Learn AI /Large Language Models /User prompt
User prompt
Also called: user message.
Back to the milestone you came from
In five seconds
What it is
The input a person gives a chat assistant to guide what kind of response it should produce.
In plain words
A user prompt is what you give the AI to tell it what you want.
An everyday example
At a water station, Amina sends a note to the control room: “Pump 3 shows 92°C for two minutes. Give me three possible explanations, ranked by urgency, and say what data I should check next.” The technician uses that note to decide what kind of reply to prepare. If the note only said “Pump problem?”, the reply would likely be broader and less useful. If the note gave the wrong pump number, the reply could be off track from the start.
A human technician may use real-world experience, ask follow-up questions on their own, and notice when the note seems unsafe or inconsistent. A language model does not reliably do that. It generates text from patterns in data and the current context, so a well-worded prompt helps, but it does not guarantee correct judgement or factual accuracy.
Remember thisA user prompt is the input you give the assistant. Its wording, attachments, examples and requested format shape the response, but they do not guarantee accuracy.
Understand it
Understand it
Why it matters
For most people, the user prompt is the main handle they have on an AI system. It affects whether the output is vague or focused, short or detailed, balanced or one-sided. Knowing what a prompt is helps you ask for format, scope, uncertainty, and source checking more clearly. It also helps you recognise a limit: better wording can improve usefulness, but it cannot turn a weak model into a trustworthy expert or remove the need for review, especially in health, law, procurement, safety, or public-sector work.
How a user prompt shapes a model’s reply
After viewing this, the reader should be able to say that a user prompt is the person's input, that it is combined with earlier context, and that its wording influences the generated reply.
01
1. Person enters a request
The user types a message such as a question, instruction, or text to transform.
Draw a text box on the left labelled 'User prompt' with a cursor icon. Under it, show one short sample request.
02
2. Prompt joins earlier context
The system reads the new message together with relevant earlier messages in the conversation.
Add a second smaller box labelled 'Earlier chat/context' feeding into the same central area as the user prompt.
03
3. Model processes the text
The model uses the combined text to predict a response token by token.
Place a central processing box labelled 'Model generates likely next tokens' with incoming arrows from prompt and context.
04
4. Reply reflects the prompt
A specific prompt often leads to a more focused output than a vague one, though not necessarily a correct one.
On the right, split into two output boxes: one labelled 'Vague prompt → broad answer' and one labelled 'Specific prompt → structured answer'.
Show a left-to-right flow with three main boxes: user input, model processing, and reply. Above the first box, place two example prompts: one vague and one specific. Use arrows to show both entering the same model box, then show two different outputs to illustrate that wording changes the result.
Hovering over each sample prompt highlights the matching output path; on touch devices, tapping a prompt toggles the highlight.
Describe this without the diagram
The diagram is a left-to-right flow. First, a box labelled 'User prompt' shows the person's message entering the system. A second box labelled 'Earlier chat/context' also feeds in. Both arrows point to a central box labelled 'Model generates likely next tokens'. From there, arrows lead to two example outputs on the right. One path shows that a vague prompt tends to produce a broad answer. The other shows that a specific prompt tends to produce a more structured answer. The main idea is that the user's wording shapes the reply, but the model is still generating text from patterns, not guaranteeing correctness.
With reduced motion: A static flow chart shows the same four stages without animation: user prompt and earlier context feed into the model, which then produces example outputs for a vague prompt and a specific prompt.
A user prompt is the input, sometimes including attached content, that a person sends to an AI system. It tells the system what task to try, what information to use, or what style to follow. In a chat tool, this is usually the message you type.
What goes in can be very short, like “Summarise this paragraph”, or more detailed, like “List three possible causes of this sensor reading and explain the limits of each guess”. The clearer the request, the easier it is for the model to continue with useful text.
What happens next is not magic and not guaranteed understanding. The model converts the input into internal representations, uses the prompt with other information in its current context, and generates a reply token by token. The wording of the prompt strongly affects that process.
What comes out is a completion: generated text, including tables or code in some systems. Connected software may act on model outputs, but those actions are not the completion itself. A good prompt can improve relevance and structure, but it cannot make the model reliable beyond its actual limits. A weak or ambiguous prompt can lead to vague, wrong, or overconfident output.
How “A note to a control room technician” maps onto the real thing
| In the story | In the AI system |
|---|---|
| Amina’s note | the user prompt |
| Details like pump number, temperature, and requested format | instructions and input that shape the response |
| The technician’s written reply | the model’s completion |
| A vague or mistaken note leading to a weak reply | an unclear or incorrect prompt leading to poor output |
A worked example
The associate expert submits the non-confidential draft terms of reference with a request to identify ambiguous deliverables before clearance.
