How to Actually Use Custom GPTs Inside ChatGPT Projects
A client of ours runs a separate ChatGPT Project for every client they work with — sensible practice if you’re managing several accounts and want your chats kept apart. We’d also just built them a handful of Custom GPTs for specific, repeatable tasks. Their question was straightforward: how do you use a Custom GPT without leaving the Project it belongs to?
We didn’t know the answer off the top of our heads. So we tested it properly, and this article — and the video above — walks through exactly what works, plus the one sequencing mistake that quietly breaks the whole thing.
Quick framing: Project vs Custom GPT
If you’re already using both, skip ahead. If not, here’s the short version: a Project is an ongoing folder that groups your chats, files, and context together for one piece of work — a client, in this case. A Custom GPT is a tool built for one specific job, with its own instructions and, often, its own reference material. On the surface they look like they should just merge. They don’t, automatically.
If you want the fuller comparison between the two, we’ve covered it in a separate video: Custom GPT vs Projects in ChatGPT: Which Should I Use?
What happens if you start with the Custom GPT
The instinct is to open the Custom GPT directly and start typing. If you do that, the chat sits outside the Project. It shows up in your recent chats instead, and there’s currently no way to pull it back into the Project afterwards. You lose the very organisation you set the Project up for in the first place.
The method that actually works
Start the chat inside the Project first — even if it’s just a plain message to begin with. From there:
- Give the chat some initial context, the same way you normally would inside a Project
- Type @ and select the Custom GPT you want to bring in
- Hand off the specific task to it
In our example, we started by asking for a welcome email to a fictional company. Once that existed, we brought in a Tone Guide Reviewer GPT with @, asked it to review the email for tone, and then asked it to rewrite the draft. The GPT applied its own instructions — reviewing what worked, what drifted from the tone of voice guide, and what to change — entirely inside the Project’s chat history. Go back to the Project folder afterwards, and the whole exchange is sitting there with everything else for that client.
The one rule that breaks it
This only works in one direction. If you start the chat by including the Custom GPT first — rather than starting plainly and then @-mentioning it — it won’t appear in the Project’s history. It’ll sit in your recent chats instead, disconnected from the client’s other files and context, with no way to move it into the Project afterwards. Start in the Project. Every time, even if the first message is nothing more than “start a chat.”
What this means in practice
Once this is set up correctly, a Custom GPT stops being a separate piece of functionality you have to remember to go and open. It plugs straight into whatever client work you’re already doing, able to use everything already sitting in that Project. For anyone running several Custom GPTs across several client Projects — proposal reviewers, tone checks, QA tools — that’s the difference between those tools actually joining up with live client work and quietly operating as a separate system you’re switching between manually.
It’s worth treating this as a workaround rather than a guaranteed platform feature — OpenAI could change how Projects and Custom GPTs interact at any point. But for now, getting the order right is a one-time habit that’s worth building properly.
