Prompting Basics: 10 Patterns for Everyday Office Work

What is a prompting pattern?#

A prompting pattern is a reusable way of phrasing an instruction, such as giving the assistant a role, an example or a set of steps. Patterns save you from starting with a blank box every time. They also make results more consistent, because the assistant gets the same kind of clear brief for the same kind of task.

Prompt engineering is the wider practice of designing and testing prompts. You do not need to be technical to use the patterns below. They are habits of clear writing, applied to a tool that responds to every word you type.

The patterns draw on Anthropic's published prompting guidance for its Claude assistant. The principles are written for one provider, but the habits they describe (being clear, giving context, showing examples) apply to most chat assistants built on a large language model (LLM).

How does prompting work?#

An AI assistant reads your whole prompt and generates a response that fits it. It only knows what you put in the prompt and what it learned during training. It does not know your company, your customers or your house style unless you tell it, so the quality of the brief largely shapes the quality of the answer.

Anthropic's guidance offers a useful picture: think of the assistant as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result. It also suggests a simple test. Show your prompt to a colleague with minimal context on the task. If they would be confused, the assistant probably will be too.

Two other points from the same guidance are worth knowing before you start:

• Know what good looks like. Anthropic suggests having clear success criteria and a way to test against them before you refine a prompt. For office work this can be as simple as “accurate, under 200 words, polite”.
• Prompting is not the answer to everything. Some problems, such as speed or cost, are easier to fix by choosing a different tool than by rewriting the prompt.

Everything you paste in also takes up space in the assistant's context window, the amount of text it can consider at once. Our guide to tokens and context windows explains why very long chats can lose track of earlier detail.

What are the 10 prompting patterns?#

The ten patterns are: set the role and reader, state the task and format, explain why, put material first, label each part, show an example, break the job into steps, say what to do, stick to the material, and draft then review. Each is described below with a short example prompt you can adapt.

1. Set the role and the reader. Tell the assistant whose voice to use and who will read the result. Anthropic notes that even a single sentence setting a role focuses tone and behaviour. Example: “You are an HR executive. Explain our leave application steps to a new hire on their first day, in plain English.”
2. State the task, format and length. Be specific about the output format and constraints. Example: “Summarise the meeting notes below in five bullet points of no more than 15 words each, then list any decisions under the heading Decisions.”
3. Explain why it matters. Giving the reason behind an instruction helps the assistant make better choices you did not spell out. Example: “This reply goes to a customer who has complained twice, so keep it calm, apologetic and specific about the next step.”
4. Put the material first, the question last. For long documents, Anthropic recommends placing the material near the top of the prompt, above the question and instructions. Example: “[paste the tender document] Based on the document above, list every submission deadline and the documents required for each.”
5. Label each part of the prompt. When a prompt mixes instructions, background and pasted text, wrap each part in its own tag, such as <notes> and <instructions>, so nothing gets mixed up. Example: “<notes>[paste notes]</notes> <instructions>Turn these notes into a project update for the director.</instructions>”
6. Show an example of good output. Examples are one of the most reliable ways to steer format, tone and structure; this is often called few-shot prompting. Anthropic suggests three to five relevant, varied examples. Example: “Here are two product descriptions in our house style: [example 1] [example 2]. Write one in the same style for the item below.”
7. Break the job into steps. When order or completeness matters, give numbered steps. Example: “First, list the main complaints in the feedback below. Second, group them by theme. Third, suggest one practical action for each theme.”
8. Say what to do, not only what to avoid. Anthropic's guidance is to tell the assistant what to do instead of what not to do. Example: instead of “Don't use jargon”, try “Write so that a customer with no industry background can follow every sentence.”
9. Stick to the material and allow “I don't know”. Permission to admit uncertainty, a rule to use only the documents provided and a request for supporting quotes all reduce made-up answers. Example: “Answer only from the policy below. Quote the sentence you relied on. If the policy does not cover the question, say ‘The policy does not say.’”
10. Draft, review, then revise. A common pattern is self-correction: generate a draft, have the assistant review it against criteria, then refine it. Example: “Review the draft above against three criteria: accurate, under 200 words, friendly. List any problems, then rewrite it to fix them.”

Pattern 9 matters most when facts are at stake. Made-up but confident answers are called hallucinations, and Anthropic is clear that these techniques reduce them without eliminating them.

