Does it learn from my chats?
Training on your data: the model is fixed while you chat, but a saved copy can train a later model · Beginner · 5 min · training, safety · after Chatbot memory: fixed weights, a fresh context per chat, saved notes pasted into the prompt, Training: loss and gradient descent
No model changes while you chat. “Training on your data” means a saved copy of your chat may go into the data for a later model. Chat apps mostly do this unless you opt out; APIs mostly don’t.
1 Guess first
Your chat can’t change the model you’re talking to: its weights are fixed and shared by everyone. Training on your data is a separate, later step: the company keeps a copy of your chat, may have people review it, and can add it to the data for a future model.
3 Look
Who trains on your chats, by default
1 / 7Is your chat used for training, by default? vendor chat app API ChatGPT (OpenAI) trains by default not by default Claude (Anthropic) you choose not by default Gemini (Google) trains by default free tier trains Copilot (Microsoft) trains by default not by default DeepSeek trains by default not stated Mistral trains by default free tier trains Vendors’ own policy pages, read 2026-10-25; policies change
Chat apps mostly train on your chats unless you opt out; APIs and work plans mostly don’t. Step through each vendor for the switch and the page.
Chat apps mostly train on your chats unless you opt out; APIs and work plans mostly don’t.
4 You are the model
You prepare the data for the next training run. For each of five chats, decide: may it go in, or is it kept out?
chat 1 of 5
A ChatGPT Free chat about a work planSettings untouched
Illustrative chats; rules from the vendors’ pages, read 2026-10-25
You sort the chats for the next training run. Which may go in?5 Check
Three rules worth knowing. Opting out covers new chats, not data already collected. A thumbs up or down can share that chat anyway: OpenAI, Anthropic and Mistral say so. And work plans and paid APIs are kept out by default, while the free API tiers of Gemini and Mistral may train.
7 Say it back
8 Mission · 2 min
In ChatGPT Free, open Settings → Data controls and find “Improve the model for everyone”. Decide whether to leave it on. Then start a Temporary Chat: it isn’t used for training while it stays temporary.
Three quick ones
Drawn at random from this lesson's pool; different next time.
Done?
Get an email when new lessons come out
New lessons arrive in the weekly digest, only in a week that has one. You confirm by email first; unsubscribe in one click.
Go deeper
Sources
- OpenAI Help Center, “How your data is used to improve model performance” (https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance, checked 2026-10-25, read in Chrome because the page refuses automated reads): Free, Plus and Pro “we may use your content to train our models”; opt out with “Improve the model for everyone” under Settings > Data controls or “Do not train on my content” in the Privacy Portal; after opting out new conversations aren’t used; feedback can share the whole conversation; temporary chats are not used while they stay temporary; by default no training on ChatGPT Business, Enterprise, Edu or the API; API organization owners can enable data sharing
- Anthropic Privacy Center, “Is my data used for model training?” consumer (https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training, updated 2026-03-16, checked 2026-10-25): chats are used if you allow it under the Model Improvement privacy setting; incognito chats are not used; and commercial (https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training, updated 2026-08-18, checked 2026-10-25): “By default, we will not use your inputs or outputs from our commercial products” (Claude for Work, Anthropic API, Claude Gov)
- Anthropic, “Updates to Consumer Terms and Privacy Policy” (2025-08-28, https://www.anthropic.com/news/updates-to-our-consumer-terms, checked 2026-10-25): Free, Pro and Max; new users choose at sign-up; five-year retention if you allow training, 30 days otherwise; Commercial Terms services excluded. Neither page says which way the sign-up switch starts
- Google, Gemini Apps Privacy Hub (https://support.google.com/gemini/answer/13594961, hub updated 2026-10-23, privacy notice 2026-10-13, checked 2026-10-25): Keep Activity controls whether data is used to improve Google AI; with it off, chats are kept 72 hours and not used for AI training unless you send feedback; human reviewers read a subset, kept up to three years. “Manage & delete your Gemini Apps activity” (https://support.google.com/gemini/answer/13278892, checked 2026-10-25): “If you’re 18 or over, Keep Activity is on by default”; turn off at gemini.google.com, Settings & help > Activity
- Gemini API Additional Terms of Service (https://ai.google.dev/gemini-api/terms, last updated 2026-04-28, checked 2026-10-25): Unpaid Services content is used to “provide, improve, and develop” Google products, “human reviewers may read, annotate, and process your API input and output”; Paid Services: “Google doesn’t use your prompts” or responses to improve its products; in the EEA, Switzerland and the UK the Paid Services terms apply to the free tier too
- Microsoft, “Privacy FAQ for Microsoft Copilot” (https://support.microsoft.com/en-us/topic/privacy-faq-for-microsoft-copilot-27b3a435-8dc9-4b55-9a4b-58eeb9647a7f, checked 2026-10-25): Copilot conversation activity is used for AI training except for Entra ID accounts, Microsoft 365 Personal or Family, signed-out users, under-18s, users who opted out, and users in Brazil, China (excluding Hong Kong), Israel, Nigeria, South Korea and Vietnam; “Copilot for individuals: privacy controls” (https://support.microsoft.com/en-us/microsoft-copilot/microsoft-copilot-privacy-controls, checked 2026-10-25): Profile > Privacy, “Training on conversation activity” and “Training on voice conversations”. Both pages say they apply to the older Copilot app; the new app (available since 2026-08-18) has its own privacy controls page (https://support.microsoft.com/en-us/privacy/microsoft-copilot/privacy-controls, checked 2026-10-25) that lists no training setting, so the row is not verified for the new app
- Microsoft Learn, “Data, privacy, and security for Foundry Models sold by Azure” (https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy, 2026-05-18, checked 2026-10-25): prompts and completions “are not used to train, retrain, or improve the base models”
- DeepSeek Privacy Policy (https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html, last update 2026-02-10, checked 2026-10-25): text input, prompts, uploaded files and chat history are used “to train and improve our technology”; users have “the right to opt-out of using your Personal Data for training our models”, requests to privacy@deepseek.com; no setting is named. The in-app “Improve the model for everyone” switch is from Kaspersky’s guide (https://www.kaspersky.com/blog/deepseek-privacy-and-security/54643/), not confirmed on DeepSeek’s pages. DeepSeek Open Platform Terms (https://cdn.deepseek.com/policies/en-US/deepseek-open-platform-terms-of-service.html, effective 2026-04-29, checked 2026-10-25) don’t say whether API inputs are used for training: that row is unverified
- Mistral Help Center, “Can I opt out of my input or output data being used for training?” (https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training, 2026-09-03, checked 2026-10-25): Vibe users are not opted out by default, Enterprise is; Admin panel toggle “Allow your interactions to be used to train our models”, mobile “Enable data sharing”; Studio/API toggle under Anonymous improvement data, separate from Vibe’s. “Do you use my user data to train…?” (https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models, checked 2026-10-25): “Input and output data are used by default” on Vibe; Studio Free mode “we may use your data (input and output) to train”; Pay-as-you-go does not state a default. Le Chat was renamed Vibe in May 2026 (heise, https://heise.de/-11311685)
- The five chats in the you-are-the-model step are illustrative; each answer follows one of the policy rows above
Published Oct 25, 2026 · updated Oct 25, 2026