Glossary
What is a prompt in AI
In short
A prompt in AI is the text with which you give a task to an AI model: a question, an instruction, examples and the data the model uses to write its answer.
Prompts · prompt
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine you ask your older brother to draw a card. If you say “draw something”, you get who knows what. If you say “draw a cat in a blue hat for Grandma”, you get what you need. A neural network also waits for a request like this. Grown-ups call this written request a prompt. The clearer the request, the closer the answer is to what you had in mind.
In plain words
An AI model does not guess what you want: it works only from the text it received. This text is the prompt. It holds the task itself and everything that helps to understand it: who the answer is for, how long it is, in which style, and which data to rely on. The model reads the whole request and continues it with the most likely answer. So “write a letter” gives a general template, and “write the client a short letter apologizing for a two-day order delay” gives an almost finished text.
For business
For a company, a prompt works like a job description for AI. Its quality decides how many answers an employee accepts without edits and how many need rewriting. Good wording for routine tasks (letters, summaries, replies to clients) goes into a shared library, so nobody has to invent it again. The cost is not the text itself but the time to choose and test it; long instructions also raise the token bill. The risk is that an employee pastes personal data or trade secrets into a public service. So before rollout, companies write rules on what may be sent to an AI model and teach the team how to set a task.
For a specialist
A prompt is the input sequence that conditions the generation of a language model: the model continues it, predicting token after token. In chat models, the input consists of messages with roles: a system message, user and assistant messages, and the results of tool calls. Typical parts are an instruction, context or documents, examples, input data and a description of the answer format. The approach took hold after the work of Brown and co-authors (2020) on GPT-3. That 175-billion-parameter model solved new tasks from a description and a few examples right in the text, with no weight updates (in-context learning). Liu and co-authors (2021) described the “pre-train, prompt, predict” paradigm: the task is rewritten as a template with gaps that the model fills. There are discrete (text) prompts and continuous ones (soft prompts, trainable vectors). The context window limits the input length.
Example
A manager asks an AI model to “make a report” and gets general words. The second time, the manager attaches the monthly sales export and asks for three conclusions for the director, half a page long. This answer can be sent almost without edits.
Related terms
Questions and answers
Sources
- Brown et al., Language Models are Few-Shot Learners, 2020arxiv.org
- Liu et al., Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in NLP, 2021arxiv.org
- Akademos, Vinogradov Russian Language Institute of the Russian Academy of Sciences, the word "prompt", checked 16 October 2026orfo.ruslang.ru
Not sure whether your company needs this?
AI solutions map