My first query got a possible bot response with the promise of escalation to a human representative. I suspect I was put in front of another AI agent, likely running a higher model. OpenAI left that vague. After five days of silence, I decided to open a new case.

In the first case, asking for escalation produced an immediate reply, too fast for a real human. The second case also kept saying “escalated” while an AI assistant gathered information, before eventually asking whether I wanted specialist review. Finally, this time, a human answered.

The supposed human representative opened with the standard greeting and a real name, unlike the AI replies. They said they had reviewed what I had already supplied about the Pro-to-Plus downgrade that left my weekly Codex allowance at 0%, and asked for any additional historical evidence I still had: billing timestamps, screenshots, logs, or task IDs. They also said I did not need to recreate evidence that no longer existed. Much of the evidence I had, it was already submitted a week ago. And the evidence I could have collected, it has likely washed away by now.

The representative said the current guidance did not establish whether the Pro weekly allowance should have been recalculated when Plus became active. They said the evidence I had supplied was sufficient for the review to continue, and that any reset, credit, or other adjustment would depend on its finding. Then the case closed. I only found out when the chat window said I would need to open a new case to continue. Again, as unclear as the transition was, neither the supposed review decision nor a remedy ever came.

Codex had become usable again when the regular weekly window reset landed. That did not account for the better part of a week I had lost on the newly paid Plus plan. The amount was less than $5 of a $20 subscription. I was not asking for damages or a full refund. OpenAI could have brushed it all aside and sent me off with a reset, or completed a proper review. Instead, even the review was cut off. A human reply simply was not sufficient, and the case closed without settling the core question of the whole debacle: billing.

The real issue at hand is not the reset. It’s the subscription model we are supposed to be buying. Technicalities aside, a new subscription starts with a new tier. And OpenAI has no annual plan for personal users, so it renews every month. If a banked reset ends up counting against the next paid period, this is not a complimentary service on the house; it is a hidden debt written against a user’s wishes.