How much context should I give an AI?

Enough that a capable stranger could do the task: the text to work on, one or two sentences about the situation, and any rules. More than that rarely helps; less than that is the usual cause of generic answers.

Can I paste a whole document into a prompt?

Yes, up to the tool's limit, which varies by plan and changes over time. Put your instruction first, then the document, then any rules. For very long documents, split them or use a plan with a larger limit.

Is it safe to paste personal information into AI?

Treat it like emailing a company you do not know. Leave out passwords, card numbers, and other people's private details. The model rarely needs them to do the task well.

Giving AI what it needs to know — How to Talk to AI

The most common reason an answer is generic is that the model had nothing specific to work with. It does not know who you are, what you are looking at, or what you already tried. Everything it knows about your situation is what you typed. This lesson is about typing the right things.

Three kinds of context, and how to give each one.
Kind	What it is	How to give it
The text	The article, email, notes or document you want it to work on	Paste it in full, after your instruction, with a blank line before it. Put quotes around it if it is short.
The situation	Who you are, who the answer is for, what has already happened	One or two plain sentences. "I am replying to a customer who has already complained twice."
The rules	What it must do, must not do, or must keep	Short, direct statements. "Do not add facts that are not in the text." "Keep the two dates."

Pasting text without confusing the model

1
Instruction first
Say what to do before the text, so the model knows what it is reading for.
2
Blank line, then the text
A gap makes it obvious where your words stop and the material starts.
3
Say what it is
"Here is the article:" or "My notes:" is enough. The model then treats it as material, not as instructions.
4
Rules at the end
Constraints placed after the text are the last thing the model reads, and they stick better there.

Try it

Same article, real context

Write a three-sentence summary of this article for a neighbourhood newsletter.

Context: the readers are local residents and shop owners. Several of the shop owners are worried about the parking. I want the summary to be fair to both sides, not cheerful.

Here is the article:

Local council approves new bike lanes on Harbour Road

The council voted 7 to 2 on Tuesday to approve protected bike lanes along the 2.4 km stretch of Harbour Road between the bridge and the shopping centre. Construction is expected to begin in February and take around five months, during which one traffic lane in each direction will be closed. The project is budgeted at $3.1 million, with $1.8 million coming from a state active-transport grant. Opponents on the council argued that the loss of 46 on-street parking spaces would hurt shops along the route; the council has promised a review of parking after twelve months. A survey of 1,200 residents earlier this year found 61% in favour of the lanes, with support strongest among those under 40.

Rules: use only facts from the article. Include the start date and the parking review. Under 80 words.

Three free runs a day on the site key. Add your own key for unlimited runs.

Now delete the Context paragraph and run it again. You will get an accurate summary that reads like a press release. The facts were the same; the situation was missing, so the model picked the default tone. Context is not extra. It is the difference between a summary and *your* summary.

Knowledge check

You paste a customer complaint and ask for a reply, and the model writes a cheerful response that offers a refund you cannot give. What was missing?

Giving AI what it needs to know

Lesson FAQ

Stop overpaying for AI tools