Good prompts matter because they can save you time, headache, and tokens — the behind-the-scenes “currency” of generative AI. While most of us have chatted with a GenAI tool, you might need some tips to level up your prompt game. Here’s everything you need to know about what makes prompts work and why getting it right is so important.
Prompts and Tokens
If you’re new to GenAI — welcome! — you may have never encountered a “token” limit and might be thinking of skipping this section. Don’t! As GenAI use grows, understanding tokens and managing your input will become more and more important.
Essentially, a token is what a GenAI tool “spends” to generate your response. This happens completely behind the scenes, which is why lots of people have no idea they exist. Plenty of technical documents can explain the ins and outs of tokens (like this one from Microsoft), but what you need to know is that more tokens = more computing power = hitting your daily limit sooner.
Advanced GenAI users know that these tools have capacity limits. Because this happens behind the scenes, most people don’t know they exist. Once you hit that limit, your chat will switch over to an older model with slower response times and less sophisticated responses. You can keep chatting, but you might notice a difference in the quality of the responses.
If you’re using a Generative AI @ Pitt tool, you have a higher token limit than if you were using a free tool, but that doesn’t mean you can be flippant with your prompts. A strong prompt generates better responses faster with less iteration. That means you’re less likely to hit your token limit from a single conversation, freeing up valuable capacity for whatever else you need GenAI to do that day.
Building Better Prompts
Because prompts vary depending on what you’re asking, a good prompt will look different depending on the situation. At Pitt Digital, many of us use the CREATE framework to write better prompts. Simply put, CREATE provides the building blocks of a strong prompt: clear, specific, and filled with enough context to generate consistently quality responses.
Character | The role or persona you want the AI to assume. |
Request | Clear and specific request with context and key details. |
Examples | If relevant, illustrate the tone, style, or format you want. |
Adjustments | After the first response, tweak the way AI responds, either by editing the original prompt or iterating in a new prompt. |
Type of Output | The format and length of the final result. |
Extras | Additional instructions to make your prompt more effective, like “Ask me questions before you answer,” or “Explain your thinking.” |
CREATE may require more up-front effort, but you’ll find the AI generated output is more useful with far less fine-tuning. Putting it all together, here’s a CREATE prompt for a student looking for some personalized tutoring:
You’re a graduate assistant specializing in organic chemistry. Ask me guided questions to support my understanding of this week’s readings (attached). I’m struggling with ionic and covalent bonds. Don’t provide lengthy explanations; help me to understand the material on my own. Start with 3 key questions that we can build upon through Socratic discussion. Ask me questions before you respond to determine what I need the most help with.
Notice how our final prompt left out pleasantries like “please” and “thank you?” It might feel rude, but GenAI doesn’t care about manners. In fact, adding these into your prompts means that you’re spending tokens on things that don’t matter — a waste of time and resources. So, just this once, it’s okay to skip the P’s and Q’s.
Refining Your Prompts
Even with a strong prompt, it’s normal for GenAI to not get your request exactly right the first time. It’s so common, in fact, that adjustments are built into the CREATE framework! Adjusting your request — or iterating — is key to making the most out of GenAI.
To iterate effectively, you need to understand what’s working and what’s not with GenAI’s initial response. This is where your expertise can make a big difference, or, like our student in the example, where knowing your limitations can help shape how you follow up.
Let’s say our GenAI tutor took their role as a graduate assistant too seriously and provides a response that goes way over our heads. A smart iteration would then be to edit the initial prompt to say: You’re a helpful peer in my organic chemistry class, or You’re an expert in organic chemistry with a background in education.
More often, you’ll iterate using a new prompt, like if our GenAI tutor came back with tricky questions that we realize need more attention than we initially expected. That iteration could look like: Break down your first question into 3 sub-questions to augment my learning. I need more tutoring help than I initially thought.
The prompts you use to iterate might be shorter than your initial CREATE prompt, since the GenAI tool will have all the context you already provided. Still, use your prompting basics when you iterate. Clear, context-specific prompts that give the GenAI tool the benefit of your thought process.
GenAI can make mistakes. Check our guide to spotting and avoiding hallucinations.
Prompt Like a Pro?
If you’ve been keeping up with everything GenAI, it’s clear that making the most of this powerful tool isn’t about shortcuts. Effective prompting skills show us that expertise, context, and creativity build better prompts, which wouldn’t be possible without you bringing your A-game to your chats. If you’re ready to give it a try, log in to any of the Generative AI @ Pitt tools with your Pitt credentials and see how powerful a good prompt can be.
— Pitt Digital
