Give the right context

What you will learn

• Which context to include: background, materials, what you already know and what to avoid
• What a context window is, and why long chats can lose details
• How to remove sensitive data before you share material

The idea

The AI only knows what is in the conversation, plus general patterns it learned in training. It does not know your company, your class, your earlier drafts or your preferences unless you say so. Context is the information you add so that it works on your actual situation, not an average one.

Useful context falls into four kinds. Background: the situation, who is involved and why the task exists. Materials: the text, notes or data the answer should be based on, pasted in or uploaded. What you already know: so it does not explain the basics or repeat what you have tried. What to avoid: topics, words, sources or approaches you do not want.

There is a limit to how much the AI can take in at once. The context window is the amount of text a model can consider at one time, including your messages, any files and its own replies. It is measured in tokens, small pieces of words. When a chat or a document goes past this limit, earlier or extra parts may be dropped or handled less carefully, sometimes without a clear warning. For long material, send it in parts, ask for a summary of each part, or start a new chat with a short recap.

Context is also where private data slips in. Before you paste, remove or replace names, contact details, ID numbers, passwords, customer data and anything your company or school does not allow in outside tools. Look at your privacy settings, the options that control whether your chats are stored or used to improve the service.

Example

A team lead wants help writing feedback after a project. Here is a first try:

Write feedback for my team about the project.

The result could fit any team and any project. Here is the same request with context:

Help me write feedback for my team about a website redesign project
that finished last week.

Background: four people, eight weeks, launched on time. The main
problem was that design changes kept arriving late.
What I already know: I want to thank the team for the testing work
and suggest a cut-off date for design changes next time.
Avoid: naming individuals, blaming anyone, corporate jargon.

Notes:
- Testing found most bugs before launch
- Design changes came in until two days before launch
- The team stayed late in the final week

The first prompt can only produce general feedback. The second gives the AI the facts, the material and the lines it should not cross. The notes contain no names, so nothing personal leaves the team.

Practice now

1. Pick a real task where the answer depends on your situation, such as a reply to a message or a summary of your notes.
2. Copy the material you will use into a plain text document first.
3. Replace names, contact details and other private data with labels such as Person A or Client B.
4. Write one line each for background, what you already know and what to avoid.
5. Send it using the template below, then compare the answer with your original material.

At the end you should have an answer based on your situation, and a cleaned copy of your material you can reuse.

Copy this template

Help me with [task].

Background: [the situation, who is involved, why this matters].
What I already know: [what I have tried or do not need explained].
Avoid: [topics, words, sources or approaches to leave out].

Use only the material below. If something you need is missing, ask
me instead of guessing. Mark anything you are unsure about.

Material:
[paste your material here, with names and private details removed]

Asking AI to flag uncertainty does not make its answer correct. Keep checking the parts that matter.

Check the output

• Does the answer use your material, or general statements that could fit anyone?
• Did it follow your "avoid" list?
• Did it skip the basics you said you already know?
• With long material, did it miss details from the start or the end? Compare with the original.
• Did any private detail end up in the output that should not be there?

Common mistakes

• Assuming the AI remembers earlier chats. Some tools keep memory between chats and some do not. Restate the key context, or check how your tool handles memory.
• Pasting everything you have. Too much material can crowd out what matters and use up the context window. Pick the parts the task needs.
• Leaving private data in pasted material. Once it is sent, you may not be able to take it back. Clean it first, and follow your company or school rules.
• Forgetting to say what to avoid. The AI fills gaps with common choices. If something is off limits, say so.