An Intro to Prompt Engineering for B2B Marketers

By Maria Geokezas, Chief Operating Officer at Heinz Marketing

If you’re like most marketers, you’re contemplating how A.I. will change the way you work.

From McKinsey Global Institute Chairman and Director James Manyika:

“It’s natural to wonder if there will be a jobless future or not. What we’ve concluded, based on much research, is that there will be jobs lost, but also gained, and changed. The number of jobs gained and changed is going to be a much larger number, so if you ask me if I worry about a jobless future, I actually don’t. That’s the least of my worries.”

The truth is, there’s no way to know with certainty how A.I. will shape the workforce moving forward. In fact, some research indicates that new job creation will likely offset any job displacement from advancements in A.I.

The World Economic Forum’s “Future of Jobs” report estimates A.I. may displace around 85 million jobs but may create as many as 97 million.

That means instead of jobs getting eliminated; they’ll be transitioned—the same way they have throughout history.

Today, the transitional skill to learn is prompt engineering, or in other words, learning how to communicate with A.I. in a way that results in usable assets, not wasted time.

But writing a functional prompt isn’t as straightforward as it sounds.

So below, we’ll look at the must-have components of a good prompt and how to layer those components to generate ready-to-use outputs.

The Must-Have Components of a Prompt

Maybe you’ve been reading up on prompt engineering or attempting to teach yourself through hands-on practice, but if you’re like many marketers, you’re not impressed with the results.

That’s because while A.I. is fast and knowledgeable, it lacks a marketer’s intuition.

Simple, single-line prompts don’t provide enough background. Rather, proven prompts contain five main components.

1. Context
2. Examples
3. Format
4. Parameters
5. Instructions

Every marketing prompt should have context—or the who and why components. Adding a target persona or goal to your prompt helps the A.I. model narrow down which resources are useful and which are not. Setting the context helps prevent the model from going rogue and incorporating irrelevant information.

Another way to provide context is through examples. A language learning model can catch on to language, voice, and tone patterns when provided with a series of similar materials to imitate. That’s why you need to feed the A.I. model multiple examples of what you do or, in some cases, don’t want.

Moreover, if you want to use A.I. to generate ready-to-use content, you’ll need to provide clear format and parameter directions. Do you want a 50-word social post or a 1,000-word blog post? Should it stick to proven themes, or do you want to test the model’s creativity?

Think of the format and parameter directions as guideposts for the A.I.

Finally, are the instructions every prompt must contain. The instructions round out your prompt and include commands you didn’t introduce in the prompt previously— and sometimes even reiterate the crucial elements listed early in the prompt if it’s particularly long or complex. Instructions can include details such as the writer persona you want the A.I. to use, the delivery style, and the core ideas of the piece.

When it comes to prompts, don’t be afraid to over-communicate. There’s a reason why generic prompts produce generic content. If you simply prompt A.I. to “write a 500-word blog post on project management,” don’t expect to be impressed with the outcome.

How to Chain Components for the Best Results

It’s important to note that before you combine different types of prompts, you should train the A.I. with simple requests. Otherwise, you risk confusing the model and having to restart your session.

You can do this by asking the A.I. model to complete background research on your target audience and verify it matches your research to iron out any miscommunication before progressing to the next step.

Once you’ve refined the results and you’re confident in the primary outputs the A.I. has provided, you can begin to chain your prompt components.

Technically, you can layer your components in any order. However, experts have discovered some combinations work better than others.

Since A.I. models are susceptible to recency bias, the information at the end of your prompt will guide the model the most. For that reason, experts recommend you should test placing your instructions at the end of your prompt like this:

I want you to write copy encouraging our audience of enterprise HR representatives to download our recent ebook.

Here are two examples of successful sales emails [Insert examples]

The emails need to be between 200 and 400 words.

Acting as a CEO, write three persuasive emails like the examples. Write them in a conversational tone with a CTA to download the ebook at the end of the email.

In the prompt above, the chain goes context > examples > instructions. That way, the model won’t get distracted by the examples and possibly forget the word limit or CTA.

As a rule of thumb, order your prompt components like you would if you were training a junior marketer on a new task—provide an overview, give a few examples, and wrap it up with clear delivery instructions.

After you’ve trained the A.I. with proven prompts, you can use questions to explore new angles.

For example, by asking the A.I. how it would change its outputs if it were creating content for a different audience or platform, you can significantly reduce the time it takes to version your campaigns.

In this way, A.I. is more of a tool, not a takeover—embrace it.