Do more beforeyou hit the limit

Chapter 01

How your usage actually gets spent

Most people hit a limit and assume they used Claude too much. Usually they used it in a way that costs far more than it needs to. Once you see what is being counted, the fixes are obvious.

Every message re-reads the whole chat

Claude does not remember a conversation the way you do. Each time you send a message, the entire thread goes back in: every earlier message, every reply and every file you attached. Message forty in a long chat costs many times more than message four, even if you only typed one line.

Paid plans run on a rolling window

On the paid Claude plans your usage is counted inside a session window that resets every five hours, with a weekly limit on top. You can see where you stand under Settings, then Usage. If you work in long bursts, the window matters more than the daily total.

Bigger models spend the window faster

Opus is the strongest model and it uses your allowance fastest. Sonnet handles most everyday work at a fraction of the cost. Haiku is the quickest and cheapest. Choosing the right one per task is the single biggest lever you have.

Chapter 02

Keep the context small

Context is everything Claude is reading on a given turn. Less context means faster answers, sharper answers and more messages before you hit a limit.

Start fresh when the task changes

If you finished the sales email and now want help with a hiring post, open a new chat. The old thread adds nothing to the new task and you pay for it on every message.

Carry forward a summary, not the whole thread

Before you leave a long chat, ask: Write a short handoff note with the decisions we made, the open questions and the next step. Paste that note into the new chat. You keep the useful part and drop the rest.

Paste the part that matters

• Send the three paragraphs you want edited, not the whole document.
• Send the error and the function it came from, not the whole codebase.
• Send the columns you need from a spreadsheet, not every tab.

Use Projects for files you reuse

If you keep uploading the same brand guide or price list, put it in a Project once. Every chat inside that Project can use it without you attaching it again.

Chapter 03

Pick the right model for the job

Treat the models like staff with different day rates. You would not pay your most senior person to rename files.

Haiku

Quick, cheap and good at simple, well-defined jobs: sorting, tagging, short rewrites, pulling fields out of text, first-pass summaries.

Sonnet

The default for real work. Writing, analysis, research, most coding and most business tasks. If you are not sure, start here.

Opus

Save it for hard problems: long multi-step reasoning, tricky code across many files, strategy where the details matter. Use it on purpose, then switch back.

Extended thinking

Thinking mode helps on problems that need working out, like planning, math and debugging. It also spends more of your allowance, so leave it off for everyday writing and lookups.

Chapter 04

Prompt for fewer, better replies

Every back-and-forth costs a full re-read of the chat. A prompt that gets it right the first time is cheaper than three that get there eventually.

Say the four things up front

The goal, who it is for, the format you want back and how long it should be. Example: Write a follow-up email to a lead who went quiet after a demo. Audience: a busy agency owner. Format: subject line plus under 120 words. Tone: friendly, direct, no pressure.

Show one example

One sample of what good looks like beats a paragraph describing it. Paste a past email you liked and say "match this voice".

Ask for the change, not the whole thing again

When you want an edit, ask Claude to return only the part that changed. Rewrite the second paragraph only. or Show me just the lines you changed. A full rewrite of a long document burns output for no reason.

Outline first, then expand

For anything long, ask for an outline, fix the outline, then ask for the full draft. Fixing a five-line outline is cheap. Fixing a two-thousand-word draft is not.

Stop it early when it goes off track

If the first lines of a reply are heading the wrong way, stop the response and redirect. Waiting for a long wrong answer and then correcting it costs you twice.

Batch your questions

Five related questions in one message are cheaper than five messages, because the chat is only re-read once.

Chapter 05

Set up projects that do the work for you

Good setup means you stop repeating yourself. The instructions live in one place and every chat starts already knowing them.

Project instructions

Write a short brief for each Project: who you are, who your customers are, how you write and what to avoid. Keep it to the rules you would give a new hire on day one. Long instruction files get read on every turn too.

Lean knowledge files

Add the documents Claude genuinely needs, such as your offer sheet, your voice guide and your FAQ. Remove anything stale. A tidy Project is faster and more accurate than a crowded one.

One Project per area of the business

Marketing, sales, operations and hiring each get their own. That keeps each set of files and rules focused on the work it serves.

Skills for repeat workflows

If you do the same multi-step job every week, turn it into a skill: a short set of instructions Claude loads only when the task calls for it. You get the same result every time without pasting the steps.

Connectors instead of copy and paste

Connectors let Claude read from tools like your drive, calendar or CRM directly. Pulling the one record it needs is cheaper and more accurate than pasting an export.

Chapter 06

Files, research and long documents

Text beats scans

A PDF full of images is heavy to read. If you only need the words, export or copy the text and send that instead.

Split big documents

Working on a long contract or report? Send it in sections and ask for a summary of each. Then work from the summaries.

Use research mode when you need it

Web search and deep research are powerful and they read a lot. Turn them on for questions that need current facts, and leave them off for writing and thinking work.

Huge source sets

If you need to ask questions across dozens of long documents, a tool built for that, like NotebookLM, will often do the job without touching your Claude allowance at all.

Chapter 07

Claude Code without the waste

Claude Code reads files, runs commands and keeps a long working context. The same rules apply, with a few specific habits.

• Plan before it edits. Use plan mode to agree the approach first. Undoing a wrong change costs more than planning it.
• Clear between tasks. Run /clear when you move to a new task so the old one stops riding along.
• Compact long sessions. Run /compact to squash a long session into a summary and keep going.
• Write a CLAUDE.md. Put your build commands, folder layout and conventions in it so Claude does not have to rediscover them every session.
• Point at files. Name the file and the function you mean. Asking it to search the whole repo for something you already know the location of is wasted reading.
• Switch models with /model. Use the lighter model for routine edits and the strongest one for hard debugging.
• Hand searches to subagents. Big searches can run in a subagent so only the answer comes back into your main session.

Chapter 08

If you use the API

Building your own tools on the API? Two features cut costs sharply.

Prompt caching

If every request starts with the same long instructions or documents, cache them. Cached input is billed at a fraction of the normal rate and comes back faster.

Batch processing

For work that does not need an instant answer, like tagging a thousand support tickets overnight, send it through the batch API at a discount on the standard price.

Right-size every call

Route simple calls to Haiku, everyday calls to Sonnet and only the hard ones to Opus. Set a sensible maximum output length so a runaway reply cannot drain your budget.

Chapter 09

The daily checklist

• One task per chat. New task, new chat.
• Start in Sonnet. Move to Opus for the hard part only.
• Goal, audience, format and length in the first message.
• Paste the part that matters, not the whole file.
• Ask for the change, not a full rewrite.
• Long chat? Get a handoff note and start fresh.
• Reused files live in a Project, not in every chat.
• Check Settings, then Usage, before a big session.