Feature request: Restore Canvas as a persistent, model-independent workspace
I have returned to ChatGPT after a pause to find that Canvas has been deprecated. Others have already given cogent reasons for retaining it, but several aspects of its value seem to be missing from the discussion.
Canvas was not simply a place where ChatGPT displayed longer pieces of text. It provided a shared, persistent object beside the conversation. The chat contained discussion, alternatives and discarded ideas. The Canvas contained the accepted document.
In other words, Canvas preserved canon.
That distinction is essential for long-running creative and professional projects. I use ChatGPT for world development, story development and screenwriting. During discussion, I might consider five different solutions to a structural problem. Only one of those solutions should enter the project document. Canvas allowed us to explore freely in the chat, then implement the chosen decision in a separate, authoritative text.
A conversation could also contain multiple named Canvases. One chat could cover several related parts of a project, each maintained as a separate document and brought together later. Because those Canvases existed outside the chronological chat stream and could be opened by name, they remained navigable.
Before ChatGPT could access other conversations within a Project, I used multiple named Canvases in a single chat to develop different aspects of a fictional culture. Separate Canvases covered material that would later become chapters or subdocuments, while the conversation allowed those subjects to interact. Each Canvas accumulated the accepted material relating to its particular subject, and the parts could eventually be brought together into a larger document.
Project-wide access to previous chats improves retrieval, but it does not replace that structure. Previous discussions contain speculation, corrections, rejected possibilities and superseded ideas. Access to every discussion is not the same as access to an organised body of canonical documents. Canvas provided the beginnings of a project architecture, not merely additional context for the model.
Yes, Canvas had problems. ChatGPT could sometimes “jump the gun” and edit the document while we were merely discussing a possible change. That was annoying, but it was an implementation problem that could be addressed through better edit controls, confirmation options, undo and version history. It did not invalidate the underlying workflow.
Writing blocks do not reproduce that workflow. They place the document inside the conversation, so discussion and content are once again mixed together. The block becomes buried as the conversation continues. When I scroll back to inspect a passage, submitting a prompt takes me to the bottom of the chat and away from the text I am discussing. Reproducing the complete block after every revision creates multiple competing versions and still leaves the document buried in the conversation.
Writing blocks are chronological rather than navigable. They are content stored somewhere in the conversation, not named project objects that can be opened directly. A series of writing blocks cannot reproduce a collection of separate, addressable Canvases within one working conversation.
Generated documents solve a different part of the problem. They can provide a durable file, but on the web they are not a shared, directly editable surface in the way Canvas was. Moving the text into Word, LibreOffice or another external editor removes the integrated collaboration that made Canvas useful in the first place.
Canvas supported a natural human working method: two people looking at the same document, discussing it, changing it and immediately inspecting the result.
To use a CAD analogy, “I think a classical portico would make the main entrance look amazing” is an exploratory conversation. “This column needs to be at least 450 mm in diameter to bear the weight of the portico” is a precise intervention in a shared design. The second conversation works because both people can see the model and know exactly which element is being discussed.
Chatting around the coffee machine is not the same as collaborating on a design document. Improving the naturalness of the conversation does not replace the shared drawing table.
The same distinction applies across devices. On mobile, I often use ChatGPT conversationally. I might sit with the phone beside me and verbally discuss story ideas. On desktop, I do the actual writing and turn the ideas I have chosen into project canon. These are different venues and different workflows, even though this change between venues and interaction style might occur within the same chat.
Mobile users already understand that a phone, tablet and desktop provide different working environments. Feature parity does not require every device to have an identical interface. Canvas could open full-screen on a small display while retaining its side-by-side document and conversation view on desktop.
If Canvas was removed to maintain consistency across devices, that is the tail wagging the dog. Crippling the desktop experience to match the limitations of a small touchscreen makes no more sense than downgrading Call of Duty’s graphics to resemble Angry Birds.
For many of us, the combination of conversation and a persistent working document was a major part of ChatGPT’s value proposition. Without it, the workflow becomes:
-
Discuss or generate text in ChatGPT.
-
Transfer the accepted material into another application.
-
Maintain the authoritative document manually.
-
Bring extracts back into ChatGPT whenever further discussion is needed.
At that point, ChatGPT is no longer part of the working document. It is merely another chat box beside it.
My feature request is therefore to restore Canvas, or provide a genuine successor with the same core interaction model:
-
Multiple persistent, named documents kept separate from the chat.
-
The ability to create and work across several documents within one conversation or Project.
-
Stable access to those documents from a Project or sidebar, rather than from one point in the chat history.
-
The ability to refer to each document by name.
-
Side-by-side document and conversation on desktop.
-
A suitable full-screen document view on mobile.
-
Direct manual editing and AI editing of the same text.
-
Precise local changes without regenerating the entire document.
-
Version history and a clearly identifiable canonical version.
-
Independence from any particular model, so retiring a model does not retire the workspace.
-
Reliable import and export in Markdown, plain text, DOCX and ODT formats.
There is also an urgent continuity issue. Existing Canvas material should remain readable and exportable after Canvas and its associated models are deprecated. Users should not discover that information stored in old project Canvases has become inaccessible because the interface or model capable of displaying it has been retired.
Canvas was not just an editor. It was where discussion became decisions, where separate parts became an organised project, and where decisions became the authoritative work. Writing blocks preserve pieces of text. Canvas preserved the work and its structure.