What is a prompt? Explained with briefing a new contractor

In one sentence: a prompt is the message you give an AI model to tell it what you want: your question or task, plus any background, examples and rules for the answer, and it is the only brief the model has to work from.

The everyday picture

Picture hiring a skilled contractor for one job in your office. They are very capable. But they have never been inside your business. They do not know your customers, your style or what "done" looks like to you.

So you write them a brief. A short note gets a guess. A good brief says what the job is, why it matters, what to include and how the finished work should look. Sometimes you hand over a sample of past work, too.

The AI model is the contractor. The prompt is the brief.

How it works, step by step

1. You decide the job (what you need done). Before you type, know the result you want: a reply to a customer, a summary, a list of ideas.
2. You write the prompt (the brief). Google's documentation says a prompt can hold questions, instructions, background information, examples and partial text for the model to finish. Anthropic's guide suggests treating the model like "a brilliant but new employee" who lacks context on how you work.
3. You add context and samples (background notes and past work). Saying why a rule matters helps the model understand your goal. Anthropic calls examples one of the most reliable ways to set the format and tone of the answer.
4. The model does the work (the contractor gets on with it). The model writes its answer using the prompt plus what it learned in training. It does not know facts about your business that you left out.
5. You check and refine (you review the job and update the brief). Anthropic's article notes that the first prompt rarely works perfectly, so test and refine. You can reuse a brief that works.

Where the picture stops

• The contractor will not phone you with questions. A real contractor asks when the brief is unclear. A model usually just answers, filling any gap with its best guess.
• The contractor does not remember last month's job. A model keeps nothing from one request to the next on its own. Apps that seem to remember usually send earlier messages or saved notes along with each new prompt.
• The brief is not the only thing that shapes the work. Apps can add their own hidden instructions, called system instructions, before your message. Google describes these as a preamble the model sees first.

Why it matters for your business

• Better results from the same tool. The same AI model can give a vague answer or a useful one. The difference is often how clear and complete the brief is.
• Less time fixing drafts. Stating the format, length and tone up front means fewer rounds of edits.
• Consistency across a team. A tested prompt can be saved and shared, so everyone gets work in the same style.
• Lower risk. Giving the right facts, and telling the model it may say when it is unsure, leaves less room for made-up answers. You still need to check the work.

An example

Imagine a 15-person travel agency that uses an AI assistant to draft replies to customer emails. At first, staff type "reply to this email". The drafts are polite but generic, too long and sometimes promise things the agency does not offer.

They write a better brief. It says who the agency is, which trips it sells and that replies should be under 150 words and warm but plain. It says to point people to the booking team for refunds, and includes two past replies the owner liked. The drafts now need only small edits, and staff save the brief so everyone uses the same one.

Frequently asked questions

What is prompt engineering?

It is the skill of writing and refining prompts to get better results from AI models. Anthropic describes it as the craft of structuring instructions to get better outputs.

Do I need special words or tricks in a prompt?

No. Clear, direct instructions in plain language work best. Anthropic suggests a simple test: would a colleague with little context be able to follow your prompt?

How long should a prompt be?

As long as it needs to be to explain the task, and no longer. A one-line question is fine for simple things. A task with rules, a format and a tone needs more detail and perhaps an example.

Can a prompt include files or documents?

Yes, many AI tools let you attach documents, and they become part of what the model reads. Everything you send counts towards the model's working space, its context window.

Related explainers

• What is an LLM? Explained with a new hire who read the whole library
• What is a context window? Explained with a clerk's desk
• What are AI hallucinations? Explained with a confident quiz guess
• What is RAG? Explained with an open-book exam

Sources

• Anthropic, Claude Platform Docs: Prompting best practices
• Anthropic: Best practices for prompt engineering
• Google, Firebase AI Logic docs: Understand prompt design