Blog · AI and automation

Before You Give an AI Tool Your Customer Data

Before you paste customer information into an AI tool, or connect one to your inbox or documents, know what data you hold, keep the most sensitive data out, and read the vendor's data terms for your exact plan. Then name who may use the tool, have a person approve its output, and write a one-page policy.

This is general information, not legal advice. Terms change, so check each vendor's current pages yourself.

Start by classifying the data you hold

The FTC's guide for business, Protecting Personal Information, lists five principles: take stock, scale down, lock it, pitch it, and plan ahead. The first one applies directly to AI. It tells you to identify the sensitive personal information you hold and everywhere it is stored, including vendor systems. An AI tool is now one of those places.

A simple three-tier sort is enough for a small team:

  • Public: your website copy, price list, published reviews, and product descriptions.
  • Internal: schedules, draft proposals, process notes, and vendor quotes with no customer details.
  • Sensitive: anything that identifies a customer or employee, plus payment, health, financial, legal, or login information.

Write the tiers down. The rest of this checklist depends on them.

What never goes into an AI tool

Unless you have checked the vendor's terms for your plan and decided the risk is acceptable, keep these out:

  • Payment card numbers, bank account numbers, and Social Security numbers.
  • Passwords, API keys, and one-time codes.
  • Health, medical, or insurance details.
  • Full customer lists, or a spreadsheet export with names, emails, and phone numbers.
  • Signed contracts, legal correspondence, and anything under a confidentiality agreement.
  • Employee records, including performance notes and anything used to decide who is hired, promoted, or let go.

The FTC's second principle, scale down, helps here. If you do not need to share a detail to get the task done, remove it. Example (hypothetical): replace "Maria Lopez, 14 Elm St." with "Customer A" before you paste a complaint into a drafting tool.

What to ask a vendor

Vendor terms differ, and they differ between a free consumer plan and a business plan from the same company. Do not assume one vendor's answer applies to another. As an example of how much they vary, here is what four pages said when we read them on November 18, 2026:

  • Anthropic's privacy policy, effective September 10, 2026, says it may use inputs and outputs to train and improve its models "unless you opt out through your account settings," and that some uses continue even after an opt-out, such as safety reviews. A separate commercial-products page, dated August 18, 2026, says that by default it will not use inputs or outputs from its commercial products to train its models, though training can happen if you opt in, for example by reporting feedback.
  • Google's Gemini Apps Privacy Hub says human reviewers review some data, and advises: "Please don't enter confidential information that you wouldn't want a reviewer to see." It says chat history is deleted after 18 months by default, and that reviewed chats are kept for up to three years.
  • Microsoft's Copilot privacy page says prompts, responses, and data accessed through Microsoft Graph are not used to train its foundation models, and that Copilot only surfaces organizational data a user already has at least view permission to see.

These pages will change, and plans have separate terms. Re-read the page for your plan before you rely on any of this. The FTC has also said, in a February 2024 staff post, that it may be unfair or deceptive to quietly and retroactively loosen data practices. That is a reason to check again later, not a reason to stop checking.

Question to askWhat a good answer sounds like
Is our content used to train your models?A plain yes or no for our plan, in writing, with the settings page that controls it.
Do people at your company or your vendors read our chats?A clear description of when human review happens and how to turn it off or avoid it.
How long do you keep our data, and can we delete it?A stated retention period and a way to delete, with the deletion timeline.
Who owns what we put in and what comes out?Your inputs stay yours. Output rights are stated in the terms.
Who can see our data inside the tool?Access follows the permissions you already set, with admin controls and logs.
What happens if you have a security incident?A commitment to notify you, in the contract. The FTC's guide says to require providers to report incidents.
Will you tell us before the terms change?A notice process, not a quiet edit.

Accounts and settings

Use a business or team account if one exists, owned by the company and not by one employee's personal email. Then check the settings:

  • Turn off training or "improve the model" options where they exist.
  • Turn off or limit chat history if you do not need it.
  • Turn on two-step sign-in for every user.
  • Review connections to your inbox, calendar, and shared drive. Connect only the folders the tool needs, not the whole drive.
  • Remember that a connected tool can reach whatever the connected account can reach. Clean up old shared folders first.

Who may use which tool

Pick the tools your team may use and say so. Allow the approved tool for public and internal data. Allow sensitive data only in a tool you have reviewed, and only for people who need it. Ban personal accounts for company data. If someone wants a new tool, they ask first.

A person approves what goes out

AI tools make mistakes, and they can invent details. Nothing should reach a customer, a contract, or a payroll decision without a named person reading it first. The NIST AI Risk Management Framework is voluntary, but it backs the idea. It says documentation "can enhance transparency, improve human review processes, and bolster accountability," and one of its subcategories (MAP 3.5) calls for processes for human oversight to be defined, assessed, and documented. For a small team, that means a short written rule on who reviews what.

