Record when you observed the training setting off and when you started the next new conversation. Treat that as your operational cutover record, not proof that older conversations were changed retroactively.
Why the new-conversation wording matters
OpenAI’s supplied evidence says sharing is enabled by default on personal Free, Plus, and Pro workspaces. Turning off “Improve the model for everyone” means new conversations are not used to train models and can remain in chat history.
The important boundary is “new conversations.” The supplied passage does not establish retroactive treatment of earlier chats. A careful record therefore describes what you observed and what you did next, rather than claiming a universal policy change across your entire history.
Build a minimal cutover log
Use this editorial worksheet without including prompt contents or confidential material:
- Workspace context: A non-sensitive label identifying which personal workspace you checked.
- Observed control: The exact setting name and whether you observed it off.
- Observation time: Date, time, and time zone, such as IST.
- Decision owner: Your initials or an approved internal reference.
- First subsequent session: A neutral label for the new conversation and its start time.
- Evidence boundary: “Record of observed preference; no retroactive treatment claimed.”
Record only what you can actually observe. Do not invent a backend activation timestamp or treat a handwritten log as provider confirmation. Avoid storing account credentials, private prompts, or screenshots containing sensitive conversations.
Hypothetical cutover evaluation
Suppose an independent researcher in Jaipur records “setting observed off, 29 October 2026, 14:10 IST.” At 14:15, they start a new conversation using an invented research brief. At 14:20, they return to a conversation begun the previous day.
The first new session fits the source’s new-conversation wording. The resumed older session should be marked “outside the conclusion supported by this log.” Its later message time alone does not establish how the earlier conversation is treated.
Evaluate the record by asking whether another reviewer could distinguish observation time, new-session time, and unknowns without reading any conversation content.
Stop if the boundary is uncertain
If you cannot tell whether a session predates the decision, do not classify it confidently. For future work, start a clearly new conversation after checking the preference, using only authorised, non-sensitive material. Keep older-session questions open pending current provider guidance.
If changing access arrangements, check access format and saved terms through the AI tools catalog before buying. Marketplace account, code, or token offers are not proof of official subscription equivalence, and offer wording does not replace provider data-control documentation.
Sources and scope
Source material checked 29 October 2026; features/policies may change. OpenAI data controls establishes the stated default and new-conversation distinction. The log is editorial process advice, not a provider audit mechanism or a claim about retroactive treatment.