ChatGPT Projects: How to Organize Your Work in One Place

Starter · Level 1 of 3

A ChatGPT Project is a workspace inside ChatGPT that keeps related chats, files, and instructions together so the model stops losing the thread of recurring work. Instead of starting a fresh chat every week and re-explaining who your client is or how you want things formatted, you set that context once at the project level, and every conversation inside the project inherits it. To set one up: open the left sidebar, click New project, name it, add custom instructions, upload your reference files, and start chatting inside it. The rest of this guide walks through each step with a worked example.

What is a ChatGPT Project, and why does context keep getting lost?

First, the platform this assumes. You need a ChatGPT account at chat.openai.com. Projects appears in the left sidebar of the web app and the desktop and mobile apps. It is available on the Free plan and the paid plans (Plus, Go, Pro, Business, Enterprise), with file and project limits that scale up as the plan does — the Free tier is the most restrictive. If you do not see Projects yet, update the app and check again, since the rollout has reached plans in stages.

A normal ChatGPT chat has no memory of your other chats. Each new conversation starts cold. If you run the same kind of task every week — a status report, a set of product descriptions, a recurring email — you end up pasting the same background, the same tone notes, and the same source documents over and over. That repetition is where the model drifts: one week it formats the report one way, the next week another way, because nothing is holding the context steady.

A Project fixes that by attaching three things to a container instead of to a single chat:

• Custom instructions that apply to every chat in the project (on top of your global instructions).
• Files — PDFs, spreadsheets, docs, images, or pasted text — that any chat in the project can read.
• The chats themselves, grouped under the project so the work stays in one place instead of scattered across your history.

Open a new chat inside the project and it already knows your standing instructions and can see your files. You stop re-explaining, and the output stays consistent week to week.

Project vs. a regular chat, at a glance

Behavior	Regular chat	Inside a Project
Remembers your standing context	No — re-paste each time	Yes — set once
Shared reference files	Per chat, re-upload each time	Project-wide, upload once
Keeps related chats together	Scattered in history	Grouped under the project
Custom instructions	Global only	Global plus project-level

What do you need before you start?

Three things make a project worth setting up:

1. A task you repeat — weekly, monthly, or just often enough that re-explaining it is annoying.
2. The reference material that task always needs: a brand guide, a client brief, a spreadsheet of data, last month’s version of the document.
3. A clear idea of how you want the output to look. You will write this down once as custom instructions.

If you already keep this kind of reference material in a notes system, pulling the key facts into a short project file is quick. If you do not, this is a good prompt to start — a tidy set of linked notes makes a strong source of truth for any AI workspace. Our guide on how to use Obsidian for linked notes covers a simple way to organize that.

How do you create a ChatGPT Project?

The exact labels shift slightly between updates, but the flow is stable:

1. Open ChatGPT in the browser or app and look at the left sidebar.
2. Click New project (near the top of the sidebar, above or beside your chat history).
3. Give the project a clear name — name it after the recurring task, not the tool. Weekly Client Report — Acme beats Reports.
4. Pick a color or icon if offered. This is only so the project is easy to spot in the sidebar.
5. Confirm. The empty project opens, with spots to add instructions and files.

That is the container. It does nothing useful yet — the value comes from what you put in it next.

How do you add files and custom instructions?

Inside the open project, look for the project settings — usually an Add files button and an Instructions or Add instructions field.

Files. Click Add files and upload the reference material the task always needs. Keep it to what is actually relevant — a client brief, a style guide, the data spreadsheet. You can paste plain text instead of uploading if the material is short. Every chat you open inside the project can read these files, so you upload once rather than per conversation.

Custom instructions. This is the part most people enable and never configure. The instructions field is where you tell the model how to behave for this project specifically — the role it should play, the format you want, the tone, and anything it should never do. These stack on top of your global custom instructions and apply to every chat in the project.

Write them as plain, direct rules. Vague instructions produce vague output. Spell out the structure you want and give the model the constraints it needs to stay consistent.

