Using ChatGPT Projects: keep work and instructions together

Using ChatGPT Projects is useful when you keep pasting the same explanation into a chat. Every Monday you may start with who you are, which client you work for, what tone you want and which files belong with the job.

In ChatGPT, there is a place where you keep that context together for each task. It is not about a magical memory that remembers everything perfectly. It is more like a work folder where chats, files and instructions stay together.

What are ChatGPT Projects and what can you do with them?

A project is a folder where chats, files and instructions for one topic are kept together. ChatGPT can use that context for ongoing work, such as writing, research or planning, as OpenAI explains in the Help Center on Projects.

You can think of a project as a fixed workspace inside ChatGPT. Not every separate question sits somewhere else, but belongs to the same task. That brings peace of mind, especially when your work spans several conversations.

Examples include client work, an annual plan, a recruitment round or a weekly report. This is especially familiar at the end of October. Many people are then working on annual plans, budgets and Q4 reports, with the same background information again and again.

A project can also help when you often need the same tone or structure. Think of newsletters, client emails, policy notes or social posts. You set the fixed agreements in the project and then start a new chat within that workspace each time.

Important: a project does not make ChatGPT perfect. It mainly helps you organise context. You still need to check the answers, especially facts, names, figures and sensitive information.

Using ChatGPT Projects: how do you create one step by step?

Using ChatGPT Projects starts with creating a project, choosing a name, adding instructions, uploading files and then starting your first chat inside that project. Menus can differ by version, subscription and workspace.

According to OpenAI, Projects and project sharing are available to signed in ChatGPT users, but availability depends on your subscription and workspace settings. So always check your own ChatGPT environment and OpenAI’s guidance if you cannot see an option.

1. Create a new project. In ChatGPT, choose the option to make a project. Dutch media, such as Bright, described Projects as a way to bring order to your ChatGPT conversations.
2. Give the project a clear name. Think of Annual plan, Weekly newsletter or Client X reports. You may also be able to choose a colour or icon if that option is available.
3. Add project instructions. Put in what ChatGPT should always know for this task. For example, audience, tone, language, length and structure.
4. Upload only files you need often. Think of a style guide, sample text or product list. Do not use a huge pile of old documents if they are no longer accurate.
5. Start your first chat inside the project. Ask your question as if you are in the right workspace. Then you do not need to paste all the fixed context again each time.

A good project name matters more than you might think. Do not just call it Marketing or Work. Choose one clear task or client, so you can quickly see where everything is later.

If you work in a business setting, the settings may be different from a personal account. An employer can decide which functions are available. If in doubt, ask your administrator or check your organisation’s AI policy.

Which instructions and files should you put in a project?

Put short instructions in your project about role, audience, tone and structure. Also add a few files you need again and again, such as a style guide, sample text or product list.

Good project instructions are specific. So do not just say that the text must be good. Instead, say who the text is for, how long it may be and what structure you expect.

The same goes for files. Less and cleaner is often better than lots and messy. An up to date style guide helps more than many old documents with conflicting style.

If you often use separate prompts, you can move from separate instructions to fixed project instructions. For that, also see the explanation with prompts for your work, but think of a project as the place where you keep recurring agreements.

A handy basic template is this:

You can put this text in your project instructions and adjust the placeholders. Keep the instruction short enough to remain readable. If everything is important, nothing is really important.

Mini checklist: is this worth a project?

• Do you do this work more than once?
• Do you keep using the same context, tone or files?
• Do you need several chats for the same task?
• Do you want to keep conversations separate by client, task or report?

If you answer yes to several of these, a project is probably useful. If it is just one quick question, a single chat is often enough.

What is the difference between a project, memory and a custom GPT?

A project is a workspace for one task, memory is about you in general and a custom GPT is a fixed assistant you can reuse or share. So do not choose based on what sounds newest, but on how you work.

A single chat is useful for one question. A project is useful when you want to keep several conversations and files together for one task. A custom GPT is a better fit if you want to build a fixed assistant for a recurring role, for example for a team or process.

Part	Single chat	Project	Custom GPT
Designed for	One question or short task.	A workspace per task, client or topic.	A fixed assistant for reuse.
Instructions saved	Only within that chat.	Yes, as project instructions within the project.	Yes, as the assistant’s fixed instructions.
Files saved	Only in the context of that chat.	Yes, files stay with the project.	Depends on your settings and subscription.
Shareable	Not as a workspace.	Yes, depending on subscription and workspace settings.	Yes, if you want to share an assistant.
When to choose	For a quick question.	For recurring work with the same context.	For a fixed assistant you want to reuse or share.

If you mainly want a workspace per task, choose a project. If you want to build a fixed assistant with its own role, read on about making your own AI assistant.

