Everything You Need to Know About Writing Gen AI Prompts (with examples)

There’s no shortage of generative AI tools that you have access to. And while generative AI can be incredibly useful, it’s not a cure to all of your problems. It’s only as powerful as the prompts that you provide it. The robots are not coming for your job—because your expertise is at the core of what’s required to make gen AI truly work. But how can you use it to boost your productivity and streamline your workflow? 

It all starts with a good AI prompt. 

What is a gen AI prompt?

Prompts are the instructions or examples you enter into a generative AI platform to produce a response. They can come in different forms. They can be single sentences or paragraphs, or even contain multiple examples. 

So a prompt could be anything from “create an audit program for me” to “create a stakeholder engagement plan for my ESG program” to “write 3 different introductory paragraphs to a blog all about creating prompts for generative AI. Write it in Workiva’s brand voice and make it persuasive.”

How to write a gen AI prompt

Like we said, your humanity and creativity are crucial to working with generative AI. The model is powerful, but what and how you ask it to respond is where your subject matter expertise is required. Here are five tips to making a good generative AI prompt: 

1. Be specific

You have to clearly state your question or request. The model can’t infer or interpret your meaning beyond what it’s given, so longer AI prompts generally mean more clarity. The more precise and specific you are, the better the model can understand and provide a relevant response.

Be specific by mentioning:

• Topic 
• Length of response (number of sentences, words, or characters)
• Style (formal or informal)
• Structure (essay, blog, song, tight 5 for your comedy debut)
• Intent (who is your audience?)
• Output format (a bulleted list, paragraph, chart)

Let’s break it down with a couple of AI prompt ideas and examples. 

Less effective AI prompt example: “Create an audit program for me.”

More effective AI prompt example: “Ask me questions to help me create an audit program. You should ask questions until you have sufficient information about my audit program. Once you have sufficient information, create the audit program for me. Ask me the first question.”

2. Provide context

Make sure you give the model some background information or context related to your question. It doesn’t have life experience to draw from. Context helps the model understand the topic and provide a more accurate response.

Less effective example: “Write me a remediation strategy for a control failure.”

More effective example: “You are a governance, risk, and compliance (GRC) analyst. The GRC analyst works closely with other security analysts on projects associated with risk assessment and mitigation and tools and controls selection, and contributes with efforts that pertain to risk. Write me a detailed remediation strategy for an accounts payable control failure and how to mitigate this control failure.”

3. Ask single questions

Don’t overwhelm your new gen AI buddy! Gen AI performs better when responding to a single question or request. If you have multiple questions, ask them one at a time for more focused and coherent responses.

Less effective example: In the convoluted landscape of assembling the constituent elements that comprise an ESG report encapsulating the intricacies of scope 1 and scope 2 emissions, can you unravel the perplexing methodologies and underlying parameters driving the identification and amalgamation of salient data points, while simultaneously shedding light on the inextricable relationship between data accuracy and its reverberating implications on the transparency and credibility of the reported emission metrics?

More effective example: Question 1—what are scope 1 and scope 2 emissions in an ESG report? And question 2—how do companies typically report Scope 1 and Scope 2 emissions?

4. Use it to brainstorm

Generative AI helps when you’re facing the dreaded blank page and nothing exists except a blinking cursor. Now you have a safe space to bounce ideas around. If you want creative responses, you can explicitly ask the model to “brainstorm,” “think creatively,” or “imagine.” This can help generate more unique and unexpected answers.

Less effective example: “Will computers take my job, and what should I say about it in my ESG report?”

More effective example: “Please provide alternative approaches to writing the following prompt while achieving the same question and answer pattern:

Initial prompt: 'What are the key ethical considerations related to AI implementation in the workplace, and how do they influence ESG reporting practices?'

Additionally, feel free to brainstorm different formats or themes to make the question and answer interactions more engaging and diverse. Compare and contrast these formats, then ask me to choose one.”

5. Experiment and iterate

If you're not satisfied with the initial response, try and try again! Rephrase your initial prompt or provide additional information to get a better answer. Experimenting with different approaches can often yield improved results.

Less effective example: “Explain large language models to me.”

More effective example: “Explain large language models to me. Assume that I am five years old.”

What not to do with your gen AI prompts 

As we embrace the power of artificial intelligence to revolutionize our industry, it is crucial that we all prioritize the safe and responsible use of this technology. AI holds immense potential to enhance our productivity, efficiency, and decision-making capabilities, but it also comes with inherent risks you need to be aware of. 

Remember that gen AI is not perfect and may sometimes produce incorrect or biased responses. Dust off those critical thinking skills and verify the information obtained from the model through reliable sources. Be better than the machine and use your subject matter expertise!

Know the limits of gen AI

Generative AI is a generational innovation that can help boost your productivity and serve as a creative spark when you're getting started. But there are current limitations to what it can currently do. Here are a few things to keep in mind:

Consider other sources for more recent information

Large language models (LLMs) have a fixed knowledge cut-off—they don’t have access to the latest issue of the New York Times or know what’s trending on social media (or any real-time data). Don’t rely on gen AI if you’re looking for up-to-date information on current events, recent developments, or live data such as stock prices.

Expect limited ability to perform complex mathematical calculations

While gen AI tools can handle simple arithmetic operations—such as addition, subtraction, multiplication, and division—they may struggle with more advanced mathematical concepts such as averaging or statistical analysis. But while this isn’t currently possible, it likely will be in the very near future. 

Know what’s allowed and what’s restricted

Ensure the data you put into any AI tool is allowed via your organization’s policy. Customer data, personally identifiable information, or information that’s covered under a non-disclosure agreement are all examples of data likely considered restricted. Never put non-public data in external AI/ML tooling without explicit approval from your legal and information security teams.

Watch out for hallucinations

AI can produce “hallucinations,” which are instances when it generates unexpected, untrue results not backed by real-world data. AI hallucinations can be false content, news, or information about people, events, or facts. We’ll reiterate, you should always verify any AI-produced content.

We’re confident that responsible use of generative AI is going to supercharge your productivity and amp up your creative output. 


Take a break from reading about generative and join the discussion virtually! Check out the Gen AI session, all Amplify 2023 sessions, until October 31st! Register here.