Generative AI tools, specifically large language models (LLMs) like OpenAI’s ChatGPT, need a bit of help to create anything usable. If you’ve played around with generative AI tools, you know that outputs tend to be lackluster at best, even with the most updated versions of these tools. Over the last year, we've learned, especially with ChatGPT, that the trick to getting better results is all about crafting better prompts.
Why Do Prompts Matter?
Imagine this: You’re managing a team of experts. They have all the knowledge and skills necessary to accomplish any goal you come up with. However, when presenting your business goals to the team, you provide vague ideas like “increase sales” and “fix company processes.”
You wouldn’t do that. Instead, you’d ask your team to increase sales by a specific percentage, by a particular date, using the tools you can access. Or you’d ask your team to identify flawed processes and give them a timeline to improve them.
The same idea is true with generative AI prompting. You need to clearly understand the problem you're trying to solve and your desired outcome before you ask an AI tool to help. If you don’t know what you want, even the best prompt is bound to fail.
Generative AI tools need exceptionally crafted and thought-out prompts to produce anything resembling quality work - though you’ll still have to edit and review everything it gives you to ensure it accomplishes the task properly. Even with a quality prompt, you'll probably find yourself tweaking them a few times over to get the results you're after. The trick is starting with a great prompt to save yourself time and energy refining it.
There are many ways to go about crafting the perfect prompt. Depending on what your objectives are and which AI tool you use, there are a few techniques that work best:
- Zero-shot prompting: Think of this like a quick Google search but instead of typing your question to Google, you ask an AI tool. his is perfect for LLMs like ChatGPT, where you're not loading a prompt with much context but just looking for a quick, straight-up answer.
- Iterative prompting: This is more like a conversation where you keep asking questions and adding bits of context after each response until it completes your objective. For example, when creating an image with DALL-E, you can iterate on the initial image it makes by adding follow-up instructions with more detail and context. This approach is best if you’re unsure of what you’re looking for or your prompt is extremely complex. The more complex the prompt, the more you should piece-meal it to your generative AI tool to ensure it captures all the details.
- Content summarization: ChatGPT-4 can give you a neat summary of an entire article, video or conversation: Plugins like Link Reader allow ChatGPT to read current links, including YouTube videos.
- Few-shot prompting: This is when you provide examples for context. This technique enables in-context learning to guide the tool to perform better.
These are just a few of the most popular techniques. However, there are many others
to try if these don’t work for you. The most important thing to note is that if you consistently get poor results, your prompt or technique is likely the problem. Try implementing a different approach or adding more detail to your initial ask.
Creating Quality Prompts
Now that we’re all on the same page and understand that bad prompts equal inadequate outputs, let’s jump into how to create quality prompts. Every generative AI tool will be slightly different, but generally, the more specific your prompt, the better your outcome will be. There’s an easy acronym you can use to guide your prompt creation:
- Context – provide as much context as possible. This helps your AI tool generate relevant and appropriate content.
- Role-play - set the scope of your chat with role-playing before asking it to write anything.
- Illuminating - be clear and avoid vague ideas or language. For example, when using an image-generating AI tool, avoid asking for something abstract and realistic, as those are opposing asks.
- Specific – include details such as your desired output format, length, audience etc.
- Plain – make it plain and simple. Keep your prompts as short and to the point as possible; the longer and more complex your prompt is, the more likely your AI tool will overlook aspects of your ask. Instead, consider breaking longer prompts and use the iterative technique to complete your objective.
CRISP prompts will help you get what you’re looking for faster with fewer back-and-forths – and isn’t that the whole point of these tools? To save time and get things down more effectively!
Of course, there will be days when writing a quality prompt is an impossible task in and of itself. In those moments, having a few premade prompts in your toolbox can save you a world of headaches. Now, every generative AI tool operates slightly differently, so you’ll have to create prompts that work best for your regular tasks with the tools you use most. However, try out this one, which we’ve found consistently useful for ChatGPT:
- You are a prompt-generating robot tasked to gather information about the user's goals, objectives, preferred output and other relevant information. Ask follow-up questions to the user until you are confident you can produce a perfect prompt. Start by asking about the user's goals, objectives and preferred output.
This lets ChatGPT generate its own prompt and produce content with follow-up questions until you are pleased. Additionally, plugins like Prompt Perfect
will rewrite your prompt with more detail and context.
There are many ways to integrate generative AI tools into your everyday workflow. However, they can’t beat the human touch. AI tools have a particular writing style and often miss the mark – so make sure your team uses it to enhance their productivity and creativity, not replace it!
If you’re interested in AI tools and learning how to integrate AI into your business, check out Volume 14 of our Marketing Trends
! Additionally, our AI Council
stays up to date with all the latest news and tools. Whenever you're ready, whatever question you have, our team is here and ready to help