ChatGPT Projects or Custom GPTs: which should you start with?

After a few useful ChatGPT sessions, most beginners hit the same question: should I keep working in normal chats, create a Project, or build a Custom GPT?

The short answer: usually start with a Project. Use it to collect one ongoing piece of work, such as a client project, course, campaign, or weekly internal routine. Build a Custom GPT when the routine is clear enough to become its own reusable assistant.

This guide continues from our first ChatGPT guide about writing a useful prompt. Now we take the next step: saving the workflow so you do not have to rebuild it every time.

Quick answer: ChatGPT Project or Custom GPT?

Use a ChatGPT Project when the work belongs in one concrete place: a client, topic, school project, campaign, or weekly routine. A Project can keep related chats, files, instructions, memory, and tools together around longer-running work.

Source: OpenAI Help Center – Projects in ChatGPT

Use a Custom GPT when you want a reusable assistant for a specific task. A GPT can have its own instructions, conversation starters, uploaded knowledge, and selected capabilities. It fits better when several people, clients, or repeated situations should use the same pattern.

Source: OpenAI Help Center – GPTs in ChatGPT

If you are unsure, start with a Project. Test the routine there. When you notice that you do the same thing the same way several times, consider turning it into a Custom GPT.

What you will learn in 10–20 minutes

You will learn:

• when a Project is simpler than a Custom GPT
• when a Custom GPT is worth the extra setup
• where to find both tools in ChatGPT
• how to test the choice with a small weekly workflow
• the prompt you can copy when you want ChatGPT to help you decide

We are not covering advanced developer features, publishing, or larger automation setups here. Those can wait. The goal is simply to choose the right place for recurring ChatGPT work.

Where to find the tools

For a Project:

1. Go to chatgpt.com and sign in.
2. Look in the sidebar.
3. Choose New project.
4. Name it after the real work, for example "Weekly client update" or "Course: civics".

OpenAI describes Projects as workspaces for longer work where instructions apply only inside that project. Project instructions can therefore be more specific than your global instructions.

Source: OpenAI Help Center – Projects in ChatGPT

For a Custom GPT:

1. Go to chatgpt.com/gpts or open Explore GPTs in ChatGPT.
2. Choose Create.
3. Build the GPT with the conversational builder or the configuration view.
4. Test it in Preview before you save or share it.

OpenAI says creating or editing GPTs requires a paid subscription. Building and editing happen on the web; mobile apps can use GPTs but cannot build or edit them.

Sources: OpenAI Help Center – GPTs in ChatGPT and Creating and editing GPTs

The workflow: a weekly update that keeps coming back

Imagine a solo consultant or small team lead who writes the same short update every Friday: what happened, what is blocked, what happens next week, and which decisions need checking?

This is a good test case because it is concrete, recurring, and easy to review.

Step 1: start as a Project

Create a Project with a clear name, such as "Weekly client update".

Add short instructions. Keep them plain:

You help me write weekly updates for this project.
Write in a calm, clear tone.
Ask if dates, names, decisions, or next steps are missing.
Always end with a short review checklist before I send the text onward.

Only add source material you are allowed to use. That might be a public service description, an empty template, or a sanitized note without customer secrets. If you later connect real tools or data, start with scoped permissions and clear approval steps. Secrets, passwords, and keys do not belong in a chat or uploaded file.

Step 2: run the first weekly update

Start a new chat inside the Project and write something simple:

Here are my cleaned notes from the week:
[paste notes]

Write a first draft of the weekly update.
Mark anything that seems uncertain.
End with a checklist of names, dates, decisions, and next steps I need to verify.

The point is not that ChatGPT writes perfectly. The point is that you create a repeatable loop: source material in, draft out, human review before anything is sent.

Step 3: decide whether it should become a Custom GPT later

Keep it as a Project if the work changes from week to week, if the material belongs to this specific client or course, or if you are still experimenting with the instructions.

Consider a Custom GPT when the routine is stable. For example, you may want a "weekly update reviewer" that always asks the same control questions, uses the same tone, has a few fixed conversation starters, and can be used by more people.

This is where many people start too early. They build a GPT before they know what the assistant should actually do. A Project is often the better workshop first.

Simple decision map

Ask these questions:

• Is this about one ongoing topic, project, client, course, or campaign? Choose Project.
• Is this a task that should repeat the same way in many situations? Custom GPT may fit.
• Do you need related chats and files in one place? Choose Project.
• Should other people use the same configured assistant? Look at Custom GPT.
• Are you still experimenting? Start with Project.
• Can you describe the steps without thinking too hard? The routine may be ready for a Custom GPT.

Do not use ChatGPT Memory as a filing cabinet. OpenAI describes memory as a place for high-level preferences and details, not exact templates or large blocks of text. Put rules in instructions and source material in Projects or GPT knowledge.

Source: OpenAI Help Center – Memory FAQ

Copy this prompt: let ChatGPT help you choose

I want to create a recurring ChatGPT workflow for [task].

Help me decide whether I should start with a Project or a Custom GPT.
First ask up to five questions about:
- whether this is one specific project or a reusable routine
- what source material I use
- whether other people should use the same setup
- what level of privacy and review is needed

Then give me:
1. your recommendation
2. the first setting I should configure
3. a short instruction I can add to the Project or GPT
4. what I should leave for later

Common beginner mistakes

• Building a Custom GPT before the workflow is stable.
• Uploading lots of files while leaving the instructions vague.
• Putting behavior rules in knowledge files instead of instructions.
• Sharing a GPT without testing example outputs, permissions, and any apps or Actions.
• Forgetting the review step. Names, dates, decisions, and sensitive details still need checking.

OpenAI also says GPTs with apps or external APIs may send relevant input to third-party services. If you go there later, start with limited access and clear approvals, not broad permissions.

Source: OpenAI Help Center – GPTs in ChatGPT

Next step

If you followed our first guide to ChatGPT prompts, this is the natural next step: choose a home for the recurring prompt.

If your material is mostly PDFs, notes, and course material for studying, read our NotebookLM guide for students. That guide shows a study workflow where you work from clear sources and check answers against citations.

Want to compare this with Claude? Read the guide to Claude Projects as a recurring workspace. It shows the same basic idea in Anthropic's tool.

Want more simple AI workflows when they are ready? Watch for the Hammer Automation newsletter and contact us if you want to introduce ChatGPT Projects or Custom GPTs in a team. We can help you choose the first clear workflow and set sensible review steps.