What is ‘meta prompting’ and how to apply it

ChatGPT, Gemini, Claude, DeepSeek and others artificial intelligences (AI) allow us to interact with them using natural language. And it’s fantastic. A decade ago this possibility plunged us into the realm of science fiction, but today, fortunately, anyone can use these AI. And you don’t need to know anything about AI or technology.

There is no doubt that the possibility of asking questions and making requests in our own language is a huge advantage, but it is important that we keep in mind that the quality of the answer that an AI will give us depends largely on the quality of our request. From the prompt we use. Because a prompt It is, precisely, an instruction or request that we give to an AI.

A prompt elaborate and meticulous define the context What AI should take into account, what role the model should take, what response format we need and what objective we pursue. The problem is that preparing such a detailed request is not always easy. And, furthermore, it can be a bit cumbersome. Fortunately, this problem has a solution: goal prompting. And it can help us get much more out of any AI in a very simple way.

The “prompt over the prompt” trick

the word goal It comes from Greek and means “beyond” or “above.” He goal prompting is literally thinking about the prompt before writing it. If we say to an AI “explain special relativity to me”, we will get a generic response. Correct, but generic. And the AI ​​does not know who the response is directed to, in what context we are going to use it, or what level of detail we need. The problem is not with the AI; lies in the instruction we have given you.

However, if we tell you the following:

“You are a physics teacher for high school students. Explain special relativity to me using only everyday analogies, without mathematical formulas. I want the students to intuitively understand why time passes more slowly at very high speeds. Maximum 300 words”

We will get a much more detailed and useful answer. Of higher quality. However, and here comes the really interesting part, the way to put the goal prompting What we propose consists of ask AI to help us generate the prompt that we are going to deliver to you.

Language models are extremely context-sensitive tools. For this reason, the same question asked in two different ways can produce radically different answers. He goal prompting Take advantage, precisely, of this sensitivity in a conscious way.

A very simple way to put this idea into practice is to add the phrase “Before responding, rewrite this prompt to make it more precise and effective. Then respond using that improved version” right after our initial request. If, for example, we say to ChatGPT:

“Explain special relativity to me. Before answering, rewrite this prompt to make it more precise and effective. Then answer using that improved version”

Will prepare the following prompt:

“Explain Albert Einstein’s theory of special relativity to me as if it were an introductory college physics class. Start with the historical context that led to its formulation, clearly state its two fundamental postulates, develop its main consequences (time dilation, length contraction, relativity of simultaneity, and mass-energy equivalence), use intuitive examples and thought experiments, include the essential equations, and explain what experimental evidence has confirmed it. End with a summary of the key ideas and their impact on physics modern.”

It is evident that this second option defines the context much better than our initial request. And it allows AI to give us a higher quality response. If we do this same experiment with Claude we will obtain the following prompt advanced:

“Explain to me Einstein’s theory of special relativity: its fundamental postulates, the most important physical consequences (time dilation, length contraction, mass-energy equivalence) and why it represented a break with classical physics. Use concrete analogies and an intermediate technical level.”

Again we will obtain a much more satisfactory answer. He goal prompting it does not make AI something it is not. It doesn’t give you capabilities you don’t have. What it does is eliminate the ambiguity between what we want and what the model interprets. The clearer we are when defining the context, our objective and the format we expect, the less the AI ​​will have to improvise and the better the result it will deliver to us.

Image | Generated by Xataka with a prompt created by Claude and submitted to ChatGPT

In Xataka | ChatGPT blocking mode: what it is, what it is for, who can use it and how to activate it

In Xataka | AI is replacing one of the most hated jobs in the world: the tailcoat collector