prompt

basicwidespread

A prompt is the request or instruction, usually text but sometimes a spoken sentence or image, that you give to artificial intelligence to get a response, write something, create an image, or perform a task.

In plain terms

A prompt is simply what you type or say to artificial intelligence. It can be a single question ('What should I cook tonight?'), but it can also be a multi-page task description with attached documents. The machine starts from this: what you do not state, and does not receive from elsewhere, it will at best guess.

It is like when you call a craftsman. If you just say "paint the room," you get some white walls. If you explain what color you want, by when, what to watch out for, and what your goal is for the room, you get much closer to what you envisioned. A good prompt similarly includes the goal, the context, and the desired form of the result.

How is it different from search? In search, you enter keywords and get a list of existing pages. For a prompt, the machine generates new text, images, or actions, so it pays to write in full sentences, in detail, and to ask follow-up questions or request clarification if needed.

How it works

You enter your request, optionally attaching a file, image, or previous text.
The system breaks the text into small pieces called tokens and usually adds the provider's own base instruction, which is invisible to you, along with the conversation history.
The language model reads all of this at once, then estimates piece by piece what would fit best as a continuation and chooses from these. This is how the response is generated, so it is not always the same.
If the result is not good, you refine it: your next message, together with the context, is sent as a new prompt, so the conversation can be refined on the fly.
For agents, the prompt gives an assignment not just for a response but for task execution: the agent independently searches, handles files, or runs programs based on your request.

An everyday example

You get home and type into the chatbot: 'There are eggs, potatoes, and onions at home, I have half an hour, I am cooking for two children, give me a simple recipe.' This sentence is the prompt. Because you provided the ingredients, the time, and who you are cooking for, you get a much more usable answer than if you asked 'what should I cook?'

Who is affected

everyoneAnyone using a chatbot, image generator, or voice assistant is writing a prompt, even if they do not call it that.

employeeThe best results from workplace AI tools come from those who can precisely state what they need and in what format. → Work & HR

parent, student, teacherIn learning, asking the right question determines whether the machine explains and provides practice or simply gives a ready-made solution. → Education & Family

developerThe behavior of models and agents embedded in applications is largely determined by the prompts written for them. → Code

A common misconception

MythA good prompt is a secret magic phrase you have to know, and then the machine suddenly answers perfectly.

In realityThere is no magic word: a good prompt is a clear, detailed task description, like one you would give a new colleague. With current models, precise context and goals matter more than any tricky formula.

Limits and risks

The best prompt does not guarantee a true answer: the model can be confidently wrong, so verify important statements.
The same prompt can yield different answers from another model, or even the same model on two occasions.
What you enter goes to the provider, so share personal, medical, or corporate confidential data only with caution.
The model can interpret every scanned text as an instruction, so a hidden foreign command embedded in a website or document can divert its operation.

Similar, but not the same

system prompt – The system prompt is written in advance by the provider or developer and sets the machine's behavior in the background, while the plain prompt is typed by you.
prompt engineering – The prompt itself is the request, while prompt engineering is the craft of consciously designing and testing requests.
search query – For a search query, you get a list of existing pages; for a prompt, the machine generates new content.

Related concepts

Broader

generative AI, large language model, human–machine communication

Narrower

system prompt, image-generation prompt, negative prompt, promptsablon

On the edge

prompt engineering, context window, token, prompt injection, chatbot

A brief history

The term has long referred to the command line prompt in computing, where the machine waits for the user's instruction. In its modern AI sense, it became widespread in the industry with large language models, especially after the launch of GPT-3 in 2020. It entered common parlance with the launch of ChatGPT in November 2022.

Where things stand

By the end of 2026, the prompt is a basic concept of using artificial intelligence, familiar even to everyday users. The emphasis is shifting from clever phrasing to what background information the machine receives and how much autonomy you give it as an agent.

Status: 2026 November

In the news

Appears in 68 Agent Age stories; the curve shows weekly occurrence in the news magazine.

TikTok's Symphony Agent produces video ads, while Search Ads gains keyword packsNov 2, 2026 · Marketing
Anthropic restricts Claude Code from cloning video gamesNov 2, 2026 · Code
Over 80% of South Korean judges have encountered AI-fabricated citations, survey findsNov 2, 2026 · Jog
Hungarian legal articles: company executives are personally liable for AI decisionsNov 2, 2026 · Jog
DeepSeek unveils proactive coding agent prototype to challenge AnthropicOct 31, 2026 · Code
'Hungarian AI tax': translation proxy reduces errors in Hungarian-language agent tasksOct 30, 2026 · Code
Óbuda University rector's instruction regulates the use of AI systems in teachingOct 28, 2026 · Education & Family
GitHub adds prompt-injection scanning to AI agent pull requestsOct 28, 2026 · Code

Also known as

prompt · promt · MI-prompt · AI-prompt · instruction · input text · user prompt · prompting

Sources

Wikipedia: Prompt engineering

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.