Garbage in, garbage out.
As technology evolves, this computing ethos remains true. The quality of the output is only as good as the quality of the input.
It’s an easy concept to understand in terms of calculations or data processing, but this sentiment is even more critical in the age of Generative AI (GenAI). Tools like ChatGPT 4.0 don’t just regurgitate facts — they offer dynamic responses based on the conversation being had. And while training data and algorithms are important, the quality of that conversation also depends heavily on how we prompt the tool.
What is a generative AI prompt?
ChatGPT defines a prompt as the input provided by a user to initiate a response from the model. Good prompts provide clarity and specificity across three buckets:
- Style: The tone and level of detail provided by a Generative AI model can vary widely, ranging from simple to technical, formal, or even creative.
- Format: Guide the tool to produce the output that’s most useful to you, whether in bullet-point form or a blog post.
- Restrictions: Focus the tool to only use sources and information that you deem relevant. For example, you could instruct ChatGPT to “only use information from academic journals”
When planning a blog, let’s say, a curious beginner might ask ChatGPT something like, “Tell me about the extinction of the dinosaurs.”
A grizzled ChatGPT veteran would ask it something like, “Write me a 1,000-word blog post summarizing the top factors that contributed to the extinction of the dinosaurs. Only use sources from 2020 or later, and write it in a style that a layperson could understand.”
How can I refine my generative AI prompts?
According to Futurepedia, there are three core prompt categories that MarComm professionals can use when leveraging Generative AI for researching, brainstorming, or shaping content.
- Role Prompting: Role prompting is when you ask ChatGPT to assume an identity in order to guide its response. It’s a great way to incorporate some of the “style” elements of specificity listed above by instructing ChatGPT to take on the lens of a specific persona. Instead of “define account-based marketing,” an example of role prompting would be:
Prompt: “You are a CMO with 20 years of enterprise software marketing experience giving a presentation on the latest account-based marketing strategies and measurement best practices to other seasoned marketers at a technology industry conference. Write an outline for a 30-minute PowerPoint presentation that tells the audience how account-based marketing has changed in the last decade, and what measurement strategies they should implement in response.”
- Few-Shot Prompting: Few-shot prompting is when you provide a content sample to ChatGPT to help it understand the style and tone you are looking for. It’s another cheat code to guide specificity. Rather than telling the tool what kind of style or tone you want, you’re actually showing it.
Prompt 1: “I’m providing a sample of my writing to help you understand my style. Please analyze this sample for tone, structure, and language.”
Follow-Up Prompt: “Now, write a blog post based on the bullet points below in accordance with my writing style.”
- Chain of Thought Prompting: One of the coolest things about GenAI, and ChatGPT specifically, is its ability to reason, not just regurgitate. A great way to guide this type of output is with “chain of thought” prompting. This kind of prompt asks the tool to break complex (often subjective) topics into logical steps.
Prompt: “Who is the more legendary performer, Taylor Swift or Madonna? Let’s think about this carefully, step by step.”
Once you have the initial output (which in this case is very impressive — I encourage you all to try it), there are really great follow-up prompts you can use to dig deeper into the arguments, test other hypotheses, or ask for better ways to understand the information. A few examples, and their corresponding ChatGPT outputs, include:
- Explain it like I’m five: Okay! Imagine there are two amazing singers, like superheroes in music! One is named Madonna, and she’s been singing and dancing for a really long time. The other is named Taylor Swift, and she’s super popular right now…
- Use an analogy: Madonna is Michael Jordan, Taylor Swift is Lebron James.
- Take a side: Taylor Swift’s impact on modern music surpasses Madonna’s through her unparalleled songwriting, genre versatility, and deep fan connection.
Another great aspect of this meta world of GenAI is that, if you’re stuck on the best way to write a prompt, you can ask ChatGPT!
Like so many things in life, with ChatGPT, you get what you put into it. Our teams have found these prompting tips extremely helpful as we strive to not just use the tool but to use it efficiently.
To learn more about how you can get the most out of Generative AI, head to the MG Labs page where we’re testing AI tools, platforms, and features to uncover impactful marketing and PR use cases. And visit our Merritt Group Artificial Intelligence Industries page to see how we help B2B brands break through in this fast-evolving market.
