# AI Chatbot Privacy in 2026: Settings to Change and What Never to Paste

> How ChatGPT, Claude, Gemini, and Perplexity use your conversations, where to find each training opt-out, what opting out doesn't cover, and safe habits for AI assistants connected to email and files.

- Canonical URL: https://agora.supplies/blog/ai-chatbot-privacy-settings/
- Category: Online Privacy
- Published: October 12, 2026
- Updated: October 12, 2026
- Publisher: Agora Supplies (https://agora.supplies/)

## Key points

- Most consumer AI chatbots can use your conversations to improve their models unless you turn a setting off.
- Opting out generally applies to future chats only. Chats may still be retained for a period for safety, abuse monitoring, or legal reasons.
- Never paste passwords, recovery phrases, full ID numbers, or other people's private information into a chatbot, whatever the settings.
- AI assistants connected to email, calendars, or files hold far more data than a chat window. Grant only the access you need and review it regularly.

People tell AI chatbots things they wouldn't type into a search engine: medical worries, contract disputes, salary negotiations, relationship problems. The conversation feels private. Unless you change some settings, it may not be.

This guide covers what the main consumer chatbots do with your conversations, where the controls are as of **October 2026**, and the habits that protect you whatever the settings. Menus change often, so if a label doesn't match, search the app's settings for "training", "improve", or "activity".

## What happens to your chats by default

On most **consumer** plans, conversations are stored in your account and may be used to train or improve future models, sometimes after review by people. **Business and enterprise** plans generally exclude customer data from training by default and give administrators stricter controls.

Two points are easy to miss:

- **Opting out usually affects future chats only.** Data that has already been used in training can't be removed from a model.
- **Opting out of training is not the same as deletion.** Providers typically keep conversations for some time for abuse monitoring, safety review, or legal obligations, even when training is off.

## Where to find the controls

### ChatGPT

Go to **Settings > Data Controls** and turn off **"Improve the model for everyone"**. The setting is on by default for Free, Plus, and Pro plans. Even with it off, authorised staff and service providers may access chats to investigate abuse, provide support, or meet legal requirements ([Malwarebytes, September 2026](https://www.malwarebytes.com/blog/how-to/2026/09/how-to-opt-out-of-ai-chatbot-training)). **Temporary Chat** keeps a conversation out of your history.

### Claude

Go to **Settings > Privacy** and turn off the model improvement setting. Anthropic says chats may be used for training if you allow it, if you join a programme such as its Trusted Tester Program, or if a conversation is flagged for safety review. **Incognito chats** are not used for training even when the setting is on. If you use the thumbs-up or thumbs-down feedback buttons, the full conversation may be stored for up to five years ([Anthropic privacy center](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training)).

### Gemini

Turn off **Gemini Apps Activity**, also labelled **Keep Activity**, from **Settings & help > Activity** at gemini.google.com, or through Google's My Activity page. Google says that even with it off, conversations are saved with your account for up to 72 hours to provide the service ([Google Help](https://support.google.com/gemini/answer/13278892)). Conversations already reviewed by human reviewers are kept separately and are not removed when you delete your activity. On work or school accounts, your administrator controls these settings.

### Perplexity

Go to **Settings > Preferences** and turn off **AI data retention**. Data collected before you opt out is not removed, and data may still be processed to run the service or comply with the law.

## What never to paste into a chatbot

Settings reduce risk. They don't remove it. A conversation can be stored, reviewed, exposed in a breach, or shown to someone who borrows your laptop. Keep these out of chats entirely:

- **Passwords, API keys, and recovery or seed phrases.** If one ever ends up in a chat, treat it as compromised and replace it.
- **Full ID numbers**: Social Security, passport, tax, or national ID numbers.
- **Full card numbers and bank login details.**
- **Other people's private information**, such as a colleague's medical situation or a client's contract, unless you are authorised to share it.
- **Confidential work documents** on a personal account. Many employers have rules about this, and a business plan with training disabled is the right place for work data.

When you need help with a document, **redact it first**. Replace names with placeholders and remove account numbers. The answer is usually just as useful.

## AI assistants with access to your accounts

The bigger privacy shift in 2026 is AI **agents**: assistants that connect to your email, calendar, cloud drive, messaging apps, or browser and act for you. Some early products ask for access to almost everything on a device.

That access is useful, and it is a lot of trust to hand over. An assistant with access to your inbox can read password resets, bank alerts, medical letters, and every conversation in it. Before connecting:

1. **Grant the narrowest permission available.** Read-only access is safer than read-and-send, and one calendar is safer than all of them.
2. **Don't connect your most sensitive accounts.** Keep the email you use for banking and account recovery separate from the one you let an agent read.
3. **Review connected apps every month.** Google, Microsoft, and Apple accounts each have a page listing third-party access. Revoke anything you no longer use.
4. **Watch for instructions hidden in content.** Agents that read web pages or emails can be manipulated by text planted there, a technique known as prompt injection. Be wary of letting an agent send messages, make purchases, or move money without your confirmation.

## A five-minute privacy routine

- [ ] Turn off training in each chatbot you use
- [ ] Use temporary or incognito chats for anything sensitive
- [ ] Delete old conversations you don't need
- [ ] Review which apps and agents can access your email and files
- [ ] Use a separate, business-grade account for work material

AI tools are useful. Handle them like any other cloud service that stores what you type: share what the task needs and nothing more. That rule runs through all our guides, from [what to do after a data breach](/blog/data-breach-response-plan/) to [how to pay online anonymously](/blog/how-to-pay-online-anonymously/).

## Frequently asked questions

### Does ChatGPT train on my conversations?

On consumer plans, conversations can be used to improve models unless you turn off "Improve the model for everyone" under Settings > Data Controls. Business and enterprise plans are excluded from training by default.

### How do I stop Claude from using my chats for training?

Open Settings > Privacy and turn off the model improvement setting. Anthropic says incognito chats are not used for training, while conversations flagged for safety review can still be used for that purpose.

### Does turning off Gemini Apps Activity delete my chats?

It stops future chats being saved to your activity. Google says conversations are still kept with your account for up to 72 hours to provide the service. Chats reviewed by human reviewers are retained separately.

### Is it safe to give an AI assistant access to my email?

It can be useful, but it gives the assistant, and the company running it, access to everything in that inbox. Use the narrowest permissions available, avoid connecting accounts that hold highly sensitive material, and revoke access you no longer use.
