OpenAI reported elevated conversation errors for ChatGPT Free and Go users on November 18, 2026 and later marked the incident resolved. The status page identifies ChatGPT conversations as the affected component. It does not establish an OpenAI API outage, a Codex outage or a failure in a particular client's website. The useful agency question is narrower: how should a client-facing workflow preserve work when a hosted conversation becomes unavailable or unreliable?
Separate the conversation from the work record
Do not make an AI conversation the only copy of a client request, acceptance decision or recovery checklist. Keep the durable record in an authorized project system: request ID, client, owner, current decision, source links, next action and last confirmed state. Avoid copying secrets or personal data into a fallback channel.
Use a bounded fallback
When a conversation errors, retry once after checking the provider status. If the task is still unavailable, switch to an approved fallback such as a human operator, an authorized API workflow or a local checklist. The fallback should receive the durable task record and the minimum context needed to continue, not an unreviewed transcript dump.
For an agency, the handoff card can contain:
- client request and business process;
- last confirmed step and evidence;
- unresolved decision;
- owner and deadline;
- fallback channel used;
- reconciliation step for any work completed during the incident.
After recovery, compare the fallback record with the restored conversation. Mark duplicates, missing outputs and decisions that still need human approval. A successful retry is not proof that an earlier response was delivered or saved.
Keep the client message precise
If the incident affects a client-facing deadline, communicate that the provider reported a conversation availability issue and state the next checkpoint. Do not promise provider restoration, reconstruct unrecorded instructions or claim that a customer account was affected unless the account owner confirms it.
Applicability and limits
This scenario applies only to workflows using the affected ChatGPT Free or Go conversation surface during the reported window. It does not generalize to paid plans, API services or Codex. The practical output is a durable task record, bounded fallback and post-recovery reconciliation.
Primary source: OpenAI Status - Elevated conversation errors for Free and Go users.