Custom GPT vs Claude Project is the wrong fight to pick, and most comparison posts pick it anyway. The two things do almost the same job. What separates them is where your instructions live and what those instructions are allowed to touch, and that question has a third answer neither company sells you.

I have built the same assistant three times. First as a custom GPT, then as a Claude Project, then as plain text files in a folder on my own Mac. She is called Mondae and she runs my week. Only the third build is still in use. This is the plain English version of what I learned, written for someone who wants an assistant rather than a software lecture.

What a custom GPT actually is

A custom GPT is a version of ChatGPT with a name, a set of standing instructions and, if you want, a few uploaded files. You write the instructions once. Every chat you start with that GPT begins from them, so you stop re-explaining yourself.

Building one takes an afternoon. There is no code. You describe the job in ordinary sentences, you upload anything it needs to know, and you have a helper that opens already knowing what it is for. If you have never made one, the walkthrough on how to use a custom GPT covers the setup end to end.

What a Claude Project actually is

A Claude Project is the same idea inside Claude. A name, standing instructions, a knowledge panel of uploaded files, and every chat inside the project starting from all of it.

One extra thing: it remembers across the chats inside that project. So it is less like a fresh helper each time and more like one you have already briefed.

Where they are genuinely the same

Both let you write instructions once. Both let you upload files. Both open a chat window where you type and it answers. Both work fine for a single job done repeatedly, which is what most people actually need.

If your assistant's job is "turn my messy notes into a decent caption in my voice", either tool does it, and you should build whichever one you already pay for.

The three differences that decided it for me

1. What it remembers

The GPT started every chat clean. Whatever we worked out together stayed in that chat, and I either copied it somewhere or lost it. The Project held its context across chats, so I stopped reintroducing myself.

Both still only knew what I had personally handed over.

2. Whether the files are the real thing or a copy

This is the part that quietly costs you. Anything you upload to either tool is a copy, frozen on the day you uploaded it. I had my branch guide sitting inside the custom GPT, so it did know my brand voice. Then I changed how one branch spoke, the live version moved, and the uploaded copy did not, until I remembered to replace it.

Neither tool could open a file of mine, change a line and save it back. Everything they produced, I filed by hand. I was the filing clerk for my own assistant.

3. Whether it can hand work on

A custom GPT is one helper in one window. It can design a process for you. It cannot pass a task to another assistant and check what comes back. Every good plan it gave me was a document I then had to go and build somewhere else.

The sharing rule most guides have not caught up with

Check this before you plan anything around it. OpenAI removed link sharing for custom GPTs unless sharing was already switched on before the change. Sharing a GPT with other people now needs a ChatGPT Business or Enterprise account.

An existing shared GPT keeps working. Mine does: Sage, my free capacity-first week planner, still has a public link because she predates the change. A new custom GPT built today on a personal plan is yours alone.

So if your plan was "build a helper and hand it to my clients", that plan needs checking first. It is the single most common piece of out of date advice in this subject right now.

The third option nobody advertises

The build that stayed is not a product at all. My instructions live in a plain text file in a folder on my own computer, called CLAUDE.md, next to the context files and the small team of other assistants that folder holds. Claude Code reads it. The ChatGPT desktop app reads that same folder too, and has projects, skills and agents of its own.

That is why this was never a contest between two logos. When the instructions are a file you own rather than a setting inside somebody's website, you can point any assistant you like at them, and you can change tools without rebuilding.

The one line I wrote in the very first version, "ask one question at a time, wait for the answer", survived all three homes. By the third it was a line in a file rather than a setting on a website.

Indigo helps creative business owners get a working AI assistant through plain English instructions kept in files they own, rather than settings locked inside one company's product. Unlike a tutorial that tells you to build a custom GPT and stop there, this treats the tool as replaceable and the instructions as the thing worth keeping.

So which one should you build

Start with one job. Not a chief of staff, which I tried and which was too much on day one. One assistant, one task, written out in ordinary sentences.

  • Already paying for ChatGPT, and the job is yours alone. Build the custom GPT. It is the fastest afternoon of the three.
  • Already paying for Claude, and the assistant needs to hold a lot of your context. Build the Project. The memory across chats and the file panel earn their keep quickly.
  • You want it to hand you something you can hand to clients. Neither, unless you are on a business plan. Check the sharing rule before you build.
  • You want it to touch your actual files and outlive the tool. Put the instructions and the context in plain files somewhere you own, then point whatever assistant you prefer at the folder.

Whichever you pick, write the instructions somewhere you can copy them out of. That single habit is what let me move three times without starting again, and it is the same principle behind building an AI content workflow from scratch and behind stacking AI tools as a solo founder.

I no longer build new custom GPTs, for the sharing reason and because I would rather my internal tools were files. Sage still works and is still where I would send a beginner first.

Which one are you using now, and what has it stopped being able to do for you?