How can you get the most effective outputs from GenAI tools?

Professor Oguz A Acar, Professor of Marketing & Innovation

In the previous step, you discussed some capabilities and potentials of generative AI tools. In this step, you will discuss prompt engineering.

The value derived from generative AI often hinges on skilful communication—often termed as ’prompting’.

Prompting research and techniques

Prompting is a vibrant research field with novel techniques emerging regularly. Here are some prevalent prompting methods to guide AI responses:

• Zero-shot prompting: you present the model with a new task without any prior examples. If your task could be addressed based on AI’s general training data, you may use this.
• Few-shot prompting: you provide a few examples (eg previous learning materials) to guide the AI before asking it the actual question. This helps the AI understand the context and the type of answer you’re seeking.
• Chain-of-thought prompting: this is a more structured approach, where you guide the AI step by step through a logical progression or sequence. This is particularly useful for complex queries where the answer requires a series of steps or considerations.

Depending on the complexity of your task and the specificity of the answer you desire, you can choose the approach that suits best.

There are other methods, but we will dedicate the remainder of this article to practical guidance.

Prompting best practices

Understand before you ask. The key to effective prompting is a strong understanding of the problem that you want to tackle. In other words, thinking about what task you want to accomplish with a specific generative AI tool. The truth is, without a well-formulated problem, even the most sophisticated prompts will fall short. However, once a problem is clearly defined, the linguistic nuances of a prompt become tangential to the solution. Before prompting, think about the problem, its root causes, components and boundaries. In this paper, AI Prompt Engineering Isn’t the Future, I explain problem formulation in more detail. In essence, you should clarify what exactly you want to achieve. Do you want to brainstorm ideas, craft an outline for a module, search the relevant sources or get counterpoints on your arguments? The clearer you are about your questions, the better the AI’s response will be.

Decompose your task. Break down your task into smaller and simpler components when possible. For example, various phases of a task, such as idea generation, searching for sources, outlining for a report, drafting and refining, can be considered distinct components to tackle individually. Similarly, treat the different sections of a report—the introduction, body and conclusion—as separate elements. Craft tailored AI prompts to address each of these individual components effectively.

Set context and boundaries for precision. When seeking information or content for a specific task, be explicit. Provide generative AI some context and background about the project or what audience the output should be suitable for. For example, if you are designing a learning activity for a module you are teaching, give the AI relevant context about the student background, learning outcomes and any specific methodologies or philosophies you are employing.

Get creative with constraints. To get some creative and out-of-the-box outputs, you can play around with constraints (eg imposing arbitrary constraints, modifying or removing existing boundaries) to explore novel perspectives. Some examples:

• You could ask the AI tool to adopt different perspectives (eg ask it to act as an entrepreneur, professor, etc, or even as real people like Maria Montessori or Isaac Newton): “How would Alan Turing comment on my arguments? Provide a feedback report, detailing strengths and weaknesses of my arguments, written by Turing”
• You could ask it to amend its output style (eg outline, bullet points, mind map, quiz, poem, using a structure like what/why/how, tweets, debates, magazine article, etc).
• You could ask it to suggest to you only out-of-the-box ideas (eg “For my upcoming learning activity on XXX [remember to provide context], brainstorm novel angles or approaches. Prioritise ideas that are uncommon or challenge the dominant narrative.”)

Interact with and give feedback to AI. Don’t rely on a single prompt to get everything you want. Engage with the AI in a back-and-forth manner, refining your questions based on its responses. If you want AI to change the tone of writing or incorporate a specific argument, tell it. If you find the output too complex, ask it to simplify (eg “Your explanation on how generative AI models work was too complex. Can you simplify it for a 10 year old?”) If you are unhappy with the solution presented, try reframing the question: ie the same problem asked from a different perspective. Sometimes, it is better to start over, as past conversations often influence outcomes.

Practice makes perfect. The more you use and interact with the AI, the better you will get at eliciting the responses you want. Use it frequently to familiarise yourself with its strengths and limitations.

Get help from the AI itself for prompting. AI tools themselves are getting better at crafting prompts. So, they can help you too. Some examples:

• You can describe your objectives to the AI and ask it what information it needs from you to help you: “I am trying to plan a workshop on AI ethics. What information do you need to suggest relevant learning activities?”
• You can ask it to put together prompts for you to address this problem: “I wish to generate various assessment methods for this workshop. Can you frame a prompt to help with that?”

Now that you have completed this step, you have developed an understanding on how to effectively communicate with generative AI systems. In the next step, you will learn about limitations of generative AI tools.

Join the conversation

Over time, many frequent users of AI develop a collection of ’go-to’ prompts that yield desirable outcomes. Have you started curating your own ‘prompt library’? If so, would you be willing to share a few prompts that you’ve found particularly effective? How did you come up with them?

If you don’t have a prompt library: please recall where you’ve engaged with an AI tool. How did the quality of your prompts influence the outcome? Did any specific prompt lead to surprisingly effective or unexpected results?