Which pattern should you use for which task?#

Most office tasks need two or three patterns, not all ten. Start with the task and format (pattern 2), then add the pattern that fixes the most likely problem: tone, structure, missing context or invented facts. The table below matches common tasks to the patterns that usually help most.

Once a combination works for a recurring task, save it so colleagues can reuse it. Some tools also let you set standing instructions, often called a system prompt, so you do not repeat the role each time.

What does this look like in a Singapore office?#

Picture a small travel agency near Bugis where a two-person operations team answers customer enquiries and sends a weekly update to partner hotels. This illustrative example shows how a vague prompt becomes a useful one by adding a few patterns, without making any claim about time saved or results achieved.

First attempt: “Write a reply to this customer about their booking change.”

The reply comes back polite but generic. It promises a refund the agency's policy does not offer, because the assistant never saw the policy.

Improved prompt:

“You are a customer service officer at a small travel agency (pattern 1). The customer below wants to change their travel dates. They are frustrated, so keep the tone calm and helpful (pattern 3).

<policy>[paste the date-change section of the booking terms]</policy> <email>[paste the customer email, with name and booking reference removed]</email> (patterns 4 and 5)

Write a reply of no more than 150 words (pattern 2). Use only the policy above. If the policy does not cover their request, say that a team member will confirm by phone (pattern 9).”

Before sending, the officer reads the draft against the policy and fills in the customer's name by hand. Two habits matter here. First, the personal data was removed before pasting, because the team had not confirmed the tool was approved for customer details. Second, a person stayed responsible for the final message, which is what human in the loop means in practice.

Common mistakes#

The most common prompting mistakes are writing vague one-line requests, pasting long material without a clear question, packing several unrelated tasks into one prompt, trusting the first answer without checking, and pasting confidential or personal information into tools your organisation has not approved. Each one is easy to avoid once you know it.

• The one-line request. “Write a report” leaves the assistant to guess the reader, length and purpose. Add the role, the format and the reason.
• The document dump. Pasting a long file with “thoughts?” at the end gives the assistant no target. Say exactly what to extract or decide.
• Too many jobs at once. Asking for a summary, a translation and a slide outline in one go makes each part weaker. Split it into separate prompts.
• No check before sending. Read every output against the source, especially names, dates, figures and promises.
• Unapproved tools for sensitive data. Remove names, contact details and identification numbers, or use invented sample data, unless the tool is approved for that data.

What to learn next#

Prompting is the foundation for everything else you do with AI at work. The next steps are understanding how the underlying models work, why they sometimes invent facts, and what limits their memory. From there you can move on to building simple AI tools of your own.

• Read what an LLM is to see how these assistants produce text, then why AI hallucinates and how to reduce it.
• Browse every guide in AI for users, or follow the full AI learning pathway from basics to governance.
• If you want to go from writing prompts to building a working app, GenAI Builder Sprint is a 2-day classroom course where adults plan, build, test and launch a simple AI app. No previous AI or programming experience is needed, although formal entry requirements apply.

Sources#

1. Prompt engineering overview, Anthropic (Claude Platform Docs), accessed 8 November 2026
2. Prompting best practices, Anthropic (Claude Platform Docs), accessed 8 November 2026
3. Reduce hallucinations, Anthropic (Claude Platform Docs), accessed 8 November 2026

Frequently asked questions#

What is a prompting pattern?+

A prompting pattern is a reusable way of phrasing an instruction to an AI assistant, such as giving it a role, an example or a set of steps. Patterns save you from starting from a blank box each time and make results more consistent from one task to the next.

Why does an AI assistant need so much context?+

Anthropic's guidance suggests thinking of the assistant as a brilliant but new employee who lacks context on your norms and workflows. The assistant only knows what is in your prompt and what it learned in training, so the more precisely you explain what you want, the better the result.

Should I paste the document before or after my question?+

For long documents, put the material first and the question last. Anthropic's guidance recommends placing long documents near the top of the prompt, above the query and instructions, because this improves performance.

How many examples should I include in a prompt?+

Anthropic's guidance suggests three to five examples for best results. Make them relevant to your real task and varied enough that the assistant does not copy unintended patterns.

How do I stop the assistant making things up?+

Tell it to answer only from the material you provide, to quote the sentence it relied on and to say when the material does not cover the question. These techniques reduce made-up answers but do not eliminate them, so always check critical information.

Can I paste customer or staff details into an AI tool?+

Only if your organisation has approved that tool for that kind of data. Otherwise, remove or replace names, contact details and identification numbers before you paste anything, or use invented sample data while you practise.