Skip to main content
Claude starts every reply from nothing. It does not remember you, your project, or what you asked five minutes ago. Everything it knows for this one answer arrives in a single bundle of text, assembled the moment you press send: standing instructions, your files, the conversation so far, and finally the message you typed. That bundle is the context. This section teaches you to see it, place it, clean it, and spend it on purpose.

​
X-ray one turn

You ask a short question inside a project that has two documents attached. Here is what Claude actually receives for that one reply. Click each layer to inspect it.
Three mechanics fall out of that picture, and they drive everything else in this section:
  1. The window is finite. Claude’s working memory is large. The newest models on paid plans hold about a million tokens, roughly 750,000 words. But it is a window, not a warehouse. Fill it and something has to give.
  2. Everything is re-read, every turn. Claude does not remember message three. It re-reads the entire bundle each time it answers, including the sixty pages you pasted an hour ago.
  3. You pay in tokens. On claude.ai the currency is your usage limit. On the API it is dollars. Either way, the meter runs on everything in view, not just on what you typed.

​
Why this section pays for itself

Most people try to get better answers by rewording the question. The bigger lever is deciding what Claude has in front of it when it reads the question. Manage the window well and you get sharper answers, longer productive sessions, and far fewer “you’ve hit your limit” surprises. Manage it badly and no amount of clever wording will save you. Three words to hold onto:
TokenThe unit Claude reads and writes. Roughly three quarters of an English word.
Context windowEverything in view for one reply. Big, finite, and yours to manage.
BurnWhat a turn costs: everything in the window, re-read, plus the reply.

​
What’s in this section

Context is the real prompt

Same question, three desks, three different answers. The desk decides.

Where context lives

Chat, project, memory, or connected source. The home decides the lifespan.

Hygiene

Long threads rot quietly. Read the vitals and reset before they do.

Token literacy

Slice text into tokens yourself, then watch a conversation print its receipt.

Compression

Put a flabby request through the press. Dense beats fluffy, every time.

Lab: Context vs no context

Five minutes. Same question with and without context. Compare quality and cost.

Quiz: Context & cost

Ten instructor-led questions before you leave this section.