Strategies for Engaging with GenAI Prompting, Pushing Back, and Critical Analysis

As "reasoning" models continue to develop, it is more important to engage in back-and-forth dialogue with an AI tool than it is use the "perfect" prompt from the start. With that said, there are still some proven strategies that have shown to get more quality responses from generative AI on the whole. In a nutshell, users are best served by going beyond simply entering short, limited prompts and accepting the outputs that result without critical analysis. 

Keep in mind that all AI outputs can be improved. Thus, users will want to examine outputs carefully, compare with other sources, pushback, and even ask the tool they are using to revise, scrutinize, critique, or edit the outputs. To build skills in prompting and engaging with generative AI, investigate prompt examples, experiment, and write your own.  

Effective use of generative AI tools requires a healthy skepticism about the biases and tendencies that often show up in AI outputs. Crafting prompts and pushbacks that yield desirable results takes time, exploration, brainstorming, and a knowledge of the relevant subjects. Effective prompts are also unique to the user because we all speak, write, and think differently. Below are some suggested steps for effectively engaging with AI. Some might call these steps "prompting" or "prompt engineering." We at CELT refer to the process, rather, as engaging with AI because, as you can see below, it is much more than just a single prompt.

1. Tell the generative AI who it is. Provide a persona, knowledge, what its expertise is, what its goals are, how it interacts with the user, etc. Doing so unlocks key learned patterns that improve outputs.
2. Explain the situational context. Describe the who, what, where, when, why, how, etc. This step also provides additional information that can help the AI provide more relevant responses.
3. Define the criteria for optimal output. Mention what would it look like for the generative AI to behave effectively for the situation at hand.
4. Control for behavior. Establish rules for what the generative AI is to do and not to do (e.g., wait for the user to respond before proceeding, do not provide direct answers but instead provide leading questions, keep the response within X length, and so on). Interrupt the AI if it does something you don't want it to do.
5. Take a step-by-step approach. Break the tasks into individual steps (they can be numbered or identified using ordinal adverbs like first, second, third) and insist that the GenAI complete each step before moving on to the next. You can also tell the tool when to pause and wait for input from the user.
6. Provide examples and references. Give the generative AI tool useful examples, references, resources, models, or templates either by reference (e.g., “use X’s framework for Y”) or by directly uploading files or copying and pasting directly into the chat interface.
7. Engage in dialogue with the GenAI. Assume AI outputs and be improved. Pushback on the output to make them better. Don’t hesitate to correct it, ask it to make adjustments, take it in a different direction, critique its output, ask it to revise etc.

When in doubt, ask the generative AI to recommend prompts or look at quality prompt examples. There are loads of prompt libraries out there for you to explore to get ideas and strategies. Some examples of libraries and additional resources are provided below. Remember, the best prompts, though, are ones that thoughtfully outline the task and adhere to your style of communication so be sure to tailor examples to your own task and approach.

• Text Gen EdLinks to an external site.
• AI Pedagogy ProjectLinks to an external site.
• Exploring AI PedagogyLinks to an external site.
• OpenAI's Prompting StrategiesLinks to an external site.
• One Useful Thing: PromptingLinks to an external site.
• More Useful Things: Prompt Library