ChatGPT Projects vs custom GPTs: what to use now

Side by side

From OpenAI's docs and help center, read 5 November 2026; the help pages through search-result text, because they block automated reads.

A worked example: three GPTs, split

Illustration: a made-up business. A consultant built three custom GPTs and moves them before December:

1. "Calder Foods assistant": the client's background, interview notes and style. That's context for one client, so it becomes a project with the files as sources and the background as instructions.
2. "Proposal reviewer": how she checks every proposal. That's a method she uses for every client, so it becomes a skill.
3. "Workshop agenda writer": used twice in a year. She keeps the instructions in a note and lets the GPT retire.

The proposal reviewer, turned into a skill:

---
name: proposal-review
description: Reviews a consulting proposal before it goes to a client: scope, price against the rate card, dates, and promises we can't keep. Use when someone asks to review, check or tighten a proposal.
---
1. Scope: every deliverable has an owner and a date. Anything vague ("as needed") is flagged.
2. Price: each line matches the rate card. A discount needs a reason written next to it.
3. Promises: no guaranteed outcomes, no dates the plan doesn't support.
4. Return must-fix, should-fix and polish, each quoting the line and giving the fix.

Checks: every finding quotes the proposal; no price that isn't on the rate card.

She tests it on three old proposals, one of them a rushed one, and adds a rule after it misses a date that clashed with the plan.

Move yours

1. List your GPTs and what each one holds.
2. Context for one client or job: a project, with the files as sources.
3. A method used everywhere: a skill, written once.
4. Actions don't move; they need a connector instead.
5. Tell people the new place before 11 December; an old GPT link won't give them access.

Next step: the team working the same way

Once methods are skills, the team needs the same ones. Getting your team to use AI the same way covers that step. For sharing a GPT's replacement outside the workspace, see custom GPTs end on 11 December.

Questions

What's the difference between a ChatGPT project and a custom GPT?

A custom GPT is a separate assistant with its own instructions that you open and share by link; it retires on 11 December 2026. A project is a workspace for one piece of work: its chats, instructions and files stay together, and every chat in it starts from them. Use a project for a client or a job, and a skill for a method you repeat everywhere.

What replaces custom GPTs?

OpenAI's retirement FAQ, read through search results, says a migrated GPT becomes a ChatGPT plugin whose instructions are a skill, and that the plugin starts private. For your own work, the GPT's context usually fits a project and its method fits a skill.

Can I share a ChatGPT project like a GPT link?

Yes, with the Anyone with a link setting, which OpenAI's help, read through search results, describes: any signed-in ChatGPT user who has the link can join. Unlike a GPT link, it brings people into the project itself, with its chats and files, so send it only to people who should see them. You can also invite people one by one, and groups on Business, Enterprise and Edu.