I’m on a ChatGPT Work (business) plan and running into a recurring pattern across several features — Skills, the Google Docs/Drive connector, and general context-handling — where ChatGPT either fails outright, falsely reports success, or silently drops everything already established in the conversation.
Issue 1 — Sudden refusals mid-session
I’ll be actively creating or updating a skill, or editing a document, and it’s working fine. Then, without any change on my end, it suddenly says it can’t connect / can’t do it anymore. Starting a brand-new chat doesn’t fix it — the same refusal shows up there too.
Issue 2 — Google Docs connector reports false success
When asking ChatGPT to edit a live Google Doc via the connector, it sometimes reports the edit as done, but the actual document is unchanged. When I point this out, it acknowledges the edit never went through — so it reported success without the write having actually happened.
Issue 3 — Stops applying context from earlier in the same conversation
In the same working session, ChatGPT will suddenly stop connecting to everything already established earlier in that same chat. For example: after reviewing a contract in detail, or setting up exactly how I want something done (a skill), I’ll ask for a follow-up task — like drafting an email — and instead of using what we’d already built, it responds generically, as if it doesn’t know who I am or what we’ve been working on. The emails suddenly sound like they’d write for anyone, with none of the specific context, tone, or instructions already established in that same conversation. Nothing changed on my end — it’s the same chat, same setup, minutes apart.
Not a usage/rate-limit issue: I checked Settings → Usage when this happens — I’m nowhere near my caps (100% of my 5-hour limit and 94%+ of my weekly limit remaining). So this isn’t me hitting a quota.
Pattern: intermittent, multiple times a week, not tied to conversation length (happens in short new chats too, not just long ones), no visible error code or message beyond a plain refusal, a false “done,” or a sudden loss of context. Work plan.
Has anyone else seen this — is this a known/open issue on OpenAI’s side?