Write better AI prompts: A 4-sentence framework

For non-developers, AI can seem like a black box. The concept of AI prompting sends a chill down our spines, especially when we’re bombarded with complicated examples and techniques. Multi-agent prompting? Prompt chaining? ReAct prompting? Sounds too much like the CompSci 101 course we failed in college. 

But according to Ethan DeWaal, Asana’s AI Program Manager, AI prompting doesn’t have to be complicated. Instead, DeWaal says to think about AI like a golden retriever: “It wants to please you, but you have to be specific about what you want.” 

And being specific doesn’t require anything complicated. You just need four simple sentences. 

How to build intelligent workflows for business agility

Get expert tips to build flexible intake processes that can adapt as your business does. Plus, leverage AI to pivot quickly and make smarter prioritization decisions.

Watch the webinar

How to create prompts for anything

When people start using large language models (think Chat GPT), they often try simple commands first—even for large tasks like writing a blog post or creating a brief. “This is like surprising your partner with the news you got a job offer that’s 3,000 miles away,” says DeWaal. “With no lead-in or background info, they’re caught totally off guard.” 

Just like your partner, AI needs context so it knows how to react. This 4-sentence framework helps you provide that context.  

The 4-sentence framework

DeWaal boils prompts down into four basic components: persona, goal, task, and context. You can use the acronym PGTC to help you remember. For most prompts, you only need one or two sentences for each component of the framework: 

• Persona: “You are an expert at _____”

• Goal: “Your goal is to _____”

• Task: “Your task is to _____” 

• Context: “Here is all the context you need” (think formatting expectations, things to avoid, things to highlight, etc.) 

And the best thing about this prompt framework? You can use it for anything, from content creation to research. Here are some examples to get you started: 

Example: Conducting customer research

Imagine you’re a customer service manager using AI to help prepare for calls. Here’s how you could implement this 4-step framework: 

• Persona: “You are an expert at customer success.” 

• Goal: “Your goal is to help Asana customer service managers prepare for calls.” 

• Task: “Your task is to research the person & company they are meeting with, using the Search with Bing plugin.”

• Context: “Your response should highlight recent news about the customer that the customer service manager should know, some good questions to ask regarding potential expansion, and share details that will help the customer service manager on their call.”

Example: Creating a campaign brief

Imagine you’re a marketer using AI to draft a brief for your upcoming campaign. Here’s how to adjust the framework to fit your needs: 

• Persona: You are an experienced marketing strategist.

• Goal: Your goal is to help marketers prepare comprehensive and compelling campaign briefs for their marketing initiatives.

• Task: Your task is to provide a detailed framework and prompts to guide marketers through the process of crafting a well-rounded campaign brief. The brief should cover key elements such as target audience, objectives, messaging, channels, and success metrics.

Context: Your response should include prompts and questions that encourage marketers to conduct thorough research on their target audience, define clear and measurable objectives, develop a compelling central message and supporting messaging pillars, identify the most effective channels and tactics for reaching their audience, outline a content plan, and establish key performance indicators (KPIs) to measure success. 

“Cheat codes” for better prompts

With prompting, a few simple strategies can significantly increase the quality of your outputs. “You don’t have to be a prompting genius,” says DeWaal. “You just need to do a couple things.” 

Provide examples

Examples are a great teaching tool, especially for AI. They tell AI exactly what you want, so it’s more likely to produce a quality result. 

If you’re creating content with AI (like a blog, press release, or email), use an existing piece as an example. Add it to the context section of your prompt, and put quotes around it. Before your example, tell AI something like “here’s the template I want you to follow.” This gives the model a better idea of the style, length, and formatting you want. 

Give feedback, and go with your gut

Giving feedback is an important part of the process. “Whatever your first gut reaction is, go with it,” says DeWaal. When you’re early in your journey, it’s normal to have a strong reaction to what AI produces. You might be blown away, or the results might seem too long, too robotic, or too formal. 

Whatever your perception is, tell the AI model directly. “This is where you can start shaping, editing, and adding additional context,” says DeWaal. 

Speak to AI like a human

The solution is usually a lot simpler than you think. Instead of trying to craft the perfect prompt, try just describing the issue you’re having and what you want instead. “I’ve had situations where I was blocked for three days because I couldn’t figure something out with AI,” says DeWaal. “Then I came back and just described the issue more clearly, and it worked.” 

If AI isn’t giving you the results you want, take a step back and try direct communication. Say “this is what I need from you, and you’re not doing it.” 

Model selection matters

Not all AI models are created equal. “The number one reason why people have a bad experience with AI is they’re using last year’s version,” says DeWaal. Instead of settling for something like Chat GPT 3.5, try using a model with the same (or better) quality as Chat GPT 4. 

Work smarter with AI

AI gets a lot of hype, but it’s worth learning about. According to a recent study from Boston Consultant Group and Harvard Business School, using AI produces measurable improvements in performance. In the study, consultants with AI access completed 12% more tasks and finished 25% faster. And the kicker? Their work was rated to be more than 40% higher quality by human evaluators. 

With a few minutes of reading, you can start partnering with AI to produce better work—faster. Your future self will thank you.