Skip to content

Prompt Engineering Foundations · Clarity, context and examples · lesson 4 of 16 · 11 min

Giving context and examples

Context: what a helpful colleague would need to know

Imagine asking a colleague to "draft a reply to this client". A good colleague would ask: Who is the client? What happened? What do we want to happen next? Is there anything sensitive? The AI needs the same information, but it cannot ask unless you invite it to. Giving context up front is the single biggest upgrade most people can make.

Useful context usually covers:

  • Who you are: "I run a two-person wedding photography business."
  • The situation: "A couple is asking for a discount because their budget changed."
  • The goal: "I want to keep the booking without cutting my price by more than 10%."
  • Constraints: "We have already paid for their venue permit."
  • Relevant material: the email thread, your price list, your policy.

How much context is enough?

Enough that the AI does not have to invent anything important. Not so much that the key point is buried. A good rule: include anything that would change the answer. Your company's founding year probably does not; your refund policy probably does.

Pasting material in

Often the best context is the actual document: the email you are replying to, your notes, a product description. Paste it in and label it clearly:

Here is the client's email:
---
(paste here)
---

Here is our package price list:
---
(paste here)
---

Draft a reply that offers the smaller package as an alternative.

Before pasting, remove anything private you would not want to share, such as personal phone numbers, ID numbers, passwords or confidential figures, and follow your workplace rules about AI tools.

Examples: show, don't just tell

Sometimes describing what you want is hard, but showing it is easy. An example of the style, format or tone you like is often worth a paragraph of instructions.

Example of tone:

Write 3 more product descriptions in the same style as this one:

"Soft as a Sunday morning. Our cotton hoodie is pre-washed, so it feels
broken-in from day one. Oversized fit. Machine-washable. Made to be lived in."

Products: linen shirt, wool beanie, canvas tote bag.

The AI picks up the short sentences, the sensory opening and the rhythm.

Example of format:

Turn my notes into this format:
Name - Role - One key point they made

Example:
Sara - Designer - Wants fewer colours on the homepage

Tips for good examples

  • Use one to three examples. For most everyday tasks, that is plenty.
  • Vary them if you want variety in the output. If all your examples are about coffee, the AI may drift toward coffee.
  • Say what to copy. "Copy the tone and length, not the content."
  • Use real examples you like, such as a past email that worked well or a post that performed well.

Worked example: a job application

Without context:

Write a cover letter for a data analyst job.

With context and an example:

Write a cover letter for a junior data analyst role at a healthcare
charity in Birmingham.

About me: recent maths graduate, built an Excel dashboard tracking
volunteer hours for my university society, basic SQL.

What they want (from the job ad):
---
(paste the key requirements)
---

Style: confident but humble, short paragraphs, under 300 words.
Here is an opening line I like the style of:
"Numbers only matter when they help people make better decisions."
Don't reuse that line; match its directness.

This letter will mention the real dashboard, match the job's requirements and sound like a person, not a template.

When the AI gets something wrong, add context

If the answer misses the mark, the fix is often more context, not different wording. Ask yourself: what did the AI not know that it needed to? Then add it: "Note: our customers are mostly retired people, so avoid slang."

Summary

  • Context answers who, what, why and what matters.
  • Paste real material, clearly labelled and cleaned of private details.
  • Use one to three examples to show style or format, and say what to copy.
  • When results miss, add the missing context.

Context that you set once: projects and custom instructions

Typing the same background into every chat gets tiring. Most assistants now let you store context:

  • Custom instructions or personalisation settings: a short description of who you are and how you like answers ("I run a small bakery in Lahore; keep answers short; use British English").
  • Projects, Gems or custom assistants: a workspace where you upload reference files (your price list, brand guide, FAQs) and write standing instructions. Every chat in that workspace can use them.
  • Memory: some assistants can remember facts from earlier conversations, and let you view, edit or switch that memory off.

Feature names and availability change and differ between free and paid plans, so check your assistant's settings. We cover these in depth in module 4. The principle stays the same: the AI can only use context it can see.

Hands-on: the context checklist

Before an important prompt, run through these five questions and add a line for each one that matters:

1. Who am I?                 "I'm the office manager at a 12-person law firm."
2. What's the situation?     "Our printer lease ends next month."
3. What's the goal?          "Choose between renewing and buying."
4. What are the limits?      "Budget 2,000 GBP; must scan to email."
5. What material exists?     [paste the two quotes, with prices]

Then add one example if style or format matters.

Worked example: examples that fix a tone problem

A UK charity asks for volunteer thank-you messages and keeps getting stiff, corporate text. Adding two real messages that volunteers loved fixes it instantly:

Write 5 short thank-you messages for volunteers who helped at our food
bank this month. Match the warmth and length of these two real messages
(copy the style, not the content):

1. "Saturday wouldn't have happened without you. 212 families went home
   with full bags. Thank you, Priya."
2. "You turned up in the rain and still made everyone laugh. That's the
   spirit of this place. Thank you, Tom."

Volunteer names and what they did: [list]

When examples backfire

Examples are powerful, so they can mislead. If both examples mention numbers, every output will include a number. If both are about food, the AI may drift toward food. Vary your examples and say what to copy: "Copy the length and warmth; the specific details should come from my list."

How to know you gave enough context

Look at the answer for invented details: a made-up company name, a guessed price, a generic audience. Each invention marks a gap in your context. Fill the gap, and the invention disappears.

Video lecture: Giving context and examples

Giving context and examples — Context: what a helpful colleague would need to know Imagine asking a colleague to "draft a reply to this client". A good colleague would ask: Who is the…

11 chapters · about 8 minutes · captions and full transcript below.

