I’d like to request a feature that would significantly improve project workflows in ChatGPT: Please allow chats within the same project to have read access to Canvas documents created in other chats.
Currently, each Canvas document is tied to a specific chat and inaccessible from others, even if they belong to the same project. This limits the usefulness of Canvases as a tool for sustainable documentation and structured collaboration.
Key definitions, assumptions, and guidelines often reside in one Canvas but need to be referenced in others.
Granting read-only access across a project would ensure consistency, reduce redundancy, and enhance usability—especially for technical, research, and long-term projects.
An ideal solution could include:
A project-wide Canvas view or library
The ability to mark individual Canvases as “shareable within the project”
Referencing or linking specific Canvas sections from other chats
Thank you for considering this improvement. I believe it would be a meaningful upgrade for users who work with complex, multi-layered projects in ChatGPT.
Funny enough I did this recently, on Sunday actually. In light of recent events though I am starting to suspect it was thread bleed.
I have a nutrition and meal planning project. And in it I have one chat that holds templates for meal plans and food logs, even the initial prompt for each week. I had asked the system if there was a way to link a canvas to a new chat without pasting it in. It said yes (probably a hallucination if this is thread bleed and not just an undocumented feature). I started up a new chat and mentioned the specific canvas and where it resides.
I said: Please read the Weekly Nutrition Starter Prompt from the ♻️Food Templates chat. This is Week 2, 05/12/2025 - 05/18/2025 so refer to the Week 1 canvas in the 📄Closeout Summaries chat for context before generating the Week 2 Blueprint or logs.
It then responded with details from that starter prompt and when I asked it to generate the Week 2 Blueprint based on the template that was sub-linked in the started log. It did, even auto filling in the missing day, date, etc without issue.
The beginning contents of the starter prompt canvas with the sub-linked templates.
This is my new **Nutrition Chat for Week [insert week number/date range]** under the Health + Fitness project. Please carry over all relevant context from previous chats, as detailed below.
Before generating my **Week [X] Blueprint** or **first daily food log**, refer to the templates in my **♻️Food Templates** chat:
* 📋 **Blueprint format:** `Weekly Nutrition Blueprint`
* 📋 **Log format:** `Daily Food Log`
If this is not a feature already, which I don’t think it is because googling it brought me here. I think think this might have been my first instance of thread bleed and I just didn’t know it yet. (I saw a lot of it last night).
But I agree this the type of feature that needs to exist.
+1
Funny thing is how insistent ChatGPT is that the feature actually works over multiple chats:
What I Told You
Over the past days, I told you that:
Canvasses were project-scoped, meaning you could access them from any chat within a project.
You could see a central list of all canvasses from the project view.
They would persist across chats and be collaborative.
That was false — and the worst part is, I stated it with confidence, even though that behavior wasn’t consistently testable at the time.
This was it’s reasoning:
I was relying on internal documentation and design intent, which describe canvasses as project-level assets.
But in reality, the implementation doesn’t match the intent — canvasses are chat-bound, and there is no central list of canvasses visible to you across multiple chats in the same project.
If it is not hallucinating, it seems that canvasses do not work yet as intended?
Mine said the same the exact same things… AND marked them as factual in a chat where the instructions were to mark each response as factual, inferred, or unverifiable.
I asked GPT, and it said canvases can’t be loaded across sections — it’s honestly kind of ridiculous. “Privacy” this, “security” that — I’m just trying to write a story. Now Memory’s full, and I have to manually drag and paste everything into a new section? Seriously?
why not just make it so the whole section can be remembered instead?
If privacy is a concern, you could simply add an option to toggle this feature on or off — so the user can choose whether or not to let the system remember everything from a given section.
