A prompt is the input you give an AI model to tell it what to produce. In ChatGPT, Claude or Gemini, it is the message you type into the box, plus anything you attach to it: a file, a screenshot, a pasted error. The model reads the prompt and writes a reply that continues from it. It has no other window into what you want, so the prompt is the whole request.
The word comes from its everyday meaning: to prompt someone is to cue them to speak or act. An AI prompt is the cue. It can be a question, an instruction, a block of text with a task attached, or a description of a picture.
What goes into a prompt
A one-line question is a valid prompt, and for a one-line answer it is often enough. As the request gets more specific, a prompt tends to collect the same few pieces:
- The task: what to do, usually a verb. Explain, write, fix, summarize, translate, compare.
- Context: facts the model cannot guess. Who you are, what you already know, what the result is for.
- Input: the material to work on. The code, the email, the table, the article.
- Format: the shape of the answer. A list, a table, three short paragraphs, JSON.
- Constraints: limits. A word count, words to avoid, parts that must not change.
The difference between a weak prompt and a strong one is rarely clever wording. It is whether those pieces are present. "Write a function that sorts a list" leaves the model to guess the language, the data and the order. Add those three facts and the guessing stops. How to write a prompt walks through each piece with an example you can switch on and off.
Examples of prompts
Prompts look different depending on what you want back. Switch between the tabs to see a prompt for writing, one for code and one for learning, each with a typical reply.
Hi, I was expecting the report on Monday and haven't received it yet. I need it by the end of today so I can send the numbers to the client on time. Could you let me know when it will be ready? Thanks.
Image prompts work the same way, but the reply is a picture. Image generators, including the ones built into some chat apps, respond best to a description of the subject, the setting, the style and the framing:
A watercolor illustration of a small robot watering plants on a city balcony at sunrise, soft warm light, wide shot.
Each detail removes a choice the model would otherwise make for you. Leave out "watercolor" and you might get a photo; leave out "wide shot" and the balcony might be cropped away.
Prompt, reply and system prompt
Three pieces of text are involved in a typical chat with an AI model, and it helps to keep their names straight:
- The prompt (also called the user message) is what you send for this request.
- The reply (also called the response or completion) is what the model writes back.
- The system prompt is a set of standing instructions that applies to the whole conversation. The app or developer sets it, and you can add to it through settings such as custom instructions or project instructions.
The block below shows all three. The system prompt sets a rule once, and the reply follows it even though the user never repeats it.
Close! Here is one hint: run print(list(range(1, 4))) and compare the output with the two numbers you passed to range.
What do you notice, and what would you change in your loop?
Without the system prompt, a model would most likely hand over the fix, range(1, 11), straight away. The system prompt changed the behavior for every message in the chat. System prompt covers where to set one and what to put in it.
The prompt is more than your last message
In a conversation, the model does not remember earlier turns the way a person does. Each time you send a message, the app sends the model the system prompt and the conversation so far, with your new message at the end. That combined text is what the model actually responds to.
This has two practical effects. First, earlier messages keep shaping later answers: if you said "use TypeScript" ten messages ago, that instruction is still in play. Second, there is a limit to how much text the model can take in at once, measured in tokens. When a conversation outgrows it, the app has to drop or compress older parts or ask you to start a new chat, and instructions from the start of the chat can stop having an effect. Tokens and context window explains how that limit works.
Prompts are not database queries
On its own, a model does not look anything up. It generates the reply one small piece of text at a time, choosing each piece based on the prompt and on patterns learned in training. (Apps with web search or file tools fetch material first and add it to the prompt; the model still writes from the text in front of it.) That is why the same prompt can give slightly different replies on two runs, and why a model can state something false with complete confidence when the prompt asks about a fact it does not have.
Two habits follow from that. Put the facts you need the answer to rely on into the prompt itself, such as the document, the numbers or the error message, instead of expecting the model to know them. And check any fact, figure or quote in the reply before you use it.
Frequently Asked Questions
What is a prompt in AI?
A prompt is the input you give an AI model: the question, instruction or material you type, and sometimes images or files you attach. The model reads the prompt and generates a reply that continues from it. Everything the model knows about your request comes from the prompt and the conversation around it.
What does prompt mean?
In everyday English, to prompt someone is to cue them to say or do something. In AI, a prompt is the text that cues the model to produce an answer, a piece of code or an image. Many languages borrow the English word: Japanese writes it プロンプト and Korean 프롬프트.
What is an example of a good prompt?
"Explain what a Python dictionary is to someone who knows lists but not dictionaries. Use one real example and keep it under 100 words." It names the task, the reader's level, what to include and how long to be. A weaker version of the same request is just "explain dictionaries", which leaves the model to guess all of that.
Is a prompt the same as a question?
A question is one kind of prompt. A prompt can also be an instruction ("rewrite this email"), a block of material with a task attached (a pasted error plus "why does this fail?"), a description of an image to generate, or the start of a text for the model to continue.
What is the difference between a prompt and a system prompt?
The prompt is the message you send for one request. A system prompt is a separate set of instructions that applies to the whole conversation, such as "you are a tutor, never give the full solution". Developers set it through the API; in chat apps, custom instructions and project instructions play the same role. See system prompt.