  1. Giving context and examples
  2. What context covers
  3. Paste real material, safely
  4. Show, don't just tell
  5. Examples that work
  6. Store context once
  7. Diagnose with inventions
  8. The five-question context check
  9. Example 1: an out-of-office email
  10. Example 2: 200 product descriptions (illustrative)
  11. Recap

Lecture transcript

Giving context and examples

The single biggest upgrade most people can make to their AI results is not a clever phrase. It is context. In this lecture you will learn what context the AI actually needs, how to paste material safely, how to use examples to show the style you want, and how modern assistants let you store context once so you stop repeating yourself. By the end, you will know how to diagnose a bad answer by spotting what the AI did not know.

What context covers

Imagine asking a colleague to draft a reply to this client. A good colleague would ask: who is the client, what happened, what do we want to happen next, and is anything sensitive? The AI needs the same information, but it will not ask unless you invite it. Useful context covers five things. Who you are. The situation. Your goal. Your constraints. And any relevant material, like the email thread or your price list. A good rule of thumb: include anything that would change the answer. Your company's founding year probably would not. Your refund policy probably would.

Paste real material, safely

Often the best context is the actual document. Paste the client's email, your notes or the product sheet, and label each piece clearly, for example: here is the client's email, then the email between dashes, then here is our price list, then the list between dashes. Labels help the AI separate your instructions from your material. But before you paste, pause. Remove anything private you would not want shared, such as personal phone numbers, ID numbers, passwords or confidential figures, and follow your workplace rules about which AI tools are approved.

Show, don't just tell

Now examples. Sometimes describing a style is hard, but showing it is easy. Say you sell clothing and you love this description: soft as a Sunday morning, our cotton hoodie is pre-washed, so it feels broken in from day one. Paste it and ask for three more descriptions in the same style, for a linen shirt, a wool beanie and a canvas tote. The AI picks up the short sentences, the sensory opening and the rhythm. That would take a paragraph of instructions to describe, and it still might not work.

Examples that work

A few rules make examples work. Use one to three for everyday tasks. Vary them, because the AI copies everything they share. If both examples mention a number, every output will include a number. If both are about coffee, the outputs may drift toward coffee. And say what to copy: copy the tone and length, not the content. A UK food bank charity used exactly this approach. Its thank-you messages to volunteers kept coming out stiff and corporate. Adding two real messages that volunteers had loved fixed the warmth instantly.

Store context once

Typing the same background into every chat gets tiring. Most assistants now let you store context. Custom instructions hold a short note about who you are and how you like answers. Projects, and similar workspaces, hold standing instructions and reference files for a recurring job, such as one client or one product line. And some assistants have memory, which remembers facts from earlier chats and lets you view or delete them. Names and features vary by product and plan, so check your settings. The principle never changes: the AI can only use context it can see.

Diagnose with inventions

Here is a powerful diagnostic habit. When an answer misses the mark, look for invented details: a made-up company name, a guessed price, a generic audience. Each invention marks a gap in your context. Ask yourself: what did the AI not know that it needed to? Then add it. For example: note, our customers are mostly retired people, so avoid slang. The fix is usually more context, not cleverer wording.

The five-question context check

A quick exercise to build the habit. Before your next important prompt, run five questions. Who am I? For example, I am the office manager at a twelve-person law firm. What is the situation? Our printer lease ends next month. What is the goal? Choose between renewing and buying. What are the limits? A budget of two thousand pounds, and it must scan to email. And what material exists? Paste the two quotes. Five short lines, and the assistant can give a real recommendation instead of a generic buying guide.

Example 1: an out-of-office email

A simple example first. You want an out-of-office email. Without context, you get a generic template. Now add context in one line: I run a small tutoring business, I am away from the fourth to the eleventh, parents can book next month's sessions through our website, and urgent questions go to my assistant, Hira. And add one example sentence you like the tone of: Thanks for your message, I am recharging and will reply when I am back. The AI now writes something that fits your business and sounds like you. Context supplied the facts. The example supplied the voice. That is the division of labour to remember.

Example 2: 200 product descriptions (illustrative)

Now a business scenario with illustrative numbers. A skincare seller in Riyadh has two hundred products and needs short descriptions for a new online store. Writing them by hand would take days. She pastes three descriptions she loves as examples, and gives context: our customers are women aged twenty-five to forty, we never make medical claims, prices are in riyals, and each description is under fifty words with one sensory detail. Then she feeds ten products at a time. The output matches her style closely. She spot-checks one in five for accuracy against the supplier sheet, and fixes a couple of invented ingredients she caught. If each description would have taken ten minutes by hand, two hundred products is over thirty hours of work; with this approach it becomes a few hours of reviewing. Those numbers are illustrative, but the pattern is real: examples scale your voice.

Recap

To recap. Context answers who, what, why and what matters. Paste real material with clear labels, after removing private details. Use one to three varied examples and say what to copy. Store stable context in custom instructions or projects. And when results miss, look for inventions and fill the gap. Try this now: take a reply you need to write this week, paste the original message with private details removed, add three sentences of context and one example of a past reply you liked. Next lesson, we learn the art of the follow-up.

Key takeaways

  • Context (who you are, the situation, the goal, constraints) is the biggest upgrade for most prompts.
  • Include anything that would change the answer; leave out what would not.
  • Paste real material with clear labels, after removing private details.
  • One to three varied examples show style or format better than long descriptions.

Try it

Take a reply you need to write. Paste in the original message (with private details removed), add three sentences of context and one example of a past reply you liked.