Repository navigation
[ChatGPT Web][Projects] Google Drive-linked Project Source stays stale until manual Resync #41287
Description
What version of the Codex / ChatGPT app are you using?
ChatGPT web at chatgpt.com in a desktop Chrome browser on macOS. This report concerns a Cloud Project's Project Sources behavior, not local Drive for desktop file materialization.
What subscription do you have?
ChatGPT Pro personal subscription.
What platform is your computer?
macOS, using ChatGPT web in Chrome.
What issue are you seeing?
A native Google Doc added to a ChatGPT Cloud Project through its Google Drive link did not expose a later edit through the Project Source representation until I manually clicked Resync on the source card.
The linked Drive object kept the same identity. Direct access to the Google Doc reflected the current edit, but a Project-source-only canary did not see the new exact value before manual Resync. After manual Resync, the same canary returned the new value.
This makes direct Google Drive currentness and the ChatGPT Project Source representation different states, without a clear freshness receipt available to the task.
What steps can reproduce the issue?
- Create a native Google Doc containing a unique canary field and value.
- Add that Google Doc to a ChatGPT Cloud Project as a Project Source using its Google Drive link.
- Confirm that a Project-source-only Chat can read the initial value.
- Edit the same native Google Doc in place and change the unique canary value.
- Start a separate Project Chat and instruct it to use only that named Project Source—no Google Drive connector, web, Library, or memory.
- Ask for the exact updated canary value.
- Observe that the new value is absent from the Project Source representation.
- Open Project Sources and manually select Resync for the same linked source.
- Repeat the Project-source-only canary in a new Chat.
- Observe that the new value is now returned.
What is the expected behavior?
One of the following should happen:
- Drive-linked Project Sources refresh automatically within a documented interval;
- ChatGPT detects the newer Drive revision and marks the source stale;
- the source card displays its source revision, last-refreshed time, and refresh state; or
- the UI and documentation clearly state that a linked source is a cached snapshot requiring manual Resync.
Ideally, a task should receive a source-freshness receipt so it can distinguish direct Drive state from the cached Project Source representation.
What is the actual behavior?
The linked Project Source remained stale until manual Resync. The original Google Doc was current when accessed directly, but the Project-source-only representation did not contain the edit.
Additional information
- Exact-account reproduction: verified.
- Same native Google Doc identity before and after editing: verified.
- Pre-Resync source-only canary lacked the new value: verified.
- Post-Resync source-only canary returned the new value: verified.
- Universal impact across other accounts, plans, source formats, or Project types: not proved.
- Internal cache lifetime and whether an undocumented background refresh would eventually occur: not proved.
- No private Project names, Drive IDs, local paths, legal records, credentials, logs, screenshots, or attachments are included.
- Related but not duplicate: [App/Linux] ChatGPT Project source names appear, but source files never sync locally and Work has incomplete context #39009 concerns Desktop/Linux materialization of Cloud Project sources, while this report concerns stale source content inside ChatGPT web until manual Resync.
Activity
kovr19687PrAgain commentedon Oct 17, 2026
Adding another affected user's report, with their permission, to describe the impact on software development.
In our coding workflow, Google Docs linked as Project Sources contain the current requirements, architecture rules, and refactoring constraints. The user has repeatedly encountered outdated content being used from the Project Sources representation while the Google Docs originals already contain newer instructions. This has required repeated corrections and additional verification during development.
The impact is more serious than an outdated summary: an agent can implement superseded requirements, miss current architecture/refactoring constraints, or evaluate changes against obsolete acceptance criteria. This creates substantial risk of incorrect code changes, regressions, and avoidable rework. We are not claiming a particular production defect was conclusively caused by this synchronization issue.
Our current mitigation is to explicitly retrieve the relevant original documents through the connected Google Drive tools before making code changes, rather than assume the Project Sources copy is current. That adds repeated work and undermines the expectation that attaching the canonical document to a Project makes the current requirements available to its chats.
This is an additional field report of the stale-source symptom and its development impact, not an independent controlled reproduction of the original report's manual-Resync experiment. We have not measured a reliable maximum synchronization delay or established the internal cause.
Please consider surfacing the source revision / last successful refresh time and an explicit stale or refresh-failed state to both the user and agent. An agent should be able to distinguish a cached Project Source from a freshly retrieved original, particularly when it is about to edit code based on that document.
No private document content, document links, project names, account details, or session logs are included.