In plain words
When you type a question into ChatGPT, Claude or Gemini, that message is called a prompt. It can be a question, an instruction, a draft to fix, or a photo with a request attached.
The AI cannot see your face, your screen or what you meant to say. It only sees the words you sent, plus some background text the app adds. So the prompt is its whole brief.
That is why small changes matter. "Write a birthday message" gets something generic. "Write a short, funny birthday message for my dad, who loves old Kishore Kumar songs" gets something you might actually send. Getting better at this is called prompt engineering.
Everyday example
Giving directions to an auto driver
Say "take me to the market" and you might end up anywhere. Say "the vegetable market near the old bus stand, via the main road, I'm in a hurry" and the driver knows exactly where to go and how. The driver is skilled either way. What changes is how much of your intention reached them. A prompt is your directions to the AI.
Where the example breaks: A driver can ask you questions and sees the road. Most AI tools simply guess when your prompt leaves gaps, and they can sound confident even when the guess is wrong.
How it actually works
Inside a chatbot, your prompt is never sent alone. The app wraps it in a larger package before the large language model reads it.
The app builds the full input
Your message is joined with a system prompt written by the app maker, the earlier turns of the chat, and anything you attached. This whole package is what the model actually reads.
Words become tokens
The whole package is split into tokens, small chunks of text. It must fit inside the model's context window, the most text it can read at once.
The model continues the text
The model predicts the most fitting next token, then the next, until the reply is done. This is next-token prediction. Your prompt shapes every one of those predictions, which is why wording, detail and examples change the result.
A useful shape for a prompt
Say the task, give the context, and name the format. For example: "I'm visiting Jaipur for 2 days by train with my parents. Plan each day with morning and evening stops. Keep walking short. Answer as a list."
- What counts as a prompt
- Any input: text, a file, an image, or a voice message turned into text
- Where the word took off
- OpenAI's 2020 API was described as "text in, text out": give it a prompt, get back a completion
- Into everyday life
- ChatGPT's launch in November 2022 put a prompt box in front of millions of people
- Simple test
- Anthropic suggests showing your prompt to a colleague: if they would be confused, the AI will be too
Even a perfect prompt cannot stop every mistake. Check anything that matters, because AI can make things up.
Questions people ask
What is a prompt in ChatGPT?
It is the message you type into the chat box: a question, a task or a request. ChatGPT reads it, along with earlier messages in the same chat, and writes a reply. The more clearly the prompt says what you want, the more useful the reply tends to be.
What is a good prompt example?
Compare "Help me with my CV" with "Rewrite this CV summary for a junior accountant role. Keep it under 60 words, plain and confident. Here is my current version: ...". The second names the task, the context, the length and the tone, and includes the material to work on.
Is a prompt the same as a question?
Not always. A prompt can be a question, but also an instruction ("summarise this"), a role ("act as a strict examiner"), examples to copy, or a pasted document. Anything you hand the AI to work from is part of the prompt.
Does the AI remember my earlier prompts?
Within one chat, yes: the app sends the earlier messages back each time. Across separate chats, only if the app has a memory feature switched on. Otherwise each new chat starts fresh.
Why does the same prompt give different answers?
Most chatbots add a little randomness when picking each next word, controlled by a setting called temperature. That keeps replies from sounding robotic, but it means the same prompt can produce different wording, and sometimes different content.