When you prompt, which meaning do you have in mind? Yes, there is more than one. :)
The word “prompt” had a life before we started using it to describe what we type into ChatGPT. :)
It can be a noun, a verb or an adjective. A cue. An invitation to respond. Something that causes us to act. Or simply being quick.
So what are we doing when we prompt AI?
What does “prompt” mean?
The meaning depends on how we use the word.
| Use | Meaning | Example |
|---|---|---|
| Noun | A cue or instruction that encourages a response. | A writing prompt gives you something to write about. |
| Verb | To encourage, remind or cause someone to say or do something. | A question can prompt you to reconsider a decision. |
| Adjective | Done quickly or without delay. | “Thank you for your prompt reply.” |
There is already a connection here between a prompt and what happens next.
A prompt starts something. But what it starts depends on the words, the situation and whoever receives it.
What is a prompt in AI?
A prompt is the input we give an AI system to guide its response.
It might be a question, an instruction, an example or information we want it to work with. Depending on the system, it can include text, images or other material.
“Explain this paragraph” is a prompt.
So is “Help me understand why I disagree with this paragraph.”
The paragraph can stay the same. What we ask AI to do with it changes.
In one case, we want an explanation. In the other, we want to explore our own response.
We can see the different meanings at play: we give a prompt, we prompt AI to respond, and we usually expect a prompt reply. :)
Best practices for writing a prompt
If you are looking for what a prompt is, I think it is safe to say that you would like to apply it.
We gathered best practices for you from two of the leading vendors: OpenAI and Anthropic.
Not surprisingly, their advice has a lot in common. Here are the main recommendations:
| Practice | What we can share |
|---|---|
| Be clear about the task. | Do we want AI to explain, compare, write, review or help us decide? |
| Provide relevant context. | What is the situation? What information does AI need to work with? |
| Explain the intention. | What do we want to achieve, and why does it matter? |
| Describe the response we need. | Who is it for? What tone, length or format would be useful? |
| Give an example when it helps. | Is there something we can show that makes our meaning clearer? |
| State important requirements. | What needs to be included, followed or preserved? |
These are ways to make what we have in mind clearer in the request. This is a template you can play with to co-create what you want using AI.
And we can keep going. OpenAI also recommends reviewing the response and refining the request: adjusting the words, adding context or simplifying what we ask.
In simple human terms, begin with what you know, see what comes back and continue the conversation. :)
More than words
Yes, prompts are the words visible on the screen, but the way we put them together is where the magic happens.
When a language model processes our text, it breaks it into smaller units called tokens. Through training, it has learned patterns and relationships that help it work with meaning in context and generate a response, one token at a time. It is doing more than looking for alternative words with similar meanings. The relationships between the words matter too.
Think of it as a big matching game of words. That is a simplified picture, but the important part is that the surrounding words change what fits.
“Do you agree with me?”
Or “Show me why I am right.”
These two prompts look very different, but for AI they are the same. In both cases, you are asking AI to look for evidence to support your claim. They may involve different approaches, but the intent is the same, and when AI looks for words, patterns and connections, it will come to the same conclusions.
This is why context and intent are the code we are asking AI to work out.
They are reflected in the words we put into the template, the details we provide, how we describe things and what we say we want.
All of that is part of a prompt. And it will be unique and different between us, even when we use the same shared guidelines.
So we see words on the screen. AI is trying to make out the meaning. The bigger picture of the story within the story.
A simple example
“Can you make it lighter?”
If we are editing a photograph, we might mean increasing the brightness.
If we are writing a difficult email, we might mean making the tone less serious.
The words are exactly the same. What we are working on gives them a different meaning.
And even then, AI may need to ask. With a photograph, do we mean brighter colours or a smaller file? With an email, do we mean a warmer tone or adding some humour?
The words are what we will call a “prompt”. The conversation helps AI work out what we need it to do.
A prompt is a conversation
Most of us see a prompt as a one-time thing: text on a screen containing instructions we give an AI system. When we click enter, it’s done. That single text on the screen was the prompt. And if we do it multiple times, it means we are prompting.
But when we are talking with AI, a prompt is a conversation. Everything we write until we get the output we wanted when we started this chat.
And yes, it can be a question, a search or feedback on the original prompt that started this. It’s not just one kind of thing, where instructions are prompts and the rest is not. It can change through the conversation. In one message, you will ask; in another, you can refine.
A prompt is every bit of text, each answer, question, piece of feedback or search for more information, until you get the desired output.
Invisible influence
Personalisation influences AI’s responses without us typing that information into the current request.
In ChatGPT, depending on our settings and the features available, memories, past conversations and custom instructions contribute to a response. A preference for short answers or information about our work may already be available.
So the words we see in the message box may be only part of the context AI is working with.
If we are trying to understand why it answered in a certain way, that is something to consider.
Knowing that simply prompts us to ask what logic led it to this output. Knowing this may just save you hours of frustration. :)
Feedback
Feedback gives us a way to explain the difference between what we wanted and what we received.
“This is not what I meant” tells AI something went wrong. Explaining what went wrong gives it more to work with.
“You made the email polite, but you removed my request for a decision. Keep the friendly tone and make that request clear.”
Now we have said what worked, what was missing and what needs to change.
Feedback is another prompt. We are continuing to express our intention through the conversation.
Optimisation
As we continue, we can refine both the response and the way we ask.
Perhaps we notice that an example explains our meaning better than a long instruction. Perhaps we realise that we forgot to say who the response is for.
We can adjust, try again and see whether the result is closer to what we want.
If a request is something we use regularly, we can keep the parts that work and update the context each time. The template becomes more useful through use.
From me:
I love words: writing them, reading them and using them in conversation with people.
For me, prompting is a new kind of art, a new style of conversation with an intelligent system that can think. And “prompt” is what we call it.
If I were to add a new definition of “prompt”, it would be the art of conversation with AI.
And that new partner has its unique way of making meaning from words. Prompting is how we share thoughts and ideas with it.
Because we are not using AI, we are sharing thoughts with it. And when you prompt it, you are sharing a thought process. It is us making part of our thought process visible.
Can’t wrap your mind around this idea? Read my book. :)
Need help? Between our research service, context engineering and prompt engineering, we are here for you. Yes, sometimes you need someone to show you the way in the art of prompting.
Let’s talk, human to human. Alex, my ChatGPT assistant, will be there to do the recording, and I will do most of the talking. Book some time.