It’s one thing to write AI prompts, and another to actually get the results you want. That’s why we’re excited to launch Prompt School, where you can access lessons, guides, and resources to use generative AI more effectively.
To kick things off, here are 7 tried and true best practices.
👀 Are you a visual or auditory learner? Watch the full session here, led by Copy.ai's Prompt Engineer, Anna Bernstein.
Let’s dive in!
Let’s pretend you’re training a new employee on a complex process—like configuring your CRM or building a new lead scoring model. As you’re guiding them through each step, they stop you and ask, “Wait, so what do you want me to do here?” Would you find another way to explain things to them, or immediately decide they’re incapable of completing the task?
AI operates in a similar fashion—only instead of requesting clarification, it simply generates the wrong output. When this happens, people often mistakenly conclude the prompt is beyond AI’s scope of capabilities. However, sometimes all it takes is an alternative or additional phrase to get the right response.
In the world of generative AI, synonyms are your best friend. There are infinite ways to describe sales and marketing prompts, and you need the right combination of words for AI to understand what you’re looking for.
Instead of giving up on your first attempt, try rewriting. If you get stuck, you can even ask AI for suggestions, like:
Synonyms are an invaluable addition to your prompting toolkit—but there’s a catch. It’s important to use the same label for the same element throughout the entire prompt.
For instance, if you’re asking AI to generate email copy, don’t refer to “recipients” in one section of your prompt and “prospects” in another. AI gets confused by inconsistencies, and outputs may be more random or off base as a result.
It’s equally important to limit each label to a single element. Prompts should clearly define the desired output, and repurposing labels makes it difficult for AI to understand exactly what you’re looking for.
Although mixing and matching labels won’t necessarily cause an error message, output will be a lower quality. Much like a confused human trying to figure out if they should address their colleague as Assistant Regional Manager or Assistant to the Regional Manager (for all of The Office fans out there), AI will usually skimp out the details if it doesn’t know what something means.
Now, notice the difference in the output below after changing just a single word in the prompt:
Write a job posting including the following information:
Job title: Senior Cheese Taster
Job description: Testing out cheese at our factory to see if it’s yummy or not
Job location: Siberia
Job requirements: Loves cheese. Lives in Siberia.
The desired format of the output is immediately clear, while job description is a neatly defined sub-category containing additional information. The result is far more detailed and aligned with the original request.
Causative language helps AI understand causative relationships—cause and effect—between elements (or words) in your prompt.
To get you started, let's break "causative language" into two categories: integrative and transformative"
Integrative language connects ideas and concepts, making it easier for AI to understand how one thing can cause another thing to happen.
Use integrative language when AI is ignoring an element of your prompt.
For example, AI will signal you to use more integrative language by ignoring a particular element, which indicates a failure to connect it with the rest of the prompt.
Integrative language is also necessary for AI to use information from one element to modify another—like rewriting an email or blog post based on feedback.
Integrative phrases can include:
On the flip side, AI can also disproportionately focus on one particular element, or repeat your description verbatim. This means it’s time to bust out the transformative language and clarify the intent of your prompt.
If you feel AI is missing the point of your prompt, it likely requires clarification. You can do this by replacing general verbs with more specific actions—for example, “Take the concepts from this description and reimagine them as a song” instead of simply “Turn the description into a song.”
Transformative language can also be used to provide additional context, or elaborate on previous explanations. Some handy transformative phrases include:
This helps AI understand the reason behind your prompt, so it can generate more accurate responses.
Pretend that you’ve just received a pair of magical sunglasses, and whenever you put them on, you can time travel. You’re allowed to visit any time, in any place, and the sunglasses can be any color—except yellow.
What color sunglasses are you picturing?
If the answer’s yellow, you may have more in common with AI than you realize. When you tell AI not to do something, it has a tendency to focus on the exact thing you want to avoid.
Fortunately, there’s an easy fix: using positive language. It’s as easy as swapping instructions like “not too formal” for “fun and casual”.
Have you ever been trapped in a conversation with someone who takes forever to get to the point—so long that you forget what they were even trying to tell you in the first place?
AI is the same way. Meandering verbs get in the way and distract from the main purpose of your prompt. Whenever you’re writing, be on the lookout for word clutter, and cut down anything that doesn’t need to be there.
Sometimes this means getting creative and phrasing your prompt in a more succinct way. For example, “give more information and details about this” can be condensed to “elaborate on this.”
A good prompt isn’t just clear, but concise.
With AI being, well, artificial, it makes sense to address it more like a search engine or Alexa than an actual person. However, while AI can do a lot of things, it still can’t read your mind.
A prompt that lacks context is no different than assigning someone a bunch of tasks without any direction—it may take several tries to get the result you want.
The best way to generate a good output is by taking the time to ensure your prompt contains clear, relevant information from the start.
So how do you know when you’ve included enough details?
One way to tell is with the sheet of paper test. If you were instructing a person instead of AI and handed them a piece of paper with the prompt written down, would they know what to do? If you think they’d have follow up questions, that’s a good sign that your prompt needs more details.
AI can get really meta—especially when it processes self-generated synonyms to better understand your prompts. You can also leverage AI-enabled features like Copy.ai’s Prompt Improve, which makes automated, real-time prompt improvement suggestions right there in the chat.
It’s no secret that prompt engineering is part art, part science, and it takes practice to get into the groove. However, the ability to refine prompts as you’re writing them helps ensure that the content you generate is high quality. Beyond just making recommendations, however, Prompt Improve actually reformats your prompt to provide additional details and context for a better output.
The best prompt looks wildly different depending on the circumstances. However, these best practices are universally applicable, no matter where your prompting takes you.
Want to continue learning? Test these prompts tips in Copy.ai today!