Insights

Is It Safe to Put Customer Data in ChatGPT, Claude or Copilot? A Plan-by-Plan Answer for Small Businesses

By Felix Rodgers, ASPC, founder of Caprock Agility

Usually yes on a business plan, rarely on a personal one. ChatGPT Business and Enterprise, Claude Team and Enterprise, and Microsoft's work Copilot don't train on your data by default and come with business terms. Free and personal paid plans can use chats for training unless someone switches it off, so keep customer data off them.

That's the short version. The longer version matters, because in most small businesses I talk to, the real question isn't "which AI is safe?" It's "what are my people already pasting, and into which account?" Here's how to answer both.

This is practical guidance, not legal advice. If you're in a regulated field such as healthcare, law or financial advice, check your own rules and talk to your attorney.

Why does the plan matter more than the tool?

Because the same chatbot comes with very different data terms depending on who pays for it.

OpenAI, Anthropic, Microsoft and Google all draw a line between personal (consumer) accounts and business accounts. On the business side, the vendor typically commits in writing not to train its models on your content by default, and your admin gets controls over users and, on some plans, retention. On the personal side, those decisions sit with each individual user's settings.

So a staff member using ChatGPT on a personal login and a staff member using your company's ChatGPT Business workspace are not doing the same thing, even if the screen looks identical.

This isn't a hypothetical problem. In Pax8's Q3 2026 SMB AI Pulse (October 12, 2026), 35% of AI-engaged small businesses said they had found AI running without approval in the past six months. If that's happening, some of it is probably on personal accounts.

What do the major AI tools do with your data?

Here's what each vendor's own pages say, as of October 26, 2026. Terms change, so check the linked page before you rely on it.

Tool and planTrains on your content by default?What the vendor says
ChatGPT Free, Plus, Pro (personal)Yes, unless you turn it offOpenAI's data controls page says data sharing is on by default for personal Free, Plus and Pro workspaces; you can turn off "Improve the model for everyone."
ChatGPT Business, Enterprise, APINoOpenAI's business data page says it does not use inputs or outputs from these products for training by default, and qualifying organizations can configure retention.
Claude Free, Pro, Max (personal)Only if the user allows itAnthropic's August 2025 consumer terms update made training a user choice; if it's on, data can be kept for five years, and if it's off, the 30-day retention period continues.
Claude Team, Enterprise, APINoAnthropic's privacy center says it won't use inputs or outputs from commercial products for training by default.
Microsoft Copilot and Copilot Chat (work account)NoMicrosoft's enterprise data protection page says prompts, responses and Microsoft Graph data aren't used to train foundation models, and your retention policies and permissions apply.
Gemini in Google Workspace (business account)Not outside your domain without permissionGoogle's Workspace privacy commitments say customer content isn't used for model training outside your domain without permission.

Two details people miss. First, the feedback buttons: Anthropic says that if you click thumbs up or down, it stores the whole conversation for up to five years and may use it for training, even on commercial plans, and OpenAI says the same feedback can make a conversation usable for training after you've opted out. Second, Microsoft notes that web search queries Copilot sends to Bing are handled under separate terms from the rest of your Copilot data.

Does turning off training make a personal account safe for customer data?

It helps, but it isn't the same as a business plan.

Turning off training on a personal account stops new chats from being used to improve the vendor's models. It doesn't give your business a contract with the vendor, admin control over who has access, a way to pull the account back when someone leaves, or (on most personal plans) control over how long data is kept. OpenAI's Temporary Chat, for example, keeps chats out of history and out of training, but OpenAI says they may still be retained for up to 30 days for safety purposes.

My rule for clients is simple: customer work happens on accounts the business owns. Personal accounts are fine for public questions and your own writing, with nothing that identifies a customer.

What should never go into an AI tool?

Some data shouldn't go into a chatbot on any plan. Here's the four-tier list we use as a starting point. Adjust it to your industry and contracts.

