AI Prompt Engineering 101: 4 Tips for Fruitful Conversations with Robots
Good conversations share the same key ingredients. They begin with a baseline understanding of who the participants are. They are two-way streets, full of follow-up questions, clarifying statements, and listening. And they must maintain a logical thread insofar as both parties agree on the logic.
In the end, a fruitful conversation should lead to some truth that makes each person’s life better.
While AI tools are not people (yet?), these same ingredients make for better results when prompt engineering with your favorite robot.
What is prompt engineering?
Insights, delivered.
Prompt engineering is a fancy term for the instructions and questions you feed an AI tool. Whenever you input something in the dialog box—e.g., What are the three largest medical associations by total members? Compare the top 3 robotics programs in Texas. Make this heading title case. Where do babies come from?—you are engineering a prompt.
Since the generative AI Big Bang exploded a little more than two years ago, we’ve come to understand the technology better and refine the way we interact with it. Prompt engineering has matured alongside the platforms themselves to yield better results.
We now know how to get more relevant, nuanced outputs. It also means there are some simple, tried-and-true tactics you can use to become a better prompt engineer. Here are four.
1. Give the robot a job
The Large Language Models (LLMs) that form the core of generative AI tools are trained on writing samples from all over the content universe. Newspaper op-eds drafted by subject matter experts. Comment threads on car repair forums. Blogs from marketing agency websites.
Without guidance, AI platforms will regurgitate a remix of language from every nook and cranny of its dataset. But by picking a role or job for the platform, you help it narrow down the pieces of its dataset it pulls from.
For example, start off your prompt with a simple statement such as:
Assume the role of a university admissions counselor.
Pretend you’re a digital marketing strategist for a medical association.
You’re a communications director for a nonprofit in a major U.S. city.
2. Give the robot some context
Context is king for AI chatbots. It’s important to build on the job or role you assign by providing more context about your expectations for the outputs. Entering the assignment parameters into a prompt is a great way to start.
In marketing, assignment parameters often include details such as channels, intended audiences, keywords, and funnel state, to name a few. Work these into a prompt to maximize the quality of the output. To build on the first example from above:
Assume the role of a university admissions counselor.
Write an outline for a five-email series targeting prospective high school juniors and seniors in your state who participated in a statewide engineering competition.
The email series should highlight the benefits of enrolling in your university’s engineering program. Those benefits include the exciting projects its students undertake, job placement metrics, and extensive financial aid options.
Include A/B subject lines and email headers in the outline.
Important reminder: AI copy will always need human oversight. AI should be used as a tool for ideation or formatting that you can then transform.
3. Tell the robot how you want it to sound
A key improvement in AI platforms in the last two years is their ability to detect and reproduce certain tones and personalities. That doesn’t mean they’re capable of writing laugh-out-loud stand-up comedy sets or sarcastic quips for sitcoms. Nor does it mean they can perfectly mimic an existing personality or voice, at least not any of the off-the-shelf platforms.
What it does mean is that AI tools can inch you closer to your brand voice. It means no more utterly bland, lifeless outputs. It means less time spent making an output sound more human. Continuing the example from above, that could look like this:
The subject lines and email headers should sound witty yet professional. They should also become more urgent as the email series progresses. Some words or phrases to consider include: “your future,” “act now,” “cool projects.”
You can also exclude certain words or phrases, too.
4. Keep the conversation going
An AI chatbot is not a one-and-done tool. Even the most seasoned prompt engineers can’t churn out anything resembling a final product without a lengthy back-and-forth. You have to keep the conversation going.
Maybe the AI tool doesn’t know that best practice for email subject lines is 40-50 characters, so it produced 90-character behemoths. Maybe the chatbot’s idea of “witty” is simply the use of the word “witty.”
Sometimes, getting a subpar output helps you, the prompt engineer, better understand what it is you want. Is “lighthearted” a better word than “witty?” Is an admissions counselor the best-suited role, or should you try “upperclassman mentor?”
The key takeaway is to make adjustments. Add more or new context as you go. Experiment. Robots are tireless assistants, they don’t mind.
Our favorite robots
Of course, not all robots are created equal. And, not all robots have the same strengths. But if you’re struggling to select a good AI tool for your organization, we’ve got some recommendations.
Note that with each of the platforms below, we’ve selected the paid version of the plan that keeps your inputs out of the training data. This is a crucial component of protecting your organization’s proprietary data and your audience’s personal information.
Writer, Teams Plan. We like Writer because its proprietary language model does not use copyrighted content in its training data. That’s an ethical win today, and possibly a legal win down the road. We also love its style guide features, which are fantastic for maintaining a consistent brand voice.
Anthropic, Team Plan. We like Anthropic’s Claude because it’s a powerful, multi-modal platform. Its outputs are top-of-class, and it also has coding capabilities.
ChatGPT, Plus Plan. ChatGPT is not just the biggest name in generative AI, it’s still arguably the highest-performing. We especially like the Plus plan because it includes access to DALL-E, which is OpenAI’s image generation tool.
Looking for more AI resources? Take a look at our Insights for everything we’ve published so far.
