Why Do We Need to Learn How to Create Prompts?

Honestly, I used to think that “prompting” to work with an AI tool was a simple task, almost like common sense (and let me tell you, common sense is crucial, no doubt!). I believed that having good writing skills was more than enough to get the expected results. But as the saying goes, ignorance is bold, because you should never underestimate this task.

Even though AI tools are becoming increasingly sophisticated, the quality of your prompt directly affects the quality of the output. There's a reason why prompt engineering exists as a field, and it’s becoming more and more in demand.

A few weeks ago, I was chatting with Nuno Seixas and Andy Gadala María about this, discussing the importance of structuring a prompt correctly. The more detailed your objective, the task you want it to perform, the expected result in the desired format, and the foundational information you provide, the better the output.

If we don’t structure our prompts well, several things can happen, for example:

1. The resulting information might have gaps, false data, or even outright hallucinations (I find the concept of “hallucinations” in AI so amusing!).
2. The information might not adapt to the context, situation, or specifications you need. This leads to people saying things like, “AI can’t even calculate a holiday bonus.”
3. AI isn’t confrontational—it tends to agree with you. If you ask confirmation questions, AI will likely align with your input and give you an answer, which could very well be made up.
4. The output might be very basic, leading you to decide the tool isn’t suitable for your needs and abandon it. Sometimes, all you needed to do was refine your search, go deeper, and structure your prompt with more detail or differently.
5. The result might seem aligned with your expectations, but if you don’t verify it and validate the sources, you could end up with inaccurate information. Imagine researching a topic, receiving a result with sources included, and spotting a link to The New York Times. The URL even displays a title related to your topic. But when you click the link, it takes you to the NYT homepage, not a specific article. The article doesn’t exist. The AI tool pieced together fragments of information to generate the output. Always validate sources and cross-check results—never underestimate this task.

Cross-referencing results, validating sources, and analyzing information in detail to understand its origin and reliability are key steps to ensuring trustworthy data. Refining your search for more robust details helps too. But as you can see, it’s a joint effort. We can’t leave everything in the hands of ChatGPT or any other tool—we need to work together to get useful data.

There are many frameworks to help structure prompts, but here are some standout recommendations from various experts:

• Assign a specific role, like a teacher: “Teach me step by step about this topic,” or an editor: “Review this text and edit…” or an expert consultant: “As a marketing expert, brainstorm ideas for a new brand about…”
• Specify the format you want for the output, such as a table, list, essay, LinkedIn post, email structure, etc.
• Indicate the tone you’re expecting—professional, casual, emotional, persuasive, etc.
• Make a clear request without confusing words or distractions, and avoid integrating multiple questions or topics into a single prompt.
• Refine your result: after the initial request, you can dive deeper into specific aspects.
• Include aspects or limitations for the output, like “Limit the result to X country.”
• Ask for justification on how the proposed result was achieved or where the provided information comes from.
• Watch out for potential biases—AI is not free from them, and it shows.
• Request sources or resources where the information was obtained—and check them! As mentioned earlier, some links may not be as real as they seem.
• Finally, never, ever share confidential or sensitive information. Everything you upload to an AI tool contributes to its training, so protect all relevant, sensitive, personal, or confidential data.

Remember, these are “tools” that can provide great support, but we must learn to use them responsibly. There’s no point in “playing around” with one or another if you’re not aiming for a real benefit for your work or project.

I’m no expert on this (obviously), but I’ve been exploring various AI tools and platforms and have found many of them useful. Yet, new ones emerge daily, and we’re becoming more immersed in new technologies every day. That hype doesn’t always bring only good things. So, to navigate it all, let's always keep our best human abilities to discern: critical thinking, analytical skills, and common sense ;)