Do AI companies read my chats?

Many providers state that conversations may be reviewed by authorized staff or contractors for safety, abuse prevention or quality, sometimes after removing identifying details. Policies differ by provider and plan, so read the privacy policy of each tool you use and assume anything you type could be seen by a person.

Does deleting a chat delete it everywhere?

Not necessarily. Deleting a conversation usually removes it from your visible history, but providers may keep copies for a period for safety or legal reasons, and data already used for training generally cannot be pulled back out of a model. Check each provider's stated retention period.

Are paid or business versions of AI tools more private?

Often, yes. Business and enterprise plans commonly exclude customer data from model training by default and offer contractual protections and admin controls. Consumer plans, including some paid ones, may use conversations for training unless you opt out. Confirm the terms for your specific plan.

AI Chatbot Privacy: Settings and Habits That Protect You

AI assistants feel like private conversations. You type, it answers, nobody else is in the room. In practice, your words travel to a company's servers, may be stored for months or longer, may be reviewed by people, and may be used to train future models unless you change a setting. None of that means you should stop using AI. It means you should use it the way you would use email: helpful, but not a diary.

What chatbots typically collect

Policies vary by company and plan, but most consumer AI assistants collect some combination of:

Your prompts and the responses, including any files, images or voice recordings you upload.
Account details such as name, email, phone number and payment information.
Usage data: device, browser, approximate location from your IP address, and how you use features.
Feedback you give, such as thumbs up or down, which may come with the full conversation.
Memory features that store facts about you across conversations to personalize responses.
Connected data if you link email, calendars, cloud storage or other apps.

The settings that matter most

Model training

Many consumer assistants may use conversations to improve their models by default, with a setting to opt out. Business, enterprise and education plans more commonly exclude data from training by default. Look in settings under labels like "data controls," "privacy" or "improve the model."

Chat history and retention

Some tools let you turn off history or use a temporary chat mode that does not save the conversation to your account. Providers may still retain these chats for a limited period for safety monitoring, so "temporary" does not always mean "never stored."

Memory

Persistent memory features remember details you share, such as your job, family members or preferences. Review what has been stored, delete items you do not want kept, or turn memory off.

Connected apps

Granting an assistant access to email, files or a calendar is powerful, and it expands what data the provider can process. Grant only what you need and review connections periodically.

Sharing links

Shared conversation links can be visible to anyone who has the link. There have been cases of shared chats being indexed by search engines. Treat any shared link as public.

What not to paste into a consumer chatbot

Type of information	Why it is risky	Safer approach
Social Security, passport or driver's license numbers	Identity theft if exposed	Never needed for AI help
Bank, card or account credentials	Fraud	Never share; describe the issue generally
Passwords and security codes	Account takeover	Never share
Detailed health records	Sensitive and usually not HIPAA-protected in consumer tools	Summarize without identifiers
Employer confidential data	May breach policy or contracts	Use employer-approved tools only
Other people's personal details	Their privacy, not yours to give	Anonymize names and specifics
Legal matters in progress	Could undermine confidentiality	Talk to your lawyer first

A useful rule: if you would not put it in an email to a stranger, do not put it in a chatbot.

How US privacy law applies

There is no single comprehensive federal privacy law in the United States. Protection comes from a patchwork:

The Federal Trade Commission can act against companies that break their own privacy promises or use unfair or deceptive data practices. Its privacy and security guidance explains how it approaches these issues.
State privacy laws, led by California's CCPA as amended, give residents rights such as knowing what data a business collects, deleting it, correcting it and opting out of its sale or sharing. A growing number of states have passed similar laws, with differences in scope.
Sector laws cover specific data: HIPAA for health information held by healthcare providers, insurers and their business associates; FERPA for student education records; COPPA for children under 13 online.

One point surprises many people: HIPAA generally does not apply when you type health information into a consumer chatbot yourself, because the AI company is usually not your healthcare provider or its business associate. The protection you have comes from the company's privacy policy and general consumer protection and state law.

Using your privacy rights

If you live in a state with a comprehensive privacy law, you can often request a copy of your data or ask for its deletion through the provider's privacy portal or support pages. Even where it is not required by law, many providers offer data export and account deletion tools to all users. Look for a "privacy request" or "data export" option in account settings or the privacy policy.

A five-minute privacy checkup

Open each AI tool you use and work through this list:

Training: Is my data used to train models? Turn it off if you prefer.
History: Do I need chat history? Delete old conversations you no longer want.
Memory: What does the assistant remember about me? Remove anything sensitive.
Connections: Which apps and accounts are linked? Remove those you do not use.
Shared links: Have I shared any chats? Revoke links you no longer need.
Account security: Is two-factor authentication on? A compromised AI account exposes your entire chat history.

At work

If your employer provides an approved AI tool, use it for work content rather than a personal account. Company plans typically come with contractual data protections that consumer accounts lack. If no tool is approved, ask before using one with work information; many data leaks start with a well-meaning employee pasting something into the wrong place.

Anonymize before you ask

You can usually get the same quality of help without the identifying details. Before pasting a document or describing a situation:

Replace names with roles or placeholders: "my manager," "Client A," "Person 1."
Round or remove numbers that could identify an account, a salary or a medical result, unless the exact figure matters to the question.
Strip metadata and headers from documents and emails, including signatures, addresses and reference numbers.
Describe rather than upload when only a summary is needed. "A lease clause that lets the landlord enter with 24 hours' notice" is enough to get an explanation.
Crop screenshots so they show only what is relevant.

These habits take seconds and dramatically reduce what is stored about you and the people around you.

Browser extensions and third-party apps

Be especially careful with AI browser extensions, "wrapper" apps that put a new interface on someone else's model, and mobile apps from unfamiliar developers. They may collect more data than the underlying AI provider, including your browsing activity. Stick to official apps from known providers or tools vetted by your employer.

Privacy with AI is mostly about habits: share less, check settings once, and assume anything you type could someday be read by someone else. With those habits in place, you can get the benefits of AI without handing over more than you meant to.

Protecting Your Privacy With AI Chatbots: Settings, Habits and Limits

What chatbots typically collect