you to adjust settings such as temperature and max tokens, allowing you to fine-tune the AI model's behavior and generate desired outputs.

If you invest the time to gain a solid understanding of generative AI and its capabilities, you'll be better equipped to craft prompts that deliver whatever you desire.

2. Practice writing more effective prompts

Writing effective prompts requires a blend of creativity, precision, and a solid understanding of your AI platform's capabilities. 

Here are some essential tips to remember:

1. Set a clear goal: Make your prompts well-defined, unambiguous, and clearly state the role or persona. In the case of ChatGPT, we can specify a role and area of expertise to set the tone and expectations for the AI's response. For example, "As an SEO expert, provide advice on optimizing website content for search engines."
2. Break down complex tasks into simpler subtasks: Divide a complex task into smaller, more manageable parts to help the model better understand your needs. For example, instead of asking for a full marketing plan, request specific elements, like "list three key components of an effective SEO strategy."
3. Provide context: Offer relevant context or background information related to the role or task. For example, "Our company specializes in eco-friendly products for environmentally conscious consumers."
4. State the task or question: Clearly state the task or question you want the model to address. For example, "Please explain how we can improve our website's presence on search engines"
5. Set constraints or limitations: Mention any constraints or limitations that should be considered while generating the response. For example, mention constraints, like "Suggest strategies that do not require a large budget."
6. Offer additional guidance: Provide further instructions or examples to guide the model towards the desired output. For example, give further instructions, such as "Focus on on-page SEO techniques like optimizing title tags and meta descriptions."
7. Specify format, tone, and style: Mention the desired format, structure, or tone in your prompt. For example, "Provide the information in a concise, actionable list with a friendly tone."
8. Use double quotes: Enclose specific phrases or instructions in double quotes to emphasize their importance. For example, "Include best practices for on-page SEO optimization."

So all together, that gives us a prompt that looks like this:

“As an SEO expert, provide advice on optimizing website content for search engines. List three key components of an effective SEO strategy for our company. As a company we specialize in eco-friendly products for environmentally conscious consumers. Please explain how we can improve our website's presence on search engines. Suggest strategies that do not require a large budget. Focus on on-page SEO techniques like optimizing title tags and meta descriptions. Provide the information in a concise, actionable list with a friendly tone. Include best practices for on-page SEO optimization.”

By following these tips, you can get more out of the AI model, enhancing your marketing campaigns and making your communication efforts more efficient and impactful. 

Remember, like any skill, prompt engineering improves with practice, so don't be afraid to experiment and learn as you go.

3. Use prompt engineering tools

The explosion in popularity of OpenAI’s ChatGPT model has led to the emergence of numerous third-party apps aimed at enhancing and simplifying prompt engineering for end users to help avoid pitfalls. They also provide additional resources, like methods for organizing and managing prompts, that go a long way in assisting individuals in designing effective prompts for AI models.

FlowGPT

FlowGPT is a collaborative platform that allows users to share and explore prompts specifically tailored for AI models like ChatGPT. This platform categorizes prompts across various domains like marketing and software development, allowing users to easily locate suitable prompts. Furthermore, FlowGPT enables interactive exchanges among users, fostering a sharing and learning environment for like-minded individuals.

AIPRM

AIPRM is a marketplace that offers 1-click prompts for ChatGPT, Midjourney, and DALL-E. It provides a quick and easy way to generate marketing content, such as blog titles, original articles, and copy for service pages. 

The platform has over 121,000 prompt engineers and also has a browser extension with curated prompt templates for SEO, SaaS, and more. The extension requires an AIPRM account and offers free and premium features, including custom lists, custom writing tones, and custom writing styles.

AI Prompt Genius

AI Prompt Genius' free browser extension takes it a step further, making it easier than ever to generate high-quality text from AI models. By providing locally synced chat history for convenient searching and offering a multitude of pre-designed prompt templates, AI Prompt Genius substantially simplifies the prompt engineering process for users. 

While all of these tools aim to make prompt engineering easier, it is ultimately up to you as the user to feed it the best information possible to receive high-quality results.

