Giving context

Context is the difference between AI that gives generic answers and AI that tailors answers to your specific situation.

Every chat starts from zero

When you open a new conversation, AI knows nothing about your work, your team or your documents. Instructions and memories are a few loose rules and facts, not knowledge of how you work. And AI was never trained on your internal world: your team structure, your reporting templates or the abbreviations your department uses every day.

New chat

Copilot doesn't know who you are.

Message Copilot

Copilot is not trained on your organisation

internal documents
roles and teams
work processes
culture and jargon

Without context → generic output

With context → relevant output

Colleague vs. language model

Think about asking a colleague for help versus asking AI. Your colleague already knows the background. AI does not.

Colleague	Language model
Knows your organization	Remembers loose facts at most
Understands your jargon	No knowledge of your situation
Familiar with the history	Guesses at meaning
Gets implicit expectations	Uses general patterns

Two sources of information

AI works with exactly two things: what it learned during training, and what you give it right now.

Training data. Billions of words of text up to a cutoff date. This is where its general knowledge of language, facts, and patterns comes from.

What you give it. Your prompt, plus any files or documents you attach. This is the only way AI learns about your specific situation.

Keep in mind. Anything after the training cutoff date is invisible to the model. Your organization, your colleagues, your processes: AI knows none of it unless you spell it out.

Where do you give context?

There are three places. Which one you choose depends on how often you need the same explanation.

In your prompt

The context building block: who is it for, what is going on, what should Copilot watch out for. For a task you do once.

Prompting in 3 building blocks

In a context document

Everything about a process that keeps coming back, in one file. You attach it to every task in that process.

How to write a context document

In your instructions and memories

Copilot always applies instructions, and memories only when they are relevant. They are loose rules and facts, not knowledge of your processes or your team.

Set up instructions

Managing context

Like any conversation, the longer it goes, the harder it is to keep track. After a long conversation, AI mixes up earlier information more easily.

Draft an outline for the March newsletter.

Here is an outline with four sections and a short intro.

Add the team outing, on 14 March.

Added under Agenda.

Can you also write an email to our suppliers?

Sure. Here is a draft for the suppliers.

Make that email more formal.

Updated, with a formal greeting.

Back to the newsletter: when was the outing?

I believe the team outing is on 21 March.

The more messages you add, the vaguer the start of the conversation becomes.

Start with a clean chat. Start new tasks in a new chat. Old context from previous conversations can unintentionally influence your new output.

Be selective with attachments. Only add relevant documents that directly help with your specific task. Not your entire archive: too much context reduces quality. Attach a document as a file, not as a link. With a link, Copilot often reads only part of it.

Split complex tasks. Create multiple focused prompts instead of one long one. This keeps the context clear and improves results per subtask.

Save successful prompts. Copy working prompt+context combinations to a prompt library for reuse.