Learn AI /Large Language Models /Prompt
Prompt
- Starter
- Large Language Models
- Communications teams
- Policy officers
Also called: prompts, prompting.
Back to the milestone you came from
In five seconds
What it is
The text, image or other input that guides a model on what kind of output to produce next.
In plain words
A prompt is what you give the AI to tell it what kind of answer you want.
An everyday example
A programme officer asks an assistant to draft a short meeting note. First, the officer says only, "Write about today’s meeting," and the result is generic. Then the officer adds the audience, the three decisions made, the desired length, and the tone: neutral and formal. With that fuller briefing, the draft is much closer to what is needed.
A human assistant may truly understand the meeting, ask follow-up questions and use judgement about missing facts. A language model does not reliably do that; it generates output from patterns in data and can produce confident but incorrect text.
Remember thisA prompt shapes the model’s next output by supplying context and instructions. It does not guarantee accuracy or permanently teach the model.
Understand it
Understand it
Why it matters
For non-specialists, the prompt is often the main control they have over an AI system. Knowing how prompts work helps people ask for clearer outputs, spot when an answer drifted from the request, and avoid confusing a polished response with a reliable one.
How a prompt shapes an output
After viewing this, a reader should be able to say that a prompt provides instructions and context that influence, but do not guarantee, the model’s output.
01
1. Prompt enters
Show a text box containing a task, background, format request and constraints.
Draw a rectangular input panel on the left with four short lines: 'Task: summarise', 'Audience: managers', 'Use only notes below', 'Length: 5 bullets'.
02
2. Input is split into tokens
The model reads the prompt as small pieces rather than as a single block.
In the second box, show the input line broken into rounded token chips, with a small caption 'tokens = small text pieces'.
03
3. Output is generated step by step
The model uses patterns learned earlier to produce likely next tokens guided by the prompt.
In the third box, animate or illustrate token chips appearing one after another in a row, with arrows from the prompt box to show influence.
04
4. Completion appears
The result can match the request more closely when the prompt is clear, but errors are still possible.
In the rightmost box, show a short bullet list output and a yellow caution icon below with the text 'Useful direction, not a guarantee of accuracy'.
A left-to-right flow diagram with four boxes: Prompt, Model reads tokens, Next-token generation, Output. Small callouts show examples of useful prompt parts such as task, context, format and limits. A warning note under Output says better prompts improve direction, not truth.
Hovering over each useful prompt part highlights where that part affects the output: task affects content, audience affects tone, format affects layout, limits affect length.
Describe this without the diagram
The diagram shows a four-stage flow from left to right. First is a prompt box with example lines: task, audience, source limit and length. Second, the prompt is shown as small token pieces to indicate how the model reads input. Third, the model generates output one token at a time, influenced by the prompt. Fourth, an output box shows a concise bullet list. A note under the output says that clearer prompts usually improve direction and format, but do not guarantee factual accuracy or good judgement.
With reduced motion: A static four-box flow replaces animation. The third box shows a row of already placed token chips with arrowheads between them to suggest step-by-step generation without movement.
A prompt is the instruction or material you give a model before it produces an answer. It can be a question, a task, a few examples, a document to work from, or a mix of these. In many chat tools, the message you type is a prompt.
What goes in matters. Clear prompts usually state the goal, the relevant background, the format wanted, and any limits such as length or audience. If the input is vague, incomplete or contradictory, the output often reflects that.
What happens next is not magic and it is not training. The model processes the prompt as pieces of text called tokens, uses patterns learned earlier during training, and then generates a likely continuation token by token. In other words, the prompt steers the response for this conversation, but usually does not permanently change the model.
What comes out is a completion: text, and in some systems also images, code or other content. A better prompt can improve usefulness, but it cannot guarantee truth, safety or good judgement. The model can still miss context, invent details or follow the wrong part of the instruction.
How “Briefing a multilingual assistant before drafting a note” maps onto the real thing
| In the story | In the AI system |
|---|---|
| the officer’s briefing | the prompt given to the model |
| audience, decisions, length and tone | constraints and context inside the prompt |
| the draft note | the model’s completion |
| the more detailed second briefing | a more specific prompt that better steers output |
A worked example
At close of business, a reporting officer in a regional bureau needs four bullets for a briefing note to the front office, based on cleared notes from an inter-agency call.
