Lesson 14 of 28 · Making AI useful
What is a prompt?
A prompt is the text you give an AI model to tell it what you want: a question, an instruction, or a request plus any context, examples or files that go with it. The model reads the prompt and generates a response from it. Clear prompts that state the task, the audience and the format you want usually get noticeably better answers.
A prompt is a brief, not a search query
The word comes from theatre, where a prompt is a cue that tells an actor what to say next. In AI, it is whatever you type or upload that the model responds to. It can be one word or several pages long, and it can include images or documents if the tool supports them.
The most useful mental model: imagine briefing a capable new colleague who knows a lot about the world but nothing about you, your project or your standards. 'Write an email' leaves them guessing. Saying who it is for, what it must achieve, how long it should be and what tone to use lets them get it right the first time. A prompt is not a search query either: full sentences with context work better than a few keywords.
What goes into a good prompt
Showing examples works because large models can pick up a pattern from a few demonstrations, as Brown et al. (2020) showed. Asking a model to work through a problem step by step can also improve answers on reasoning tasks (Wei et al., 2022). The prompt engineering guide covers these techniques in depth.
Most strong prompts combine some of these parts. You rarely need all of them:
- Task: what you want done, stated as a clear instruction.
- Context: background the model cannot know, such as your audience or goal.
- Examples: one or two samples of the output you want (called few-shot prompting).
- Format: length, structure, tone, or a template to fill in.
- Constraints: what to avoid, and what to do if information is missing.
System prompts: the part you never see
What you type is usually not the whole prompt. Most AI products add a system prompt, a set of hidden instructions from the app's developer that sets the assistant's role, rules and style. The conversation so far is also sent along each time, and so are any files or search results the tool retrieves. All of it must fit in the model's context window, the limit on how much text it can read at once.
This is why the same model can behave differently in two products, and why very long chats can start to lose track of early instructions. Some apps let you add your own standing instructions, which work like a personal system prompt.
Where you write prompts, and what they are not
Prompts are not only for chat assistants. Image generators turn a text prompt into a picture, coding assistants take prompts in your editor, and automation tools send prompts to a model as one step of a workflow. Some products hide the prompt behind buttons such as 'summarise' or 'make shorter', each of which sends a ready-made prompt for you.
A prompt is also not magic wording. Clarity, context and checking the result matter far more than secret phrases, and a prompt that works well with one model may need adjusting for another.
Example
Try it: a vague prompt vs a clear one
Try this promptVague: Write an email about the meeting. Clear: Write a short email to my team of five designers. Our Thursday 3 pm review is moving to Friday at 10 am because the client needs more time. Keep it friendly, under 100 words, and end by asking anyone who cannot attend to reply by Wednesday.
Paste both into any AI assistant and compare. The second prompt states the audience, the facts, the tone, the length and the action you want.
Key points
- A prompt is everything you give a model to respond to.
- Write it like a brief to a smart colleague who lacks your context.
- Task, context, examples, format and constraints make prompts stronger.
- Apps add hidden system prompts, so the same model can behave differently.
- Clarity beats magic phrases; check and refine the result.
Try it on Mindber
Related terms
Related topics
Frequently asked questions
What is the difference between a prompt and prompt engineering?
A prompt is a single input you give a model. Prompt engineering is the practice of designing, testing and improving prompts so they produce reliable results, especially when the same prompt runs many times inside an app or workflow. Everyone who uses an AI assistant writes prompts; prompt engineering is the more systematic version of the same skill.
How long should a prompt be?
As long as it needs to be to remove guesswork, and no longer. A simple question can be one line. A task with a specific audience, format or set of facts needs those details spelled out. Long prompts are fine when the content is relevant; padding them with vague or repeated instructions tends to make results worse, not better.
What is a system prompt?
A system prompt is a set of instructions, usually written by an app's developer and hidden from users, that the model receives before your message. It sets the assistant's role, rules, tone and limits. It explains why the same underlying model can behave differently in different products, and many tools let users add their own standing instructions on top.
Sources
- Language Models are Few-Shot Learners. Brown et al., arXiv, May 2020
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Wei et al., arXiv, January 2022
- Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing. Liu et al., arXiv, July 2021
Last updated