  1. Open (any account, training off): public information, industry questions, templates, and your own drafts with names and numbers removed.
  2. Business plan only: customer names, emails, order history, project notes, quotes and internal documents, on a business account where the vendor's terms say no training by default.
  3. Business plan plus a named approver: contracts, pricing and deal terms, employee and personnel matters, anything under an NDA, and financial details about a specific customer.
  4. Never paste: Social Security numbers, full card or bank account numbers, passwords and login codes, health information (unless you have a vendor agreement that covers it, such as a HIPAA business associate agreement), and anything a contract or court order says can't leave your systems.

Put this list on one page and share it with everyone who uses AI at work. A rule nobody can find is not a rule.

What should a small business do about AI and customer data this week?

You don't need a big governance program. You need five things done:

  1. Find out what's in use. Ask the team, without blame, which AI tools they use and on which accounts. You'll learn more from an honest five-minute conversation than from a policy memo.
  2. Pick one business plan per job. For most small teams that means one general assistant on a business plan, plus the AI already built into the software you pay for, such as Microsoft 365 or Google Workspace. Fewer tools means fewer places for data to go.
  3. Change the settings once. Turn off training anywhere it's on, turn off feedback buttons where your admin can, set retention where your plan allows, and make sure everyone signs in with a work account.
  4. Publish the four-tier list. Use the list above as a draft and make it yours.
  5. Keep a one-page register. Tool, plan, admin owner, retention setting, date of last review. Check it every six months, because vendors change their terms. Anthropic's consumer change in 2025 is a recent example.

Then have a person check AI output before anything reaches a customer. No plan does that part for you.

How does Caprock Agility help with AI and customer data?

Most of the businesses we work with don't need a 40-page AI policy. They need to know where AI is already touching customer data and how to put real work on safe ground.

Our AI Readiness Audit ($2,500, 3–5 business days) maps how work actually moves today, including your tools and data, and gives you risk flags alongside ranked AI opportunities. If you already know the workflow you want to fix, EDGE Starter (starting at $8,500, 4–6 weeks) builds one workflow end to end with training and a Day 30 outcomes review, set up on accounts your business controls.

Still deciding whether outside help is worth it? Read Is an AI Consultant Worth It for a Small Business?, or compare every option on our services and pricing page.

Frequently asked questions

Is it safe to put customer data in ChatGPT?

On ChatGPT Business or Enterprise, OpenAI says it doesn't train on your data by default, so customer work is reasonable with sensible rules. On personal Free, Plus or Pro accounts, training is on by default unless turned off, and your business has no contract or admin control, so keep customer data off them.

Does ChatGPT Plus train on my data?

By default, yes. OpenAI says data sharing is enabled by default on personal Free, Plus and Pro workspaces. You can turn it off under Settings, then Data controls, then "Improve the model for everyone." ChatGPT Business and Enterprise don't train on your data by default.

Is Claude safer than ChatGPT for business data?

On business plans they're broadly similar: both Anthropic and OpenAI say they don't train on commercial customers' data by default. On personal plans the details differ. Anthropic made training a user choice in 2025 and keeps data for up to five years if it's on. Compare the plans, not the brands.

Is Microsoft Copilot safe for customer data?

When staff sign in with a work account, Microsoft says prompts, responses and Microsoft Graph data aren't used to train foundation models, and your organization's permissions and retention policies apply. Web searches Copilot sends to Bing are covered by separate terms, so keep sensitive details out of prompts that need the web.

Can employees use personal AI accounts for work?

We recommend against it for anything involving customers. Even with training turned off, a personal account gives the business no contract, no admin control and no way to recover the history when someone leaves. Personal accounts are fine for public questions and drafts with identifying details removed.

Do I need an AI policy if I'm a small business?

Yes, but it can be one page: which tools and plans are approved, what data can go in each, who approves exceptions, and who reviews the list. Pax8 found 35% of AI-engaged small businesses discovered unapproved AI in the past six months, so a short written rule is worth it.

Does Caprock Agility only work with businesses in Texas?

No. Caprock Agility is based in Lubbock, Texas, and works with businesses across the United States.

Not sure where AI is touching your customer data? Book a free Workflow Fit Call and we'll help you find the gaps and the safest first workflow.