A one-page policy outline

Keep it to one page. Cover these points:

  1. Purpose: why the business uses AI tools.
  2. Approved tools: the list, the plan for each, and the owner of each account.
  3. Data tiers: what is public, internal, and sensitive, and which tier each tool may touch.
  4. Never paste: your list from above.
  5. Review: who approves outputs before they go to a customer or affect a person.
  6. Connections: who may connect a tool to email or documents, and to which folders.
  7. Incidents: who to tell, and how fast.
  8. Review date: when you will re-check vendor terms. Put a recurring reminder on the calendar.

If something is pasted by mistake

The FTC's last principle is plan ahead, and its breach response guide is a useful model. Act in this order:

  1. Stop using that chat and tell the person in charge.
  2. Write down what was pasted, which tool, which account, and when.
  3. Delete the chat or use the vendor's deletion tools, and read the vendor's terms to see how deleted data is handled.
  4. Change any password or key that was pasted.
  5. Decide whether customers or others must be told. Many states have breach notice rules, so for sensitive data ask a lawyer.
  6. Update the policy so it does not happen again.

A note for Illinois employers

If you employ people in Illinois, the Illinois Human Rights Act now has a section on AI. Public Act 103-0804 took effect January 1, 2026. It makes it a civil rights violation for an employer to use AI for recruitment, hiring, promotion, discharge, discipline, and similar decisions if it has the effect of discriminating against protected classes, or to use zip codes as a proxy for protected classes. It is also a violation to fail to give an employee notice that AI is used for those purposes. The law tells the Illinois Department of Human Rights to write rules on when, how, and in what time period notice is given. Check the current Department rules and ask a lawyer before you use AI in any employment decision.

Where we fit

We build and run AI workflows for small businesses, and our AI and automation plans are priced publicly: Monitor is $149 a month with setup from $349. You own your accounts and data. If you are also curious how AI tools describe your business, see how to check what ChatGPT, Gemini and Perplexity say about you. Prices are on our pricing page.

Quick answers

Is it safe to paste customer information into ChatGPT, Claude or Gemini?

It depends on the product, the plan and the settings, and the terms differ by vendor and change over time. Read the vendor's current data terms, use a business account where one exists, and keep the most sensitive data out of any tool you have not checked.

Do AI tools train on what I type?

Some do by default on some plans and some do not. For example, as of November 2026 Anthropic's privacy policy says consumer chats may be used for training unless you opt out, while its page on commercial products says it does not train on their inputs or outputs by default. Check the exact page for the exact plan you use.

What should I do if an employee pasted something they should not have?

Stop using that chat, tell the person in charge, and write down what was pasted, where and when. Then check the vendor's controls for deleting it, review the vendor's terms for how that data is handled, and consider legal advice if the data was sensitive.

Do I need a written AI policy for a small team?

A one-page policy is enough for most small teams. It should say which tools are allowed, what data never goes in, who approves outputs, and who to tell if something goes wrong.

Is this legal advice?

No. This is general information for small business owners. Rules depend on your industry, your state and your contracts, so ask a lawyer about your own situation.

Sources

  1. FTC: Protecting Personal Information: A Guide for Business The five principles (take stock, scale down, lock it, pitch it, plan ahead) and service provider guidance; checked Nov 18, 2026
  2. FTC: Data Breach Response: A Guide for Business First steps after personal information is exposed; checked Nov 18, 2026
  3. FTC Office of Technology: AI (and other) Companies: Quietly Changing Your Terms of Service Could Be Unfair or Deceptive Feb 13, 2024 FTC staff post on retroactive changes to data terms; checked Nov 18, 2026
  4. NIST: Artificial Intelligence Risk Management Framework (AI RMF 1.0) Voluntary framework, January 2023; four functions; documentation and human oversight; checked Nov 18, 2026
  5. Anthropic: Privacy Policy Consumer training and opt-out wording, effective Sep 10, 2026; checked Nov 18, 2026
  6. Anthropic Privacy Center: opting out of model training Commercial products and training, article dated Aug 18, 2026; checked Nov 18, 2026
  7. Google: Gemini Apps Privacy Hub Human review, retention and the confidential information warning; checked Nov 18, 2026
  8. Microsoft Learn: Data, Privacy, and Security for Microsoft Copilot Training statement and permission model for Microsoft 365 Copilot; checked Nov 18, 2026
  9. Illinois Public Act 103-0804 (HB 3773) Section on employer use of artificial intelligence; effective Jan 1, 2026; checked Nov 18, 2026

Keep reading

Not sure what you need?

Take the survey to find out exactly what your business needs.

Seven quick questions, about two minutes. We recommend a package and write your message for you. A person replies within 2 business days.

Take the survey
Ready when you are

Have something to grow?

Tell us what you are working on and where you want it to go.

We reply within 2 business days.

Say hello
↗