A worked example: a weekly client report project

Here is a full setup you can copy. The recurring task: every Friday you write a short status report for a client, Acme, based on that week's notes. You want the same structure and tone each time without re-briefing the model.

Step 1 — Create the project. New project, name it Weekly Client Report — Acme.

Step 2 — Add files. Upload two things: the client's one-page brief (so the model knows who Acme is and what the engagement covers), and last week's report as a format reference (so it copies the structure).

Step 3 — Add custom instructions. Paste this into the project instructions field and edit the bracketed parts:

Role: You are a project manager writing a weekly status report for our client, Acme.

Audience: Acme's marketing lead. Professional, concise, no filler.

Always structure the report in exactly these sections:
1. Summary (2-3 sentences, plain language)
2. Completed this week (bullet list, past tense)
3. In progress (bullet list, with expected finish date)
4. Blockers / needs from Acme (bullet list; write "None this week" if empty)
5. Next week's focus (3 bullets max)

Rules:
- Keep the whole report under 300 words.
- Use the client's own product names exactly as written in the brief file.
- Do not invent metrics or dates. If a number is not in my notes, write "[confirm]".
- Match the tone and section order of last week's report in the files.
- Second person is fine when addressing Acme; never use "I".

Step 4 — Use it. Each Friday, open a new chat inside the project and paste only that week's raw notes:

This week's notes:
- Finished the homepage redesign, shipped Wednesday
- Started the email template build, ~60% done, expect to finish next Tue
- Waiting on Acme to approve the new logo files before I can update the footer
- Next week: finish email templates, start the blog migration

The model already has the brief, the format, and the rules, so it returns a clean report in your exact structure. You never re-explain who Acme is or how the report should look. Next Friday you open another chat in the same project and paste the new notes. The setup pays for itself by the second week.

One detail worth noticing: the instructions tell the model to write [confirm] instead of guessing at any number it was not given. That single rule prevents the most common problem with AI-written reports — confidently stated figures that nobody actually verified.

How does project memory work?

Two things to understand so you set expectations correctly.

First, files and instructions are project-wide. Every chat inside the project sees them. This is the behavior you want — it is what keeps the output consistent.

Second, chats inside a project are still separate conversations. The model does not automatically remember what you discussed in a different chat within the same project unless that information lives in the files or instructions. If a decision matters across weeks — say, a formatting choice you settled on — add it to the project instructions or a project file, not just a one-off message in a single chat. The container holds what you put into its instructions and files; it does not silently absorb everything you have ever typed.

This is the same principle behind any good system for repeated work: the durable context lives in a defined place, and the day-to-day work references it. The same logic powers most time-saving setups, including the ones in our roundup of workflow automations that save hours a week.

What are the common mistakes and limits?

• Treating instructions as optional. A project with no custom instructions is just a folder. The instructions are what make the output consistent — write them.
• Overloading files. Uploading ten documents when two are relevant makes answers worse, not better. Give the model only what the task needs.
• Expecting cross-chat memory. As covered above, separate chats in a project do not share their conversation history. Put anything that must persist into the instructions or files.
• Hitting plan limits. Free and lower paid tiers cap how many files a project holds and how many projects you can keep. If you bump into a ceiling, prune old files or consolidate projects rather than assuming the feature is broken.
• Naming projects after tools instead of tasks. A project called ChatGPT stuff helps no one. Name it after the job it does.

The short version

A ChatGPT Project moves your standing context — the files, the format, the rules — out of individual chats and onto a container that every chat in the project inherits. Set it up once for any task you repeat, and the model stops re-asking and stops drifting. Start with one recurring job, write clear project instructions, upload only the files that job needs, and open a fresh chat inside the project each time you run it.

If you want the reference material behind your projects to be just as organized as the projects themselves, a linked-notes system is a good next step — see the beginner’s guide to Obsidian for a simple way to keep your source-of-truth notes in order.

Evidence and verification

Last verified: July 12, 2026. This guide was checked against OpenAI’s official Projects documentation and OpenAI Academy guide. Interfaces, plan limits, pricing, and feature availability can change. Confirm any feature or cost that determines your setup before relying on it, and test the workflow with a non-critical example first.