A colleague says: "I turned off model training in Data Controls, so my conversations are private and nothing is kept." Judged against what the setting does, what is wrong with that?

Hardchatgptdata-controlsprivacy

Last reviewed 11 Aug 2026

The setting stops training on conversations, but they still sit in history and inform memoryCorrect answer
The setting only takes effect for conversations started after it, so earlier ones remain in training
The setting cannot be changed back, so the real problem is that the decision is permanent
The setting is per device, so it protects the laptop it was changed on and nothing else

Show answer & explanation

Correct answer: A. The setting stops training on conversations, but they still sit in history and inform memory

Explanation

The principle — privacy in ChatGPT is not one switch. Deciding what your conversations are used for, whether they are stored, and whether they shape later answers are three separate decisions with three separate controls, and a claim that collapses them is wrong even when the switch itself was set correctly.

Why the key is correct — OpenAI is explicit about what this setting leaves untouched: your conversations will still appear in your chat history but will not be used to train ChatGPT. So "nothing is kept" does not follow. Memory is a further gap in the claim, because memory forms from your chats and is governed separately under Settings and then Personalization. The control that actually delivers what the colleague described is Temporary Chat, which does not show in history, does not use or create memories, is not used to train models, and is deleted after 30 days.

Why the others are wrong — the setting is not per device; OpenAI states it applies to your entire account regardless of which device you are using. Timing is not the flaw either, since opting out means new conversations are not used for training, and presenting that as the problem draws attention away from the unsupported half of the claim. And it is not irreversible; you can change the setting at any time with no restrictions.

Remember this — when someone states a privacy conclusion, check which of the three questions their control actually answers. Training, storage and personalisation are decided in different places, and the only tool that answers all three at once for a single conversation is Temporary Chat. One further caveat worth carrying: OpenAI notes that turning off memory and personalisation features does not disable safety features that may use limited, safety-relevant context in rare, high-risk situations.

Sources — OpenAI's Data Controls FAQ and What is ChatGPT FAQ.

Sources

Primary sourceData Controls FAQ — OpenAI · accessed 23 Jul 2026
“Your conversations will still appear in your chat history but won't be used to train ChatGPT.”
Primary sourceWhat is ChatGPT: FAQ — OpenAI · accessed 27 Mar 2026
“Temporary chats don't show in history, don't use/create memories, and aren't used to train models.”

Practise 10 questions on this topic

Take ChatGPT Basics — Timed Test 1 (10 questions) — scored instantly, explanation for every question, no login.

Start the test

A colleague says: "I turned off model training in Data Controls, so my conversations are private and nothing is kept." Judged against what the setting does, what is wrong with that?

Related questions

A colleague says: "I turned off model training in Data Controls, so my conversations are private and nothing is kept." Judged against what the setting does, what is wrong with that?

Related questions