4. Human oversight is still needed

It’s a best practice with any AI model, whether that be translation, transcription, or facial recogntion, to keep human oversight in the loop. In other words, don’t take any LLM’s output at face value. Whatever piece of content you are creating should be reviewed by yourself or someone else just as you would working with your human counter-parts. AI is here to make our lives easier, not replace us (or so they say).

5. Avoid using confidential information

It’s important to remember that any data input into a public LLM like ChatGPT is not secure. This means you should avoid putting in any proprietary, client, or otherwise senstive information as data security nor confidentiality are guaranteed.

5 real-life examples of prompt engineering for marketers

For a more practical understanding of the power and versatility of prompt engineering, let's explore a variety of common marketing use cases. These examples cover a mix of common marketing activities:

1. Blog post topic ideas

Prompt:

Generate {Number} unique and captivating blog post topic ideas in the field of your choice based on my company name and product description. The topics should be specific and attention-grabbing, providing enough material to write a full blog post, and is SEO friendly. 

My company name is {Company Name}. 

The product description is {Product Description}.

The target audience is {Target Audience}.

Example:

This is an example for OtterAI, the popular meeting notes recording software.

2. Homepage title and meta descriptions for SEO

Prompt:

Act as an SEO expert tasked with optimizing the homepage of a company/product for search engines. Your client has provided you with the company/product name, a brief product description, and a primary keyword. Your job is to create a compelling title and meta description for the homepage that will both accurately reflect the content and appeal to potential visitors. Remember to keep in mind the target audience and the search engine algorithms when crafting your title and meta description. Provide {Number} results.

{Company/Product Name}

{Product Description}

{List of Keywords}

Example:

This is an example for a solar panel company.

3. Press release title & introduction

Prompt:

Act as a News Editing Expert, I will create a compelling press release title and introduction for your news story. Please provide me with the details of your press release, including the subject, company or product name, and a relevant keyword. Provide {Number} results.

{What is your Press Release about?}

{Company/Product Name}

{Keywords to Include}

Example:

This is an example for the launch of a new home automation product.

4. LinkedIn connection request

Prompt:

Using the best practices for LinkedIn connection requests, such as personalization, brevity, clarity on mutual benefits, and a distinct call to action, please generate a connection request. Use the following details:

1. Prospect's name: {Prospect's Name}

2. Mention how we're connected or how I found them: {How We’re Connected}

3. State a clear reason for wanting to connect that provides value for them: {Clear Reason for Connecting}

4. What action I want them to take after reading the message (call to action): {Desired Action}

Remember the total length of the output message should be ideally 140 characters or less.

Example:

This is an example with a mutual connection.

5. Google Ads writer (Two-Part)

For this example, we will use chain-of-thought prompting by first asking the LLM to tell use specifications and best practices for what we are trying to achieve–in this case, Google Ads.

Prompt 1:

List out Google Ads specifications for text ads, and concise best practices.

Prompt 2:

Using the given guidelines above for creating text ads on Google Ads, please design a compelling ad for {Company Name}. 

The ad should promote {Product Name}, which is best described as {Product Description}. 

Ensure that the ad includes relevant keywords, a strong call-to-action, and mention if we have any offers or discounts. Use concise language and ensure that the content is both appealing and relevant to our landing page. Please keep in mind the character limits for the headlines, display path, and descriptions.

The ad should look like:

Headline 1: ____

Headline 2: ____

Headline 3: ____

Display Path 1: ____

Display Path 2: ____

Description 1: ____

Description 2: ____

Example:

This is an example for a fictional cleaning supply company.

Final thoughts

As we've explored in this guide, prompt engineering is not just a smart addition to your marketing toolkit, it's quickly becoming an essential skill. The strategic application of zero-shot, few-shot, and chain-of-thought prompts can significantly transform the way you approach content creation, SEO optimization, PR activities, social media management, and ad campaigns. 

By gaining a strong understanding of the AI platform's capabilities, practicing effective prompt writing, and leveraging available platforms, you have the potential to make your marketing efforts more efficient, engaging, and impactful. 

Remember, mastering prompt engineering isn't about getting it right the first time, but about learning and refining as you go. So go ahead, experiment, and uncover the potential of AI in amplifying your marketing strategies. Happy prompting!