ChatGPT’s memory is something different. It is not just about one project, but about information ChatGPT can use more broadly, depending on your setting. A project helps you define context per task.

What about privacy and memory in a project?

When you create one, you choose between standard memory and project only memory. You can change this later in the project settings, although OpenAI says changes may take several hours.

Standard memory means the project can work with ChatGPT’s broader memory. Project only memory means the project works separately from memory outside the project. OpenAI describes this choice in the explanation of Projects.

If you switch to project only memory, information from that project is removed from the memory used outside the project. Chats and files remain in the project. That matters if you want to keep work separate by client or topic.

Shared projects are automatically set to project only memory. They cannot be switched to standard memory. They also do not have access to a member’s personal context, custom instructions or memories outside the project.

Pay extra attention to personal data in the Netherlands and Belgium. The GDPR applies to personal data in uploads. So do not upload customer data, staff files or other sensitive information without clear agreements.

Use a business account if your employer has one. Also check your organisation’s AI policy. Organisations that use AI must also pay attention to AI literacy under the AI Act.

A practical rule is this: upload only what you really need. Anonymise where possible. For more on risks, see the article on using AI safely.

Using ChatGPT Projects with colleagues: can you share a project?

Yes, projects can be shared, depending on subscription and workspace settings. A shared project does not see any member’s personal context, instructions or memories outside the project.

This is useful when a team works from the same style guide. Think of marketers using the same newsletter style, or a project team working on the same report. Everyone then works from the same project context.

The boundary is important, though. A shared project is not a gateway to someone’s personal ChatGPT memory. The shared project uses project only memory and therefore stays separate from personal settings outside the project.

Make clear agreements before you share. Agree which files may go in, who cleans up outdated files and which kinds of data must not be uploaded. That way, a useful workspace does not slowly turn into a messy dumping ground.

Does it also work with Claude or Copilot?

For Claude, yes: Claude has a similar feature, also called Projects. According to Claude Academy, these are independent workspaces with their own memory, chat history, knowledge base and custom instructions.

In Claude, uploaded documents are available in all chats within that project. In practice, that is similar to the idea of a workspace per topic. The details differ by service, so do not compare the features one to one without checking.

We do not have verified facts about Copilot in this plan. So use the documentation for your own work environment if you want to know whether there is a similar feature. Settings can differ by workplace, especially in organisations.

Practical example: weekly newsletter

This is what using ChatGPT Projects looks like in a recurring task. Marloes is a marketer at a Dutch webshop for garden furniture. She writes the newsletter every week and used to paste the same explanation into a new chat each time.

Now she creates a project called Weekly newsletter. In that project, she puts these instructions:

You help me write a weekly newsletter for a Dutch webshop selling garden furniture. Write in English, informal but professional, using you and your. Maximum 150 words. Structure: subject line of up to 50 characters, opening of two sentences, three products with one sentence each, a clear button text. Do not use exclamation marks. Ask for the weekly offer first if it is missing.

Then she uploads the style guide and two earlier good newsletters. These are files she needs all the time. She does not upload her entire marketing folder, because it contains too much old and possibly unnecessary information.

In a new chat within the project, she now only types:

Week 45: 15 percent off parasols, new: Bergen lounge set.

The expected result is a draft newsletter in the house style, without Marloes having to paste all the context again. Even so, she always reads the draft herself and adjusts it. The result can differ each time, even with the same project instructions.

Common mistakes when using ChatGPT Projects

The biggest mistake is seeing a project as a digital attic. A good project is tightly defined, up to date and set up safely.

Putting everything into one project

A project with all clients, reports and ideas mixed together quickly becomes unusable. It is better to make one project per task or client. That keeps the context clear.

Uploading too many or outdated files

Old files can give the wrong examples. Upload a few current, clean files and clear things out regularly. Remove what is no longer correct.

Using vague instructions

An instruction like write well is not very helpful. Name the role, audience, tone, length and structure. Then ChatGPT knows better what you mean.

Uploading confidential data without thinking

Personal data and customer data need extra care. Anonymise, follow your employer’s policy and choose carefully between standard memory and project only memory.

Frequently asked questions

Are ChatGPT Projects free?

Availability depends on your subscription and workspace settings. Check OpenAI for the latest options and terms.

Does a project remember everything I say?

No, that is not guaranteed. A project uses the chats and files in the project as context, according to the chosen memory setting.

Can I share a project with a colleague?

Yes, depending on your subscription and settings. A shared project uses project only memory and does not have access to personal context outside the project.

Which is better: Project or custom GPT?

Choose a project for your own workspace per task. Choose a custom GPT if you want a fixed assistant to reuse or share.

Create one project today for the work you repeat most often, paste the instruction template into it and test it for a week. If you then want to go further, read how to go about making your own AI assistant.

• ChatGPT
• Privacy
• Prompts