Starting from: Review the attached draft. Return a three-column table headed “Clause”, “Ambiguity” and “Question for the requesting office”. Focus only on deliverables and timelines. Do not rewrite the text or assess potential vendors.
- The interface treats the message, attachment and any tool parameters as the user prompt.
- The system places that prompt in context with earlier messages and its system or developer instructions.
- The model tokenises the context and generates a completion token by token. Phrases such as “three-column table”, “focus only” and “do not rewrite” condition the likely structure and scope.
- The requested constraints can make the draft easier to review, but they do not ensure that every ambiguity is found or that each suggestion is correct.
Result: The response is likely to be a review table rather than a rewritten document or vendor assessment. The associate expert still checks each entry against the file before sending the terms of reference for clearance.
In a United Nations workplace
An associate expert in a procurement unit is revising terms of reference that must reach the front office before the end of the week. In an approved internal assistant, they enter: “Review the attached non-confidential draft. List ambiguous deliverables in a table with the clause, the ambiguity and one question for the requesting office. Do not rewrite the text.”
The entire message, including the requested table, limits and attached draft, is the user prompt. Its wording conditions what the model generates, while earlier messages and system instructions also remain part of the context. A precise prompt may produce a more usable review, not a more authoritative one.
Before you try thisDo not upload draft solicitations, vendor submissions, evaluation material, personal data or other procurement-sensitive content to a tool that is not approved for it. Follow applicable access and handling rules even in an approved system. The output may flag useful questions, but it must not replace procurement review, clearance or accountable judgement.
Where you may hear it
- At a vendor demo for a procurement unit
- During an internal session on using an approved chat assistant
- While a focal point helps colleagues improve a recurring drafting request
- In an ICT governance board discussion on controls for generative AI tools
Avoid confusion
Avoid confusion
Often confused with
Prompt
The text, image or other input that guides a model on what kind of output to produce next.
System prompt
A high-priority instruction that sets the assistant’s role, rules and priorities before it responds to later messages.
Completion
The text a language model produces after a prompt, built token by token from patterns learned during training.
Context
The text, images or other inputs a model can currently use to shape its next output in a session.
Common misconceptions
Common misconception: A prompt is just a question.
Correction: It can be a question, but it can also be an instruction, a request to rewrite text, a list of constraints, an example to imitate, or attached content to analyse. Thinking of it only as a question hides how much the wording and structure matter.
Common misconception: If I write a very detailed prompt, the answer will be correct.
Correction: A detailed prompt often improves relevance and format, but accuracy still depends on the model, the available context, and the facts it can access. Clear instructions reduce some mistakes; they do not remove hallucinations or outdated information.
Common misconception: The user prompt is the same as the system prompt.
Correction: A user prompt comes from the person using the tool. A system prompt is a higher-level instruction set by the developer or organisation to guide behaviour. Users often do not see the full system prompt, and the two can interact.
Quick check
Which statement best describes a user prompt?
A user prompt is the input from the person using the system. It guides the output, but it is not the same as the hidden system prompt, it does not guarantee accuracy, and it is not the reply itself.
Go deeper
Go deeper
The simple version
A user prompt is what you give the AI to tell it what you want.
The technical version
In a language-model interface, the user prompt is the user's input message, usually plain text but sometimes combined with attached content or tool parameters. The system places this input into the current context alongside prior messages and any system or developer instructions. The model then tokenises that context and generates a completion by predicting likely next tokens. Because generation is conditioned on the prompt, wording, order, examples, constraints, and requested output format can materially change the response. However, the prompt only guides generation; it does not guarantee factual accuracy, faithful reasoning, or access to missing information.
Reviewed external resources
EncyclopaediaWikipedia search: User prompt (opens in a new tab)Wikipedia
This Wikipedia search results page helps readers quickly see how the term is used across related topics and suits readers who want a broad starting point rather than one fixed article.
search results, not one pagelink checked 2026-07-31
CourseLLM Course (opens in a new tab)Hugging Face
This course gives structured lessons on how language models use prompts and suits readers who want a guided, practical introduction.
Deeperlink checked 2026-07-31
VideoLarge language models, explained visually (opens in a new tab)3Blue1Brown
This video offers a visual explanation of how large language models work and suits readers who learn best through diagrams and intuition.
Intermediatelink checked 2026-07-31
VideoVideo explanations: User prompt (opens in a new tab)YouTube search
This YouTube search results page lets readers compare different video explanations of the term and suits readers who want to choose a teaching style that fits them.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: User prompt (opens in a new tab)Reddit search
This Reddit search results page shows community questions and examples about the term and suits readers who want informal discussion, while remembering that posts may be uneven in quality.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
A user prompt is the input you give the assistant. Its wording, attachments, examples and requested format shape the response, but they do not guarantee accuracy.