Storytelling Meets AI: How Product Managers Can Elevate Generative AI through Contextual Insights

The hype cycle around generative AI has passed, but curious and forward-thinking professionals know something fundamental has changed. They're left sorting through the jungle of instant influencers and keynote speeches, sifting for the signs of real, enduring value.

Better problem shaping will become the enduringly valuable skill of the generative AI age.

Thousands of fortune seekers rushed to become overnight Prompt Engineers during the height of the buzz. As cited in this June Harvard Business Review article, no less than the World Economic Forum and Open AI's CEO heralded the burgeoning discipline as the "job of the future" and a "high-leverage skill." If you're wondering, prompt engineering is the search engine optimizer of the AI age - an expert divining the structure and language that will solicit more relevant results from generative AI.

Indeed, learning to prompt ChatGPT is a shift from writing good search engine queries. However, author Oguz A. Acar effectively pierced the bubble. AI models will adapt to fit users' ways of asking for help or information (that's part of what sets generative AI apart.) There is no "magic prompt" that will find gold in this AI gold rush.

Phrasing and structure may change, but Acar posits that better problem shaping will become the enduringly valuable skill of the generative AI age.

As a product manager and former journalist, I read the article with more than a little gratitude that my past has prepared me well for my future. Both disciplines tap the same skillset: the ability to describe the essence of a problem out of many facts, stats, opinions and biases. It turns out that ChatGPT also loves to read a well-written story, and can help you become a better problem framer.

What's important context that ChatGPT needs to read so it can work for you? (disclaimer, I used ChatGPT 4.0 as a starting point for the following tips.)

1. Historical Context

AI models thrive on understanding the backdrop. If you're analyzing market trends or predicting future behaviors, the AI needs to know what has transpired before.

Tip: Always give a brief of past relevant events or trends before diving into the current problem. This equips the AI to predict more accurately. These events can be general and historical, like relevant news articles. If you're describing past events within your company or product, remember to describe the company or product first before framing the events.

2. Cultural & Societal Nuances

Generative AI models, despite their vast training data, may lack understanding of current cultural shifts or specific societal nuances. Always keep in mind that ChatGPT only knows events up to 2021.

Tip: If your prompt revolves around a particular culture or recent societal event, offer a concise overview or any relevant cultural sentiments to guide the AI's response. Specifically ask the model to consider or change its output to account for those described events.

3. Stakeholder Perspectives

Every problem or scenario involves multiple stakeholders, each with their viewpoint. Identifying and conveying these perspectives can greatly influence the output.

Tips:

• Detail out the major players involved in your prompt and their possible motivations or biases.
• Note the strength or influence of a particular stakeholder's opinion. The model can weight its description of the problem accounting for that dynamic.

4. Leading Facts

Vague prompts can lead to general outputs. Lay out the specific facts and stats that inspired you to query the model. The AI model may reframe the same information in an unexpectedly helpful way. It may oddly interpret your facts, but your correction helps you reflect on what's important to you.

Tips:

• Use precise figures, dates, or specific events where possible. Instead of saying "recent economic downturn," specify "the 2020 economic downturn due to the COVID-19 pandemic."
• Lay out some interesting facts you've gathered and examples of articles that have gone viral in your domain before. Give the model important information instead of just telling the model what's important.

5. Desired Outcome

AI models don't inherently know your end goal unless specified. If you have a particular format, tone, or direction in mind for the AI’s output, state it explicitly.

Tip: Along with your main prompt, provide guidance on what you wish to achieve. For instance, "Provide a summary in a neutral tone" or "Suggest solutions with a focus on sustainability." If the first try doesn't hit the mark, ask the model to re-write with a different goal.

6. Address Potential Biases

Every AI model, no matter how advanced, can carry biases based on its training data. By addressing potential biases in your prompt, you can guide the model towards a more balanced response.

Tip: If you're discussing a sensitive issue, instruct the AI to consider multiple viewpoints or to avoid common stereotypes. Even when asking for code or technical information, try asking the model to re-write the article as if it were a person with different experience or education, or from a different gender or culture. It's innovation on tap.

7. Risks and Constraints

Tell the model up front what it or you can't accept. The model has access to a huge body of information - it may draw from unhelpful information in forming its reply. At best, it will cause to spend more time correcting the model. At worst, you'll miss the subtlety in the model's bias and find out later that it (and therefore you) went off course.

Tip: Describe what statements or types of facts a "bad" reply would include.

How do you tell AI a story?

I'd love to hear more about the types of information you include when prompting a generative AI model. What helps you get more value?