Prompt: the art of suggesting answers to generative artificial intelligence

When I was a child, the phrase of a little poster that an aunt of mine had in her bedroom stuck in my mind. On it, next to a drawing of the Pink Panther, there was a text: “When I got all the answers, all the questions changed”. It seemed to me an ingenious way of saying nothing.

Today we can make more sense of it thanks to artificial intelligence because generative AI systems have given new meaning to the importance of asking questions SEO had taught us that, if our content was well targeted, it would answer the questions asked by the people we wanted to reach. Today, the opposite is true: we must learn to ask those questions that will provide us with the answers that we intuitively believe to be correct. Welcome to the art of prompts or prompting.

The term prompt has its current origin in the computer world, where it refers to a message or indication that is displayed on a system to indicate that it is ready to receive an input or an action from the user. Thus, an = sign in a table conveys that what will follow is a command, while in a form, the prompt requests a specific action, such as “Enter your name” or “Enter your suggestion”. In some ways, the prompt resembles the nudge of behavioural science, albeit without any subtlety.

Outside the computer world, prompts are also used, but we didn’t call them that, at least in Spanish. We used to refer to them as “suggestions” or “indications”. In a theatre performance, it is what the prompter does when an actor forgets the text, while in the field of psychology, it is a kind of stimulus given to an individual to help him or her answer a question or carry out a task. In short, prompting is not exactly the art of asking the right questions, but the ability to guide a system (human or otherwise) to provide an appropriate response… even if it is not correct.

“Prompting is the ability to guide a system to provide an appropriate response”

The difficulties in formulating a good prompt

With the advent of generative artificial intelligences, formulating prompts has become the new “must” in SEO. We are all trying to get the most out of these environments and, in just a few weeks, we have gone from playing with absurd questions to ChatGPT and random requests to Dall-E, to discovering that other people are capable of achieving useful results for the world of communication and marketing. The key was – who knew – just asking the right questions. That’s prompting, folks.

Formulating prompts or “guiding” artificial intelligences in the development of viable answers is not easy. In fact, I have already commented in an article that one of the essential requirements to obtain valid results lies in knowing the subject matter and, in the case of images, having in mind what kind of result we are imagining. As someone once said, “we only know what we want when we are faced with what we don’t like”.

The social networks are now full of experts sharing their prompting techniques: prompts to generate editorial calendars, prompts to create marketing plans, prompts to calculate costs and profits, prompts to establish SEO strategies… Most of the white papers being shared on LinkedIn and the successful practical threads focused on this sector have prompting as a protagonist, with predefined sequences of questions for us to fill in the spaces with the details of our case.

How to make a good prompt

So how do you generate a proper prompt? If we ask the king of viable answers, ChatGPT, he gives us the keys:

  • Clarity. We must be clear and, be careful, concise, and precise when formulating our request. Being brief is easy; being concrete is not.
  • Relevance. The prompt must be related to the way the model has been trained and the type of content it is expected to generate. The latter we will find out when we read it. The former is a mere assumption, because only those who train the model know how it is trained.
  • Specificity. The more specific, the more accurate the response. The key is to add complements in the same sentence, to narrow it down as much as possible.
  • Creativity. For me, this is what separates the experts from the amateurs. It means applying some creative freedom to the AI model to generate original content… or as original as a generative AI can be.
  • Consistency. We must maintain a consistent structure to ensure that the model knows what to expect and can respond more accurately.
  • Length. We should not formulate prompts that are too long or complex, as this can confuse the model and decrease the accuracy of its response.

Tips on how to use generative AI well

Is this all? No, of course not. To what ChatGPT tells us, let’s add three tips:

  1. If the prompt is longer than the answer that can be generated, then we are not prompting properly. We probably know the answer we want. So why are we asking? And, above all, if it is going to take longer to ask than to get the expected answer, we are wasting our time.
  2. Beware of using prompts offered to us by other people. Indeed, there is an artificial intelligence training model, but it learns from us and considers what we have been formulating, our interests and our style. In conclusion: what worked for that “expert” on LinkedIn who promised to share his magical prompts with you is likely to work badly for you.
  3. You’d better know the subject matter you’re asking the AI about. Otherwise, any wrong answer will seem possible. Therefore, prompting is like a process of mayeutics: a way of guiding the artificial intelligence so that, step by step, it gets to where you already know, but without you having to repeat the path.

“If the prompt is longer than the response that can be generated, then we are not doing the right prompting”

Generative AIs are supposed to help us optimise working time, but we should not rely on them to replace creative processes. In fact, without creativity, it is impossible to take advantage of them.

So now you know learn how to make prompts and keep an eye on upcoming solutions in this field, such as Perplexity, which will allow you to ask questions and get answers with perfectly identified sources of information. Will ChatGPT beat that?

Related Posts