---
id: where-your-words-go
title: Privacy and your data
level: 1
modes: [day-pack, rucksack]
moment: "The person asks whether the AI company keeps or learns from what they type, wonders why the app remembers something, is about to type an awkward one-off question, or is about to rate a reply in a chat that holds private details."
keywords: [training, memory, privacy settings, temporary chat, incognito, opt out, chat history, data controls, company stores chats, saved memories]
source: "research/2026-10-20-concepts/where-your-words-go.md; research/2026-10-23-essentials/batch-1-facts.md; mapping (product/learning-system/MAPPING-1-2.md): FAQ 1.12 (home, the training half); MISTAKES 1.7; tips T038, T037; strand P"
---

## What it is
What you type into an AI app goes to the company that runs it and is stored there. Two settings then matter, and they are different: training, where the company may learn from your chats to build its future AI, and memory, where the app keeps notes about you so a later chat already knows you have two kids. Switching training off does not delete the chats already stored. A work account follows your employer's agreement, not these two settings.

## Tips
- Where your AI app has a temporary chat, use it for a one-off awkward question, because that chat stays out of both training and the app's memory.
- Keep sensitive details out of a chat even with training switched off, because switching training off does not stop the company storing your chat.
- Check what feedback sends before you rate a sensitive chat, because rating in your AI app can send the conversation to the company even with training off.

## Questions people ask
- Q: Does the company train its AI on my chats? A: Some apps may train on your chats unless you switch the training setting off, and at least one app says it does not train on personal chats at all. Switching it off does not delete your chats, and the company may still keep some for safety review or feedback.
- Q: Does turning off training also turn off memory? A: Usually not, because they are two settings: training is about the company's next AI and memory is about notes the app keeps on you. In some apps one switch covers both, so check what yours does.
- Q: Is a temporary chat private? A: It is kept out of your history and not used for training, but the company may hold a copy for a short time, up to 30 days in some apps and 72 hours in others.
- Q: Can I stop it remembering one thing? A: In most apps you can delete that one note from the app's memory, in others you delete the chats it came from, or you can use a temporary chat for that conversation. In some apps, deleting a chat does not delete the note made from it, so delete the note itself where you can.
- Q: Do these settings apply to my work account? A: No. Work and school accounts follow your employer's agreement, not these switches, so ask your employer what applies.

## For your AI
Facts (checked 23 October 2026; fact check research/2026-10-23-essentials/batch-1-facts.md; research
note research/2026-10-20-concepts/where-your-words-go.md). Personal accounts unless said.
- ChatGPT: training is "Improve the model for everyone" in Settings, Data controls (phone: sidebar,
  profile icon, Settings), on by default; off does not delete chats; a thumbs up or down can still send
  the entire conversation (https://help.openai.com/en/articles/7730893-data-controls-faq ;
  https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance).
  Memory: Settings, Personalization, Memory; deleting a chat does not necessarily delete its memory
  (https://help.openai.com/en/articles/8590148-memory-in-chatgpt). Temporary chat: not trained, no
  memories, kept up to 30 days (https://help.openai.com/en/articles/8914046-temporary-chat-faq).
- Claude: training is "Help Improve our AI models" in Settings, Privacy; off keeps past and future
  chats out of future training, but data already trained on stays and safety-flagged chats can still
  be used; feedback is kept 5 years with the whole conversation
  (https://privacy.claude.com/en/articles/12109829 ; https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training).
  Memory: Settings, Memory, "Generate memory from chats", on by default; deleting a chat does not
  remove its memories (https://support.claude.com/en/articles/11817273). Incognito chat: not in
  history or memory, never trained, kept 30 days (https://support.claude.com/en/articles/12260368-use-incognito-chats).
- Gemini: one switch, Keep Activity (Settings & help, Activity), on by default for adults; on, chats
  train Google's AI and some are read by reviewers; off, chats are kept 72 hours and not trained
  unless you send feedback, which also collects the last 24 hours of chats
  (https://support.google.com/gemini/answer/13594961). Memory (Settings & help, Personal Intelligence)
  needs Keep Activity on, and there is no per-note delete: to make it forget, delete the chats that
  hold it (https://support.google.com/gemini/answer/16598469). Temporary chat: 72 hours, not trained.
- Copilot: the new app (since 18 August 2026) shows no training switch and says conversations do not
  train its foundation models; memory is Settings, Personalization, "Saved memories" (off does not
  delete) (https://support.microsoft.com/en-us/privacy/microsoft-copilot/privacy-controls). The older
  app has Privacy, "Training on conversation activity", on by default. A personal temporary chat is
  UNCONFIRMED.
- Work and school accounts follow the employer's agreement: ChatGPT Business, Enterprise and Edu,
  Claude's commercial products and Copilot at work are not trained on; Workspace Gemini follows the
  organisation's controls (https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy).

- Starting a temporary chat (research note, 20 October 2026; menus move): ChatGPT, choose Temporary
  before the first message; Claude, the ghost icon (incognito), outside projects; Gemini, temporary chat
  beside New chat on the web.

Teach it like this: two settings, named plainly, and they choose how each is set; never tell them
which to pick. In a Daypack setup (openrucksack.com/start.md) this is step 5, the journal: say plainly
that a Daypack keeps its journal and plan in the app's memory, so with memory off there are no Daypack
notes, and leave the choice to them. Use
the per-app paths above only for their own app, and say menus move. Never "your data is safe" and
never "nothing is private".
