OpenAI has given you two different ways to organise your AI work and explained the difference to almost nobody. If you have opened ChatGPT recently and noticed "Projects" in the sidebar, then remembered you keep hearing about "Custom GPTs" on LinkedIn but never actually built one, you are not behind. You have just stumbled into one of the more confusing bits of product design in consumer AI.
The honest answer for most non-technical professionals: you need a Project. Custom GPTs solve a problem you probably don't have. Here's the difference in plain English.
What a Project actually is
A Project is a folder for your chats with shared context built in. You set it up once with some instructions about who you are and what you do, you upload any reference files (a style guide, a tone document, last quarter's report), and every new chat you start inside that project inherits all of it. No re-explaining yourself every Monday.
It's the thing most people thought ChatGPT was supposed to do from the start: remember the basics so you can get on with the work. You can have lots of Projects, one for client work, one for personal admin, one for the side project nobody at work knows about. They sit in the sidebar like folders and you click into them when you want their context.
Projects are now available on the ChatGPT free tier with some limits, and the full version sits inside ChatGPT Plus.
What a Custom GPT actually is
A Custom GPT is a saved chatbot designed for one specific repeated task, that you can also share with other people. The builder gives you a longer instructions field, lets you upload knowledge files, and lets you connect to other tools through what OpenAI calls Actions.
The clearest use case: you have a task you do exactly the same way every week, the steps are the same, the tone is the same, the inputs follow a pattern. You build a GPT called something like "Weekly Status Update Writer" that takes a list of bullet points and turns them into your standard update, use it every Friday, and share it with the rest of your team so they do the same.
Custom GPTs also sit in a marketplace called the GPT Store, where anyone can browse and use yours if you make it public. They are also slightly fiddly to build well, which is why most people open the builder, get halfway through, and never finish.
How to pick
A flat rule that works for nine out of ten non-technical professionals:
- Ongoing and varied work. Lots of different chats that all need the same background context call for a Project.
- Repeated and identical work. The exact same task, the exact same shape, every time, especially if you want to share it, calls for a Custom GPT.
- If in doubt, start with a Project. Custom GPTs are the move you make after you've already worked out what your repeatable tasks actually are.
What this looks like by role
A flat rule is helpful. A worked example is more helpful. Here's what Projects and Custom GPTs look like across the kinds of work non-technical professionals actually do.
| Role | Use a Project for | Use a Custom GPT for |
|---|---|---|
| Marketing and comms | Campaign work, content production, ongoing brand or client accounts | Brand voice checker, social caption formatter, subject line first-draft writer |
| Ops and finance | Monthly close, budgeting cycle, vendor management, KPI reporting | Board pack one-pager writer, expense narrative builder, weekly status update generator |
| Management and people leaders | Team planning, performance review cycles, OKR work, hiring rounds | 1:1 prep question generator, feedback redrafter, meeting recap formatter |
| Chief of staff, generalist, EA | A specific stakeholder's ongoing work, a programme you're running | Pre-meeting brief writer, daily diary briefer, weekly email digest formatter |
These are starting points, not a fixed list. Most people end up with a couple of Projects doing the heavy lifting and one or two Custom GPTs they actually use.
Why this is more confusing than it should be
It is not just you. Every major AI company has invented its own name for roughly the same idea. Anthropic calls them Projects too, Google calls them Gems, and Anthropic has also added something called Skills, which is a different thing again. Microsoft has its own system inside Copilot. Same underlying concept, four different names, no two products quite the same.
The reason they don't agree on the naming is that each company is competing for branding, not because the features genuinely do different things in your week. Most non-technical professionals need precisely one container per tool, not all of them. You can safely ignore the rest until you have a specific reason to care.
If you're still working out which tool to use in the first place, that's the question we covered in our piece on ChatGPT vs Claude.
The setup that actually makes this work
A Project only helps you if you've told it who you are and what good looks like. Default ChatGPT, with no instructions, will write you generic copy no matter how cleverly you organise it. The real lift is in the setup layer: the role you do, the tone you want, the things to skip, the constraints to apply.
Once that's in place, Projects become useful. Without it, they're just better-looking folders. And once it is producing usable output, the next question is how hard to check what comes back, which matters more than where you organised the chat.
The AI Starter Kit. AI that actually works for you. ChatGPT and Claude.
Get the Kit