The Claude and ChatGPT connection is a door into your workspace, not an assistant we sell you. You add GoFeed as a connector, pick one workspace on the consent screen, and your own assistant schedules, approves, imports and answers questions as you, holding exactly the permissions your role holds and not one more.
A door, not an assistant
The distinction matters more than it sounds, so it is worth settling before anything else. There is no model of ours here. Nothing is generated: no caption, no image, no copy. There is not a single language model call anywhere in this layer, and that absence is the proof that the product does not write for you.
What opened is a door. The intelligence is yours, on the subscription you already pay Claude or ChatGPT for, and on the other side sits your workspace with its brands, its calendar and its files. When you ask your assistant to move a brand's week around, it decides what to say; what checks that you have permission, that the network accepts that format and that the piece still goes through the usual approvals is GoFeed.
Put differently: what changed is not that there is now AI in the product, it is that your AI stopped working blind. Until now you pasted a list of posts into a chat and copied the answer back by hand, typos included. Now it reads what is there and writes where it belongs, and the translation step in the middle, which is where the hours went, disappears.
It is also why this does not age with us. The day a better model arrives you switch it in your own app and the door is unchanged, because the door does not know what is on the other side and does not need to.
Connecting it, in three screens
You add it from the connector screen in Claude, on the web, on the desktop or in Claude Code, and from the equivalent screen in ChatGPT, using your workspace address. There is no key to copy and no token to paste into a file: the sign in is the GoFeed one, the same you use every morning.
Then comes the consent screen, and the two things that define this connection happen there. The first is that you pick one workspace, a single one: a connection is a workspace, and another workspace needs another connection. The second is that you see written down how far it will reach, which is exactly as far as you reach. The step by step, with screenshots of both apps, is in the guide to connecting your AI.
A connection being one workspace rather than one account is not lazy scoping. Somebody running three workspaces, their own and two partners', does not want an assistant hopping between them because it misread a sentence. If you want to work in the other one, you open the other connector, and the act of changing places stays yours.
Which areas it reaches
We are not going to print a number of tools, and not out of modesty: the catalogue grows every few weeks and any figure written here would be stale before you read it. What matters is what it opens onto, and those are the same areas you work in with a mouse.
| Area | What you can ask for |
|---|---|
| Calendar and posts | See the week, create a piece with its time and networks, reschedule what slipped |
| Approvals | Pull up what is waiting, approve it or send it back with a written reason |
| Inbox | Go through messages, comments and reviews, and reply |
| CRM | Find companies and contacts, move opportunities, load a whole CSV |
| Drive and uploads | Tidy folders and put large files away without pushing them through the chat |
| Tasks and briefs | Create and assign tasks, read what the client answered in their brief |
| Analytics | Ask about the month, about one network, or about a single piece |
| Notifications | Know what is waiting without opening the app |
The door hands out no modules. It reaches as far as your workspace and your role reach: if your workspace does not have a module, your assistant cannot see it, and if your role cannot write there, your assistant cannot write either. A creator workspace has no clients to ask about, and that is as true from the chat as it is from the screen.
Reading never asks, writing always does
This is the rule that makes the whole thing usable without being frightening. Looking things up does not interrupt: asking how many pieces are scheduled this week, or how the month's reach went, is answered and that is that.
Changing something is different. Every write asks for your confirmation in the chat, with what it is about to do laid out, and lands in the history of the piece or the record it touched under your name. There is no robot user to blame afterwards: your account did it, at your request, and the history says so.
And a post created this way skips nothing. It goes through the same approvals as one typed by hand, with the same wait if the brand has approvals switched on, because it would be absurd for the door to be a way around the control you put on your own team.
Everything in this article happens in one place in GoFeed.
Try it freeTwo things that only make sense once you try them
The first is large files. A four gigabyte video does not travel through a chat conversation, and pretending otherwise is the usual way an integration looks complete and falls over on its first real day. Here the assistant hands you an upload link, the file goes straight into the folder you named, and your storage quota counts it exactly as it always did.
The second is the CRM bulk import. There is no CSV import in the interface, and through this door there is: thousands of rows with progress as it goes, a warning on every row that does not add up, and the option to undo the whole batch. It is the route for bringing a full book of contacts over from another tool, and it is covered end to end in the article on importing your data.
What people actually use it for
The three audiences open with different things, and none of them is "write me a caption", which is precisely what this door is not needed for.
An agency with twenty brands uses it to avoid twenty tabs: asking what is still unscheduled this week across every brand at once, and filling the gaps by typing. That is the crosswise question the screen handles badly, because the screen is built to work one brand at a time.
A creator uses it to close the week in one sitting without opening the calendar: describing the pieces already filmed, saying which days they should go out, and confirming the proposal one piece at a time. A business almost always opens with the import, because it is the quickest way to bring in a contact base that has been sitting in a spreadsheet for years.
There is a fourth use, less impressive and probably the best: asking. What happened last month, which brand dropped, which piece beat its own average. Those are questions nobody asks when they mean opening four panels, and that ask themselves when typing them is enough.
With your permissions, and nobody else's
The assistant has no account of its own here, and that is more than a security detail: it is what makes the module safe to hand to the whole team. Each person connects their own, and each connection inherits the role of whoever created it, so somebody who cannot invoice does not invoice by talking either, and somebody who only sees one brand still sees one brand.
It is worth reviewing your roles before handing this out, not because of the AI but because this is the moment people find out what was badly shared all along. How permissions are organised is set out in the roles and permissions article.
How to turn it off
Two ways, both immediate. Each person sees their own connections and cuts whichever they want whenever they want, with nobody to notify. And the workspace owner has a switch that closes the whole door for everybody, which exists for a bad night rather than for daily use.
On top of that the module is switched on per plan, so a workspace that has not bought it has nothing to turn off. What is worth deciding in advance is who may connect and with which role, because this is one of those features that spreads through a team by itself the moment somebody demonstrates it in a meeting. What the whole thing looks like screen by screen is on the module page.