If you use ChatGPT or Claude regularly, you have probably done this: opened a new chat, re-uploaded the same document, retyped the same instructions, and burned through part of your daily limit before getting any real work done. There is a better way, and it is hiding in plain sight — the Projects feature.

Projects turn a chat assistant into a persistent workspace. Set up properly, they also help you get far more out of your message limits and tokens. Here is how they work, and how to use them well.

What exactly is a "Project"?

A Project is a dedicated space inside an AI tool that bundles three things together:

• Custom instructions — standing rules that apply to every chat in that space (tone, role, format, audience)
• A knowledge base — files and documents the AI can reference in any conversation within the project
• Related chats — all conversations on that topic, grouped in one place

Instead of starting from zero every time, the AI enters each conversation already knowing your background material and preferences.

Projects in ChatGPT

ChatGPT Projects (available across OpenAI's plans, with file limits varying by subscription) let you group chats, upload files, and set project-level instructions. When you ask a question inside a project, ChatGPT prioritises that project's chats and files over general context, which keeps responses focused and relevant. Paid users can even enable project-only memory, so context from one client or workstream never bleeds into another.

See OpenAI's official guide: Projects in ChatGPT

Projects in Claude

Anthropic's Claude offers Projects with the same core idea: project instructions plus a knowledge base of documents, code, or notes that every chat in the project can draw on. Each project supports a large context window (roughly a 500-page book's worth of material), and on paid plans, Retrieval Augmented Generation (RAG) kicks in automatically when your knowledge base grows — expanding capacity up to 10x by retrieving only the relevant parts of your documents instead of loading everything at once.

See Anthropic's documentation: What are Projects?

Other tools: Gemini Gems and beyond

Google's Gemini has Gems — saved configurations with a name, instructions, and knowledge files, essentially the same concept with a different label. Most serious AI platforms are converging on this pattern because it solves a universal problem: context that should persist, but doesn't.

Learn more: Gemini Gems overview

The part most people miss: Projects save your limits and tokens

This is where Projects go from "nice organisational feature" to genuinely valuable. Every AI plan comes with usage limits, and those limits are consumed by tokens — the chunks of text the model reads and writes. Here is how Projects help you spend them wisely:

1. No more re-uploading files. Every time you paste a 20-page document into a fresh chat, the model processes all of it again — and that counts against your usage. Upload it once to a project's knowledge base and it is simply there for every future conversation.

2. Cached knowledge costs less. Project documents are cached after the first use. In Claude, for example, only new or uncached material counts fully against usage limits, so repeated work inside a project is significantly cheaper than repeating the same context in standalone chats.

3. Shorter chats, cleaner context. Long, meandering conversations get expensive because the model re-reads the entire history with every message. Projects encourage a healthier pattern: start a fresh, short chat for each task, and let the project's instructions and files supply the background. You get the continuity without the token bloat.

4. RAG retrieves only what's needed. When knowledge bases grow large, retrieval-based systems pull in only the relevant sections of your documents rather than the whole library, which results in more capacity, fewer wasted tokens.

5. Instructions you never retype. "Write in British English, keep it under 500 words, cite sources" — typed once into project instructions instead of pasted into every chat. Small savings that compound over dozens of conversations.

Quick setup checklist

• Create one project per ongoing area of work (a client, a course, a blog, a codebase)
• Write clear, specific project instructions — role, tone, format, and any standing rules
• Upload only the documents that matter; trim outdated files monthly
• Start a new chat inside the project for each distinct task instead of one endless thread
• Move related old chats into the project so everything lives in one place

The bottom line

Projects are the difference between renting an AI assistant by the hour and hiring one that actually knows your work. They keep your context organised, your outputs consistent, and — crucially — your tokens and message limits working on new problems instead of re-reading old ones. If you are on any AI plan with usage caps, setting up Projects is the single easiest optimisation you can make today.

References

1. OpenAI Help Center — Projects in ChatGPT
2. OpenAI Academy — Using Projects in ChatGPT
3. Anthropic — Collaborate with Claude on Projects
4. Claude Help Center — What are Projects?
5. Google — Gemini Gems