Prompt engineering is ultimately a skill that can be learned and that you can get better at over time. In this blog post, I’d like to dive into some of the mindsets and practices that will help you improve your interactions with generative AI, and lead to amazing outcomes that you wouldn’t be able to arrive at on your own. At the end of this post, I provide an example of some output generated using Gemini for Workspace (formerly known as Duet AI for Google Workspace), but know that these tips can be applied to any generative AI tools or programs you’re using. Let’s get started!
Step 1: Clarify your intention
The first step to writing an effective prompt is to think about the outcome you want to achieve. What do you want to produce? A one-paragraph description of a product? A blog post that demystifies a technical process? Your aim at this stage should be to provide the level of detail necessary to generate a satisfying output, without providing so many details that you might as well have written the entire thing yourself.
There are a lot of reasons why someone might gravitate toward using generative AI. Maybe you need a short paragraph on a topic and the words aren’t coming. Or maybe you want your incomplete notes to sound more professional. You might even turn to GenAI when you’re exploring a topic you’re less familiar with, and you want to augment your understanding.
Step 2: Get personal
There are a couple of ways to improve your prompts using Gemini for Workspace. The first is to establish a persona. In other words, tell the AI who it’s supposed to be. Is it the Chief Marketing Officer of a Fortune 500 company? A tech blogger with a sardonic sense of humor? Or maybe it’s simply an extension of your own persona–you know, a stylish, sophisticated thought leader who likes to read SADA’s blog.
Step 3: Set the tone
Once you’ve established your persona, think about the tone you wish to convey. Are you super excited? Clear-eyed? Skeptical? Discerning? At this stage, it’s all about the adjectives, the more specific the better.
Step 4: Define the structure
Next, provide details that will help define the structure of the output. One great prompt I saw recently included the message, “Design a learning program, expressed in the style of executive business writing, that includes a course description, course objectives, dates and times for the course, and topics to be covered.”
Keep in mind, you’re really telling the system exactly what you’re looking for. I tend to think about it much the same way we learned to write essays back in third or fourth grade: opening introduction, two or three body paragraphs, and a conclusion. Remember when we learned how to present information in an orderly fashion? Think about submitting a prompt as giving the AI an assignment.
Another way to improve your prompts is to provide headers and subheaders detailing the topics you want to cover. Don’t hesitate to spell out what you want these topics to be in plain language.
Step 5: Keep trying
One of the biggest mistakes I see people make when submitting prompts is being impatient and expecting the perfect output right away. GenAI is still young and learning and it’s going to continue to get better. Be aware that you may want to craft prompts in a different way depending on whether you’re writing something from scratch or you’re poring over data or trying to create spreadsheets. So you have to do a little bit of the critical thinking in your prompts.
Other things to consider when drafting GenAI prompts
The importance of personas in prompt engineering
I think not providing a persona is a huge mistake. You’re going to get much stronger results if you can be more specific in the type of perspective or point of view that you want your output coming from. Just think about it–creating prompts that use personas is a way to use imposter’s syndrome to your advantage. You’re literally pretending to be someone else in order to achieve better quality results.
Welcome the oddness
Just for fun, I decided to use GenAI to write something a little fantastical for my team’s amusement, so I asked Duet AI to write an email campaign about monster trucks raining from the sky. The email went on to suggest that if you were lucky enough, you could in fact capture one of these raining monster trucks and take one home to be your pet. Creativity is about creating unexpected connections, and giving yourself a wacky assignment like this might help you get your own gears turning, too.
How to get better at writing generative AI prompts
It takes practice, patience, and making little adjustments to your prompts, through rounds of submission, so that you can notice the effect these subtle shifts have on your output. Add an adjective or an adverb, and re-generate the results. Add a few words for clarification. Change the persona a bit. You’ll start to notice how making a tiny change can have a tremendous impact on the final result.
Let’s say you’re a leader in your organization, it’s the beginning of the year, and you’re trying to inspire your team. You want to write an email saying that these are our key areas of focus. You’ve gotten your KPIs from the org, but you need your team to rally and you want to write an inspiring email. And maybe crafting that kind of message isn’t really your strong suit as a leader. AI can definitely help you get there.
I’m also seeing a lot of instances where customers are using GenAI to write blog posts like this one. In my case, I used Google Meet to produce a transcript of my spoken responses to a series of questions, then used Gemini for Workspace to summarize and organize my answers.
If you’re reading this, you’re obviously seeing the end results. It’s important to understand that as owners of our work, we are responsible for the ultimate output and what we send out into the world. It’s on us to validate, check, clean up, and apply our own critical thinking to everything generative AI provides. GenAI is valuable as a way to write something from a slightly different perspective, give us new ideas to work with, or as a jump start to the first draft.
