Skip to content

Tech explained for non-geeks

What does AI actually know about me?

Written by Daniel Gélinas, checked against primary sources. Last checked .

The short answer

Less than you fear, and more than you might think. What a chatbot learned in training was finished before you met, so your chat isn't in there. But the company can keep your chats, and some use them to train future versions: in ChatGPT that's on by default for personal accounts until you turn it off. Memory can also bring details up in later chats, and so can anything you connect, like your email. Check two settings: memory, and training.

The video

Watch it on YouTube

Dan wanted a week of dinners without peanuts. So he told a chatbot: “I'm allergic to peanuts. Meal plan?” It came back with a whole peanut-free week.

Then he stopped. “I just told a chatbot about my allergy,” he says. “Does it know me now?”

“Not the way you think,” says Liz.

What it learned, it learned before you met

A chatbot's “knowledge” comes from training: it learned patterns from a huge pile of text: public websites, data from partner companies, and material from people who help train it. That training was finished before Dan ever opened the app. Typing into a chat doesn't change the model he's talking to.

So Dan's peanut chat isn't in there. “Phew!”

One honest footnote: if you've written something public online, like a public post or a blog, it may have been in that pile. OpenAI says it filters out sites that collect personal information and takes steps to reduce personal details in training, but public is public.

But the company can keep your chat

Here's the part that matters more. Your chat doesn't vanish when you close the app. In ChatGPT, your chats stay in your history until you delete them.

And some companies can use those chats to train their next version. In ChatGPT, that's on by default for personal Free, Plus and Pro accounts. The setting is called Improve the model for everyone, under Settings → Data controls. Turning it off doesn't delete your chats; it just keeps them out of training.

A few catches worth knowing:

  • Thumbs up or down counts. If you rate an answer, OpenAI says the whole conversation can be used for training, even with the setting off.
  • Work accounts are different. ChatGPT's business, enterprise and education plans aren't used for training by default.
  • Old chats. To be safe, treat the switch as “from now on”. If there's a chat you don't want kept, delete it.

Memory: what it can bring up later

“So it remembers my allergy?” asks Dan.

“If memory is on, it can bring it up in later chats,” says Liz.

That's a different setting. With Memory on, ChatGPT can remember useful details, like an allergy or your kids' names, so you don't have to repeat yourself. You'll find it under Settings → Personalization → Memory, where you can see what it remembers and delete things.

One surprise: memories are stored separately from chats. Deleting the chat where you mentioned your allergy doesn't delete the memory it made. To remove it fully, delete both.

And what you let it open

Plus anything you connect. Depending on your plan, ChatGPT can use connected apps, like your Gmail, and files you've saved in it. Gemini can connect to your Google apps too. That's handy, and it's also access you gave it. If you don't want a chatbot reading your email, disconnect the app in its settings.

The other chatbots

Every big chatbot has the same two kinds of settings, under different names:

ChatbotTraining settingWhat to know
ChatGPTImprove the model for everyoneOn by default for personal accounts
GeminiKeep ActivityWith it off, chats are kept for 72 hours and aren't used for training. Some chats are reviewed, and those are kept up to three years
ClaudeModel Improvement (Privacy)With it on, chats are kept up to five years; with it off, 30 days
CopilotTraining on text, and on voice (Privacy)Turning them off covers future chats

These settings change often, and menus move. If you can't find one, search the app's help for “training” or “memory”.

Dan's choice

Dan opens his settings. “Training, off, from now on. Memory, I'll keep.” He likes not having to say “no peanuts” every time.

“And your meal plan?” asks Liz.

“Still peanut-free!”

His sum-up: what it learned is one thing. What he tells it, or lets it open, he can check.

What you can do

  • Check two settings: memory, and training. In ChatGPT: Settings → Personalization → Memory, and Settings → Data controls → Improve the model for everyone.
  • Look at what it remembers. Ask the chatbot what it remembers about you, or open its memory settings, and delete what you don't want kept.
  • Delete both the memory and the chat to remove something fully.
  • Disconnect apps you no longer want it to open, like your email.
  • Think before you share health, money or ID details. You can ask for a peanut-free meal plan without your full medical history.

Sources

  1. OpenAI Help Center: How ChatGPT and our foundation models are developed (help.openai.com)

    The models are developed from publicly available internet information, information from third-party partners, and information users, trainers and researchers provide. The models don't store copies of their training data; training adjusts the model's parameters. OpenAI says it takes steps to reduce personal information in training and filters out sites that aggregate it.

  2. OpenAI Help Center: Data controls in ChatGPT (help.openai.com)

    On personal ChatGPT Free, Plus and Pro workspaces, data sharing is on by default; you can turn off Improve the model for everyone (Settings, Data controls) and keep your chat history. Turning it off doesn't delete chats. Thumbs up or down can still share that whole conversation for training. Temporary chats aren't used for training but may be kept up to 30 days. Business, Enterprise and Edu workspaces aren't used for training by default.

  3. OpenAI Help Center: Memory in ChatGPT (help.openai.com)

    When Memory is on, ChatGPT can remember details from your chats and other sources, which can include connected apps such as Gmail and saved files. Manage it in Settings, Personalization, Memory. Saved memories are stored separately from chats: to remove something, delete the saved memory and the chat where you shared it.

  4. Google: Gemini Apps Privacy Hub (support.google.com)

    With Keep Activity on, Google uses your activity to improve its services, including training generative AI models. With it off, chats are kept with your account for 72 hours and future chats aren't used for training unless you send feedback. A subset of chats is reviewed, and reviewed chats are kept for up to three years.

  5. Anthropic: Updates to Consumer Terms and Privacy Policy (2025) (anthropic.com)

    Claude Free, Pro and Max users choose whether their chats can be used to improve Claude, and can change it in Privacy Settings. If they allow it, data is kept for five years; if not, the existing 30-day retention applies. Deleted conversations aren't used for future training. Work plans are excluded.

  6. Anthropic Privacy Center: Is my data used for model training? (privacy.claude.com)

    The setting is called Model Improvement, under Privacy Settings. Incognito chats aren't used to improve Claude, even with Model Improvement on.

  7. Microsoft Support: Microsoft Copilot privacy controls (support.microsoft.com)

    Copilot has separate settings for training on your conversations and on voice conversations, under Privacy. Opting out excludes your future conversations from training. Memory can be turned off, and you can delete all of it or ask Copilot to forget one thing. The page describes the older Copilot app.

Everything in AI

More on YouTube

Dan and Liz answer one everyday tech question at a time, most of them in under a minute.