A training opt-out is a setting or contractual term that excludes your conversations from being used to train or fine-tune an AI provider's models. With it in place, your prompts may still be processed to answer you — but they are not fed back into the next version of the model.
Opt-out vs. opt-in: the default matters
The key question is what happens if you do nothing. Many consumer AI products are effectively opt-in: your conversations are used for model improvement unless you find and disable the relevant control (often labelled something like "improve the model for everyone" in the data settings). Business and API tiers are typically the opposite — training is excluded by default under the provider's terms. So "we don't train on your data" means very different things depending on which door you came through.
The defaults differ enough per provider that the general rule is only a starting point. As of August 2026: ChatGPT personal workspaces train by default and the toggle affects new conversations, while business and API data is excluded by default. Claude asks you to choose at signup and applies it to new or resumed chats, with data you allowed retained up to five years. Gemini keeps activity on by default, and turning it off excludes future chats but not feedback you submitted. Grok, Microsoft Copilot and Perplexity all train consumer data by default with a forward-looking opt-out. Mistral trains on Free, Pro and Education by default; Team and Enterprise are excluded. Check the provider you actually use, on the tier you actually pay for — this is the single fastest-moving fact in AI privacy.
Opt-out is not deletion, and not zero retention
A training opt-out only addresses training. It usually does not remove your past conversations, and it does not mean the provider stops storing them — chats can be retained and human-reviewable even when they are excluded from training. Storage is a separate question (see zero-retention API), and so is deletion. Our 2026 data-retention audit tracks training, retention, and human access as three independent columns for exactly this reason.
How Secret Chat AI handles training
Secret Chat AI opts out of third-party training across the providers it routes to, so your prompts are not used to train or fine-tune the models — you do not have to hunt for a setting to make that true. Combined with local-only history and anonymized, email-only sessions, that is part of the site's core promise: it does not profile you, and nothing you ask trains a model. See the full picture on the home page.
Frequently Asked Questions
- Does opting out of training delete my past conversations?
No, on every provider we have checked. A training opt-out applies going forward and does not erase data already collected, nor withdraw it from a model that has already been trained — no deployed technique removes a specific person's data from finished weights. That is an engineering limit, not a legal one: your erasure rights are unaffected by the fact that the remedy is hard. Deleting past conversations is a separate action from opting out of training.
- If I opt out of training, is my data still stored?
It can be. Opting out of training does not automatically stop retention — a provider may still store and even human-review chats it has excluded from training. Retention and training are separate guarantees.
- Does Secret Chat AI train on my chats?
No. Secret Chat AI opts out of third-party training across the providers it uses, so your prompts are never used to train or fine-tune the models.
Related terms: Zero-Retention API · BAA · all glossary terms