Skip to content

Shared AI workspace

One space the AI keeps working inside — same files, same standing instructions, same running chats — so it holds the group's context instead of each person re-explaining from scratch. Everyone opens it in their own Claude or ChatGPT; what one person adds, the rest see. Two flavours of the same thing: Claude Projects and ChatGPT Projects.

Reach for it when the value is a persistent place the group keeps chatting in. Skip it when you're handing over a finished thing to click (Artifact, website) or files to build on (repo) — a project has no code, just context and conversations.

Last verified: 2026-06-07 · Confidence: high on the access model and paid-seat gate (Claude checked against Anthropic's docs; ChatGPT against OpenAI's).

It allows you to

  • Load the same context for everyone. Files and standing instructions sit in the space, so every chat already knows the brief, the tone, the rules — nobody pastes them again.
  • Keep working in one place. Not a one-off output; the group returns to it and the workspace grows over time.
  • Pool what people learn. Refine the instructions once and everyone's next chat picks it up. → fine print.
  • Stay private per person. The project is shared; each person's own chats stay theirs unless they choose to share one. [confirmed]

Ideal for

  • A comms team's house voice — the style guide and twenty past posts plus "write in our voice, never these words", so every drafter's AI sounds like the org from the first message, not the fifth edit.
  • A research group's standing questions — the literature and "always cite the source file", so anyone's chat starts steeped in the lit review instead of re-uploading PDFs.
  • A grants team's funder context — the funder's priorities, past winning proposals, and required sections, so every officer drafts to the same brief.
  • An ops team's institutional memory — runbooks, org chart, the "how we do X" docs, so a new hire's AI answers day-one questions.
  • A forecasting desk's scoring method — baked into the instructions so the approach never drifts between people. ChatGPT frames its own launch as a team campaign workspace; the pattern generalises.

Who can get in

This isn't a public-link tool — a shared project lives inside a paid team workspace, shared within that org. Outsiders can't be pulled in without an admin adding them to the workspace first. [confirmed]

  • Open it to the whole workspace — Claude's "Everyone at [org]", ChatGPT's workspace link. [confirmed]
  • Or name a few. Invite individual members, each set view-only or allowed-to-edit. [confirmed]
  • Lock it back down. Switch to "only people invited", or remove a person — they're out at once. (Anything they already copied stays with them — true everywhere.) [confirmed]
  • Two access levels — view/chat uses the space; edit also changes files, instructions, and invites. Fine print. [confirmed]

Which rungs it can hold. Named people in your org, or the whole org — rungs 2 and 3 only. No just-you, no anyone-with-the-link, no public web rung. → Who can see it? [confirmed]

Handing data to the host. Hosted by Anthropic (Claude) or OpenAI (ChatGPT); on the paid plans these live on, neither trains on your data by default. → Can you trust the company? [confirmed]

What you do to set it up

Your agent can prep the contents, but the sharing is yours — these are web apps, and creating the project, uploading files, and inviting people are clicks only a human can do.

  • Ask: tell Claude Code "draft the standing instructions for a shared team AI project where we do X, and tidy these files to upload." It hands you the instructions to paste and the cleaned files — the delegable half.
  • Then, in the browser (un-delegable): create the project, paste the instructions, upload the files, hit Share. ~10–20 min once. [estimate]
  • You need a paid team seat for the whole-workspace path — Claude Team/Enterprise, ChatGPT Business/Enterprise/Edu. Which plans, and the personal-plan exception. [confirmed]

Full step-by-step, tabbed by provider: Share an AI project. The clicks above are the manual path.

What the other person does

  • Open it in their own Claude or ChatGPT. It shows in their projects list; they start chatting with the context already loaded. ~0–1 min, no setup. [confirmed]
  • They need a seat on that platform and to be in the same workspace — a teammate, not an outsider. New to it? They sign in once.
  • Pay: nothing beyond their own workspace seat — the project adds no cost. (View/chat access can't change files; bump them to edit in the sharing pane.) [confirmed]

Other ways to share

  • A one-off output to click and use, not an ongoing chat space? → a Claude Artifact: opens in any browser, no seat, nothing to set up.
    • One packaged assistant on Google's side, not a shared Claude/ChatGPT space? → a Gemini Gem bundles instructions into a single bot you share by link.
  • A document people read and refine, not the AI's context? → a Google Doc: one tap to open, no account to view, comments in the margin.
  • A reusable routine an agent absorbs, not a chat space? → a Claude Code skill installs into the recipient's own agent and fires across all their work.

Sources

Good to know

  • Updates reach people on their next chat, not mid-conversation, and only an edit-access person can make them. [estimate]
  • There's a file cap per project — ChatGPT scales it with the owner's plan [confirmed]; Claude switches big knowledge bases to retrieval rather than hard-blocking, exact limit unpublished [unclear].
  • ChatGPT per-person sharing now reaches personal plans (since 2025-10-22); the whole-org workspace-link path is still Business/Enterprise/Edu. [confirmed] → full fine print.