AI and your customer data: privacy basics

Every time someone on your team pastes a customer email, a spreadsheet, or a contract into an AI tool, that information leaves your systems and goes to another company. Whether that's fine depends on which tool, which plan, which settings, and what the data is. Most of the risk comes from not having thought about it at all.

This guide explains what to check and gives you rules your team can follow. For general data handling outside AI, see data privacy basics for your business.

What actually happens to data you put in

When you type or upload something into an AI tool, the provider receives and processes it. What happens next depends on the provider's terms and your settings:

Training. Some consumer AI services may use conversations to improve their models unless you opt out. Business, team, and API plans from major providers generally say they don't train on your data by default. Check the specific terms for the plan you're on, because they differ and they change.
Retention. Providers keep conversations for some period, for your history and often for abuse monitoring. Some business plans let admins set retention or turn history off.
Human review. Some providers allow staff to review conversations in limited circumstances, such as investigating abuse.
Subprocessors. The AI provider may use other companies (cloud hosts, for example) to run the service. AI features inside other software may send your data to a separate AI provider.
Location. Data may be processed in a different country from yours.

None of this is automatically a problem. It's information you need in order to decide what can go where.

Personal versus business accounts

The single most useful rule: company data goes into company-controlled AI accounts.

When staff use personal AI accounts for work:

You can't see or control what's been shared
The data may be used for training under consumer terms
When the person leaves, the conversation history leaves with them

Business plans typically add admin controls, clearer data terms, and the option of a data processing agreement. Compare the current plans of the tools you use, and include AI tools in your vendor review.

Three questions for any AI tool

Before approving a tool for customer data, find the answers (usually in the provider's privacy policy, trust center, or business terms):

Does it use our inputs and outputs to train its models? Is that off by default on our plan, or do we have to opt out?
How long does it keep our data, and can we control or delete it?
Will it sign a data processing agreement (DPA), and does it list its subprocessors?

If a tool can't answer these clearly, don't use it for customer data.

Help me choose

Can this AI tool take customer data?

On your plan, is training on your inputs and outputs off, by default or because you opted out?

Even with a good tool, the less you share, the less can go wrong.

Remove identifiers where they're not needed. "Summarize this complaint" works just as well with the customer's name, email, and order number removed.
Use placeholders like [CUSTOMER] or [ADDRESS] in drafts.
Summarize instead of uploading whole databases. Ask about patterns using anonymized or aggregated data.
Don't upload entire mailboxes or CRM exports to a general-purpose AI tool.

Never paste these into a general AI tool

Payment card numbers, bank details, government ID or Social Security numbers, health or medical information, passwords or API keys, and anything you're legally or contractually bound to keep confidential. Some specialized tools are built and contracted for regulated data (for example, under a HIPAA business associate agreement). A general AI chat tool usually isn't.

The legal side, briefly

Privacy laws apply to personal data whether or not AI is involved. A few points to be aware of:

Transparency. Your privacy notice should describe how you use personal data, including the kinds of service providers you share it with. If AI tools process customer data, make sure your notice covers that. See GDPR basics and CCPA & CPRA basics.
Lawful basis and purpose. Under GDPR, you need a lawful basis for processing personal data, and you shouldn't use it for purposes customers wouldn't reasonably expect.
Contracts with processors. GDPR requires a written contract (a DPA) with service providers that process personal data on your behalf.
Automated decisions. GDPR gives people rights around decisions made solely by automated means that significantly affect them, like refusing credit. If AI is making decisions about people, take advice.
Health data. If you're covered by HIPAA, don't put protected health information into any tool without a business associate agreement. See HIPAA basics for websites handling health info.
Client contracts. Some clients forbid you from putting their data into AI tools, or require approval. Check your contracts.

This is general information, not legal advice. If AI is central to how you handle personal data, talk to a privacy lawyer.

A simple team rule

Add a short section to your security policy. For example:

Use only the AI tools on the approved list, signed in with your work account.
Customer and employee personal data may only go into tools marked "approved for personal data."
Remove names and identifiers when you can.
Never enter payment, health, ID, or password information.
Review all AI output before it's sent to a customer.
If you're unsure, ask before you paste.

Keep the approved list short and visible, and tell staff why. People take shortcuts when the rules feel arbitrary or the approved tool is hard to use.

AI features inside your existing software

AI features are appearing inside email, CRM, help desk, and accounting tools. They're often switched on by updates. For each one:

Check whether it's enabled for your account, and whether an admin can turn it off
Read the vendor's notes on which AI provider it uses and whether your data is used for training
Decide whether it's appropriate for the data in that system

Frequently asked questions

Is it safe to use AI for customer emails?

It can be, with a business-plan tool that doesn't train on your data, sensible retention settings, and identifiers removed where practical. Always review replies before sending. See AI for customer support.

What about AI meeting note-takers?

Tell everyone on the call that it's being recorded and transcribed, and get consent where the law requires it. Recording consent rules vary by location. Check where the recordings and transcripts are stored, and who can access them.

A staff member already pasted customer data into a personal AI account. What now?

Don't panic. Find out what was shared and which tool. Ask them to delete the conversation and check the tool's data settings. If sensitive data was involved, treat it as a possible incident and consider whether you need legal advice. Then fix the rule and the approved tool list so it doesn't happen again.