I like to remind folks that, when you’re working with Gemini for Workspace and using prompts to create content, you shouldn’t think of this as a shortcut but as a starting block. You probably want to get to the end result faster, but if you take a shortcut, you might miss something. If you’re not rereading what is provided to you for accuracy, adding your own voice, and applying critical thinking, you’re going to end up with problems.
That starting block is what you’re pushing off from to give you momentum. Ultimately we’re trying to get to that same end result–a great letter written to your team, an inspiring email about a new marketing campaign, a spreadsheet that’s going to pull some critical data, or some top-notch thought leadership–but it is still our responsibility to put in the work of learning how to engineer an effective prompt.
A prompt is just a tool, and it doesn’t eliminate our responsibility for effective communication. When you stretch your prompting muscles, try a hundred times. It’s not your hundredth prompt that wins the day. It’s the 99 that came before that got you closer, step by step, to fully expressing your ideas.
Using GenAI to summarize content
I decided to run my transcript of this blog post through Gemini for Workspace and prompted it to pull out various points I made. Here’s the result. And remember, this is a great exercise for any copy that you might be working with, as well. Getting GenAI to summarize your work might prove to be a helpful method for clarifying your ideas as you develop your content.
Prompt: Help me summarize the key tips for creating good prompts
Key tips for creating good prompts:
- Provide your persona: Specify the perspective or point of view you want the response to come from.
- Use adjectives and adverbs: Add color and tone to your prompt by using descriptive language.
- Provide structure: Outline the desired format or structure of the response.
- Be patient: Understand that AI is still learning and improving, and it may take time to get the desired results.
- Practice and iterate: Experiment with different prompts and observe the impact of small changes.
Prompt: Summarize prompt engineering
Prompt engineering is the process of designing and crafting prompts that are used to instruct AI models to generate text, images, code, or other creative content. By carefully crafting prompts, users can guide the AI model toward the desired output and achieve more specific and relevant results.
Key Elements of Prompt Engineering
- Persona and Tone: Specifying the desired tone and perspective of the generated content.
- Categories and Subheaders: Organizing the prompt into clear sections for better understanding and control.
- Clarity and Communication: Ensuring that the prompt is clear, concise, and communicates the desired outcome effectively.
Common Mistakes to Avoid
- Lack of Patience: Expecting perfect results immediately without allowing the AI model to learn and improve.
- Neglecting Personas: Failing to provide a clear persona or point of view for the generated content.
- Drastic Changes: Making significant changes to the prompt instead of gradually refining it.
Tips for Writing Effective Prompts
- Practice and Patience: Continuously experiment with different prompts to refine and improve results.
- Targeted Adjustments: Make small adjustments to the prompt to observe their impact on the output.
- Critical Thinking: Apply critical thinking when evaluating the AI-generated content and making necessary edits.
Value of Prompt Engineering
- Messaging and Inspiration: Crafting compelling and inspiring messages for various purposes.
- Meaningful Feedback: Generating personalized and constructive feedback for individuals.
- Summarization and Synthesis: Extracting key points and summarizing long-form content.
- Prompt engineering is a tool that requires practice and critical thinking to achieve the best results.
- It serves as a starting block to accelerate the content generation process, but the responsibility lies with the user to validate, edit, and add their own voice to the output.
- Prompt engineering should be approached as a collaborative effort between humans and AI to produce high-quality and accurate content.
Prompt: Summarize common mistakes that people make when submitting a prompt to gen AI
The common mistakes people make when submitting a prompt to a generative AI are:
- Thinking of prompting as a shortcut: It should be considered a starting block rather than a shortcut, as shortcuts can lead to inaccuracies and errors.
- Not editing or re-reading the AI’s response: It is important to review the AI’s output, check for accuracy, and add a personal voice.
- Not taking responsibility for the final product: While the AI can generate content, it is still the user’s responsibility to ensure the quality and accuracy of the final product.
Getting started with Gemini for Workspace
Now is an excellent time to roll up your sleeves and start experimenting with Gemini for Workspace. Remember, learning how to use GenAI requires a different approach and set of skills than mastering other software tools. It involves a lot of experimentation, trial and error, necessary false starts, and iteration over time. We’re now in the age of emergent computer behavior, which means that tinkering, wandering, and iteration are all part of your skillset.
Get started with Gemini for Workspace by scheduling a discovery call with SADA’s AI team. They’ll be excited to hear what you have in mind for your organization, and ready to equip you with the resources to establish a thriving GenAI practice.