Starting from: Using only the notes below, draft four bullets for the briefing note. Put agreed actions first, label anything unresolved, and do not add reasons or dates. Notes: Participants agreed to translate the guidance, ask the regional hub to test the registration form, and circulate revised talking points. Clearance of the budget remains pending.
- The prompt supplies the source text, audience, task and required format.
- The model tokenises this context and conditions its next-token predictions on those instructions and notes.
- The constraints make concise bullets and the stated actions more likely, although they do not guarantee perfect compliance.
- Generating the bullets does not normally change the model’s weights. The reporting officer still checks the draft against the cleared notes.
Result: - Translate the guidance. - Ask the regional hub to test the registration form. - Circulate the revised talking points. - Unresolved: budget clearance remains pending.
In a United Nations workplace
During a records retention review, a newly arrived associate expert in an evaluation office uses an approved internal assistant to draft a note for the file identifying document categories that need manual review before close of business.
The prompt can specify the permitted inventory fields, the retention categories to apply and the required table format. Those instructions shape the model’s output for this task, but do not update its weights or make its classifications authoritative.
Before you try thisDo not enter file contents or metadata containing names, witness material, personnel information, confidential evaluation evidence or procurement-sensitive material into a tool not approved for them. Even with an approved tool, provide only the minimum necessary information. A model must not make the final decision to retain or delete records.
Where you may hear it
- A vendor demonstration to an ICT governance board
- A drafting clinic before a first reading
- An office-hours session on an approved chat assistant
- A procurement meeting reviewing an AI service’s test results
Avoid confusion
Avoid confusion
Often confused with
System prompt
A high-priority instruction that sets the assistant’s role, rules and priorities before it responds to later messages.
Training
Training is the process of adjusting a model's parameters using data or feedback to improve performance on a learning objective.
Fine-tuning
Updating a pre-trained model's parameters with additional examples to improve performance on a narrower task, style or domain.
Prompt engineering
The practice of writing and refining instructions so a language model is more likely to produce a useful, safe and well-structured reply.
Common misconceptions
Common misconception: A prompt is the same as training the model.
Correction: A prompt usually affects only the current interaction. Training changes the model’s internal parameters using many examples over time.
Common misconception: If I write the perfect prompt, the answer will be true.
Correction: A strong prompt can improve relevance and format, but the model can still generate mistakes or unsupported claims because it predicts likely continuations, not verified facts.
Common misconception: A prompt must be long to be good.
Correction: Longer is not always better. The useful part is clear instructions and relevant context; extra text can distract the model or bury the key request.
Quick check
Which statement best describes a prompt?
A prompt is the input used for the current response. It can guide content, tone and format, but it is not the same as training, not the same as a system rule, and not a guarantee of accuracy.
Go deeper
Go deeper
The simple version
A prompt is what you give the AI to tell it what kind of answer you want.
The technical version
A prompt is the input context presented to a language model before generation. The model tokenises that input and conditions next-token prediction on it, which shapes the completion without usually changing the model weights.
Reviewed external resources
EncyclopaediaPrompt engineering (opens in a new tab)Wikipedia
This Wikipedia page gives a broad overview of prompt engineering and suits readers who want a quick general orientation.
Intermediatelink checked 2026-07-31
CourseLLM Course (opens in a new tab)Hugging Face
This Hugging Face course offers more structured technical learning about working with language models and suits learners ready for hands-on study.
Deeperlink checked 2026-07-31
VideoLarge language models, explained visually (opens in a new tab)3Blue1Brown
This 3Blue1Brown video explains large language models visually and suits readers who learn best through intuitive graphics and narration.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Prompt (opens in a new tab)YouTube search
This YouTube search results page helps readers find multiple video explanations of prompts and suits people who want to compare teaching styles before choosing one.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Prompt (opens in a new tab)Reddit search
This Reddit search results page shows community discussions about prompts and suits readers who want informal questions, examples and debates from other users.
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 prompt shapes the model’s next output by supplying context and instructions. It does not guarantee accuracy or permanently teach the model.