Providing relevant context

Why context matters

A model fills every gap in your prompt with the most typical assumption. If you ask for "a welcome message for new members", you get a generic welcome for a generic organisation. Context replaces those typical assumptions with your facts.

Five kinds of useful context

1. Purpose — why you need the output. "This summary will help my manager decide whether to renew the contract" leads to a different summary than "this is for my own revision notes".
2. Audience — who will read or use it, what they know already and what they care about.
3. Situation — relevant facts: the organisation, the event, the constraints, what has already happened.
4. Source material — the actual text, data or notes the answer should be based on. Asking a model to work from material you provide is one of the most effective ways to reduce invented content.
5. Preferences and standards — style guides, terminology, things that worked before, things to avoid.

Text description

Context for a prompt branches into 6 parts. Purpose: Why you need the output; The decision it supports. Audience: Who will read or use it; What they already know. Situation: Relevant facts and constraints; What has already happened. Source material: Text, data or notes to work from; Reduces invented content. Preferences and standards: Style guide and terminology; What worked before, what to avoid. Keep out: Personal data and passwords; Confidential information; Detail that would not change the answer.

Example

Without context: "Write a message reminding people about the meeting."

With context: "Write a short message reminding volunteers about Saturday's planning meeting for the community garden (10:00, in the library meeting room). Most volunteers are retired and prefer clear, friendly language. Last time, people forgot to bring their availability for the summer rota — remind them to bring it. Keep it under 80 words."

The second prompt cannot be answered generically: purpose, audience, situation and a specific past problem are all there.

How much context?

More is not always better. Each piece of context should change the output in a way you want. Ask: if I removed this sentence, would the answer get worse? If not, remove it.

Too little context gives generic answers. Too much — especially irrelevant detail — can distract the model, push important instructions out of a long context window, or cause it to mention things you never wanted mentioned.

Separate context from instructions

When you paste material, make clear where it starts and ends and what the model should do with it. A simple way is to label sections and use delimiters such as triple quotes or headings:

Task: Summarise the customer email below in 3 bullet points for our support team.

Customer email:
"""
(paste email here)
"""

You will formalise this structure in Module 3.

Context you should NOT give

Do not paste personal data, passwords, confidential business information, or other people's private messages into an AI tool unless you are authorised to do so and the tool's data terms allow it. Often you can anonymise instead: replace names with roles ("the client", "Employee A") and remove identifying numbers. Module 10 covers this in depth — for now, make anonymising a habit.

Asking the model what it needs

When you are unsure what context matters, you can ask: "Before you answer, ask me up to five questions about anything you need to know to do this well." This clarifying-questions pattern turns the model into an interviewer. Answer its questions, then let it complete the task.