Safety & privacy

AI chatbot privacy: what happens to your data, and how to protect it

Chatbots feel like private conversations, but your messages go to a company’s servers. Here’s what may happen to them and how to stay in control.

Key points

  • Your chats are processed and usually stored on the provider’s servers.
  • Depending on the provider and your settings, chats may be used to improve models and some may be reviewed by humans.
  • Check three settings in every chatbot: training/model improvement, history, and memory. Use temporary or incognito chats for sensitive topics.
  • Best protection: don’t share what you wouldn’t want stored — passwords, ID numbers, health records, other people’s data.

What happens when you send a message

When you type into a chatbot, your message travels to the provider’s servers, where the model processes it and sends back a reply. Unless you use a mode that says otherwise, the conversation is usually saved to your account history so you can return to it. From there, three things may happen, depending on the provider and your settings:

  1. Storage for your history, for a set period or until you delete it.
  2. Model improvement — your conversations may be used to train or improve future models.
  3. Human review — a sample of conversations may be read by trained reviewers, often to improve quality and safety.

Policies differ between companies, between free and business plans, and they change. That’s why the most important habit is to check the settings yourself, in the app, from time to time.

What the major providers say (checked November 19, 2026)

Google Gemini

Google’s Gemini Apps Privacy Hub says that human reviewers, including trained reviewers from service providers, review some data, and asks users not to enter confidential information they wouldn’t want a reviewer to see. Its “History” setting controls whether your data is used to train generative AI models. Google says temporary chats are not used to train its AI models, that reviewed chats can be retained for up to three years, and that chats are kept for around two months even when History is off.

Anthropic Claude

Anthropic’s privacy centre describes a “Model Improvement” setting in Privacy Settings, and says incognito chats are not used to improve Claude. Its privacy policy (effective September 10, 2026) says Anthropic may use inputs and outputs to train and improve models unless you opt out in your account settings, and that deleted conversations are removed from its back-end within 30 days. The help article and the policy word the default differently, so check the setting in your own account rather than assuming.

OpenAI ChatGPT

OpenAI’s privacy and data-control pages blocked our verification tools when we checked, so we aren’t stating its current defaults. In ChatGPT, open Settings and look for data controls to see whether your chats are used to improve models and what temporary chat options exist, and read OpenAI’s privacy policy directly.

Coding assistants

GitHub’s Copilot plans page says individual subscribers can opt out of having their data used for AI model training at any time, via their Copilot settings.

The privacy checklist

Settings to check today

  • Training / model improvement: decide whether you’re comfortable contributing your chats. Turn it off if not.
  • History / activity: shorter retention or auto-delete, if offered.
  • Memory: review what the chatbot remembers about you and delete anything you don’t want stored.
  • Connected apps: remove access to email, drives or calendars you no longer use with AI.
  • Shared links: if you’ve shared a chat link, remember anyone with the link may see it; delete links you no longer need.

Never paste

  • Passwords, PINs, recovery codes or API keys.
  • Full card numbers, bank details, passport or national ID numbers.
  • Other people’s personal information without their consent — clients, patients, students, colleagues.
  • Confidential work material, unless your employer has approved that tool and plan.

Reduce what you share

  • Anonymise: replace names with roles (“my manager”, “Client A”), remove addresses and account numbers.
  • Summarise sensitive situations instead of pasting raw documents.
  • Use temporary or incognito chats for one-off sensitive questions.
  • Consider a local model for highly private material — see open-source AI models.

Work and business accounts

Many providers offer business or enterprise plans with different data terms than personal accounts — for example, Anthropic’s Enterprise plan lists custom data retention and compliance features. If you use AI for work, ask your employer which tools are approved. Using a personal chatbot account for confidential client data can breach contracts and data-protection laws even if the chatbot itself is secure.

Privacy risks from agents and connected apps

When you connect a chatbot to your email, files or browser, it can read more — and an AI agent that reads untrusted content (a web page, an incoming email) can be manipulated by hidden instructions, a risk called prompt injection. OWASP ranks prompt injection as the top risk for LLM applications. Connect only what you need, and require approval before an agent sends or shares anything.

Children and teens

Check each service’s minimum age in its terms, use family or education accounts where available, and talk with young people about what not to share. A good rule for everyone: if you wouldn’t post it publicly or tell a stranger, don’t tell a chatbot.

Depending on where you live — for example under the EU and UK GDPR or California’s privacy law — you may have rights to access, correct or delete personal data that providers hold about you. Providers describe how to make requests in their privacy policies and privacy centres.

A 10-minute privacy check-up

  1. Open each chatbot you use and find its privacy or data settings.
  2. Decide on model training and history, and change them if needed.
  3. Open the memory section and delete anything you don’t want kept.
  4. Review connected apps and remove the ones you don’t use.
  5. Delete old shared links and conversations that contain personal details.
Editor’s note — add your own experience: Walk through the privacy settings in two chatbots you use, take screenshots of the relevant screens (blur personal info), and add a short “where to click” guide for each. Re-check every few months — menus change.
Remove this box by adding your content here, or hide all notes with "showEditorNotes": false in site.config.json.

The bottom line

AI chatbots can be used privately enough for most everyday tasks if you check your settings and keep sensitive details out. Treat every chat as something that may be stored and read, and you’ll rarely regret what you shared.

Frequently asked questions

Does ChatGPT use my chats for training?

OpenAI’s policy pages could not be loaded by our verification tools when we checked, so we won’t state its current defaults from memory. Open ChatGPT’s settings (look for data controls) and OpenAI’s privacy policy to see your current options.

Are temporary or incognito chats completely private?

They reduce what’s kept and used. Google says Gemini temporary chats are not used to train its models, and Anthropic says Claude incognito chats are not used to improve Claude. Providers may still retain them for a limited time for safety and legal reasons.

Can humans read my chatbot conversations?

It depends on the provider and your settings. Google’s Gemini privacy hub says human reviewers review some data and asks users not to enter confidential information they wouldn’t want a reviewer to see.

Is it safe to paste work documents into a chatbot?

Only if your employer allows it. Many organisations use business plans with different data terms. Never paste confidential client or company data into a personal account.

What’s the most private option?

Running an open-weight model locally on your own computer, so prompts never leave your device. It requires capable hardware and some setup.

Should I delete old chatbot conversations?

If they contain personal details you no longer need, yes. Deleting reduces what’s stored, although providers may keep data for a limited period for safety and legal reasons.

Sources

  1. Anthropic Privacy Center — Is my data used for model training? (opens in a new tab)
  2. Anthropic — Privacy policy (opens in a new tab)
  3. Google — Gemini Apps Privacy Hub (opens in a new tab)
  4. Claude plans (Enterprise data retention features) (opens in a new tab)
  5. GitHub Copilot plans (training opt-out note) (opens in a new tab)
  6. OWASP — LLM01:2025 Prompt Injection (opens in a new tab)
  7. NIST — AI Risk Management Framework (opens in a new tab)

Facts, prices and features were checked against these sources on November 19, 2026. AI products change quickly; if you spot something out of date, please tell us.