Get Creative With Prompt Iterations

Honing your AI prompt skills is one of the fastest ways to start realizing value from frontier models such as ChatGPT or Claude. Prompts are — as of this writing — a primary means of interacting with GPT-class AIs and a necessary part of using AI to work faster. When done well, crafting prompts can also expand how you think about a situation in ways that illuminate opportunities for innovation. This happens in part because writing a good prompt requires thinking creatively about the problem to be solved. Below, you’ll find some specific strategies for doing just that.

Before applying problem solving strategies, though, there first must be a defined problem to solve. Some important aspects of understanding a problem in ways that can help you refine prompts that help solve it include breaking down the problem into smaller sections, some examples of (at least approximate) success, knowledge of the constraints on the situation, and a recognition of what insights would be most useful.

There are many ways to break down different aspects of a problem statement, some as broad as the 50–150 jobs identified in the Jobs-To-Be-Done framework (JTBD) and some as tightly scoped as a UX-style task analysis exercise. A problem statement can also be reframed with respect to the people, process(es), or product(s) associated with its target. Breaking down the problem statement gives you more directions in which to experiment with prompts. For example, if I’ve identified a functional, social, and an emotional job-to-be-done, I can reframe a prompt using each lens. Here’s a breakdown of three JTBD by the pitch deck:

Functional job: Persuade audience to invest in the pitched idea
Social job: Demonstrate reliability and intelligence of presenter to audience
Emotional job: Evoke satisfaction from a well-presented, well-framed pitch deck

And here are some example prompts:

Functional: Analyze this pitch deck focusing on its effectiveness in persuading the audience to invest. Highlight elements that strengthen or weaken the deck’s ability to convince investors.

Social: Evaluate how this pitch deck portrays the presenter’s reliability and intelligence. Identify aspects that enhance or diminish the presenter’s credibility in the eyes of the audience.

Emotional: Assess the emotional impact of this pitch deck. Pinpoint elements that contribute to audience satisfaction and engagement, as well as areas that might detract from a positive emotional response

This is just one example using one framework, but there are many possibilities for applying this strategy to think more creatively about the problem at hand while expanding your prompt repertoire. (If you explore some, I’d love to hear about it in the comments!)

Another compendium useful for prompt troubleshooting is a set of successfully resolved example problems. Identifying examples of successful resolution allows you to take advantage of one- or few-shot prompting, in which example cases are given to the AI as part of the prompt. The prompt also includes instructions to apply the patterns detected to the remaining scenarios. Identifying these successfully resolved similar scenarios can help highlight adjacent possibilities, further expanding creativity and your prompt library.

Here’s an example prompt for writing a SQL query:

Here are two SQL queries that solve common database problems:
Example 1:
SELECT event_name, total_raised FROM fundraisers ORDER BY total_raised DESC LIMIT 5;
Example 2:
SELECT event_name, COUNT(donation_id) AS donation_count FROM donations WHERE YEAR(donation_date) = 2021 GROUP BY event_name HAVING donation_count > 0;

Now, using these examples as a guide, please help me write a SQL query optimized for counting the number of attendees at the highest-raising fundraiser events

A third strategy for inciting creative thinking through refining prompts involves identifying, and remixing, the problem’s constraints. “Constraints” broadly refers to anything that limits the solution possibility space. This could be technical resources, or project timelines, available technologies, or regulations. You can experiment with grouping constraints differently to produce potentially novel suggestions from an LLM and explore the possibility space even further. While not all constraint remixes will be immediately practical, they are useful for enhancing understanding of a problem statement, and seeing how different constraints might overlap or interfere with each other.

For example:

Let’s consider a mobile app development project with the following constraints:

1. Limited budget
2. 3-month timeline
3. Must work on both iOS and Android
4. Needs to integrate with an existing backend API

Now, let’s remix these constraints: A. What if we had unlimited budget but only 1 month? B. What if we had 6 months but could only develop for one platform? C. What if we had to build a new backend but had 9 months to do it?

For each scenario (A, B, and C), suggest how the app development approach might change and what new possibilities or challenges could arise.

One more strategy for using prompt iteration to expand creative thinking is framing the prompt in different types of desired insights. These insights can be analytical, strategic, or reflective. Since the AI isn’t explicitly performing “strategy” or “analysis”, but is instead looking for high probability of association with these keywords, it can expand your discovery outcomes to experiment with prompts using different keywords to describe the desired insights.

For example:

Analytical: “Analyze the current market trends in renewable energy. What patterns emerge when comparing adoption rates across different countries and industries?”

Strategic: “Considering the analysis of renewable energy trends, what strategic moves could a traditional energy company make to position itself for long-term success in this changing landscape?”

Reflective: “Reflect on the historical resistance to renewable energy adoption. What underlying factors or assumptions may have contributed to this resistance, and how might these insights inform future clean energy initiatives?”

Seen this way, prompts are not just a tool for generating AI output, but can also help frame and re-frame questions to illuminate new possibilities. Crafting prompts can be about exercising curiosity, creativity, and analytical rigor, in addition to efficiency or speed gains. These benefits can be realized through breaking down the problem, providing examples, remixing constraints, or including analysis keywords in the prompt. All of these methods can help you deepen your understanding of a problem while solving it faster.

This piece was written with assistance from Claude 3.5 — Sonnet and ChatGPT-4o.