Is it safe for my team to paste customer and company information into ChatGPT or Claude?

By Ranjit Bhinge, founder of Blur Studio · Updated · 6 min read

Short answer: It depends on which account your team uses and what they paste. ChatGPT is OpenAI's chat tool and Claude is Anthropic's. On OpenAI's individual plans (Free, Plus, Pro), your conversations may be used to train its models unless you turn that off. On OpenAI's business plans and its API (the way software connects to the model), inputs and outputs are not used for training by default. Anthropic works the same way for its commercial products. Even so, never paste passwords, card numbers, government ID numbers or health records, and ask a person with authority before pasting client data.

Ranjit Bhinge

Your staff are already using AI chat tools, and you want to know what is safe to paste and what is not.

I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.

If you want a clear picture of where AI fits in your business without risky shortcuts, the AI Strategy Sprint covers it in 3 days.

The plain answer

Pasting company data into an AI tool is not automatically unsafe, and it is not automatically fine. Three things decide it: which account type you use, what the data is, and what you promised your clients. This page gives you a simple rule sheet. It is general information, not legal advice. If you handle client files under confidentiality duties, ask a lawyer.

Personal accounts, business plans and the API

The same tool behaves differently depending on the plan. As of November 2026, based on the vendors' own help pages:

Type of accountUsed to train models by default?What to do
ChatGPT Free, Plus or Pro (OpenAI)May be. OpenAI says it may use your content to train its models, and you can opt out with "Improve the model for everyone" under Settings, Data controls, or in its Privacy Portal.Turn the setting off. Use Temporary chat for sensitive one-off questions: OpenAI says temporary chats are not used to improve its models while they stay temporary.
ChatGPT Business, Enterprise or EduNo by default. OpenAI says it does not use inputs or outputs from these plans, or its API, to improve its models.Use this for team work. Check your admin settings.
OpenAI APINo by default, unless your organisation chooses to share data. Logs for abuse monitoring are kept for up to 30 days by default, with an approval-based Zero Data Retention option.Use for custom tools. Ask your developer about retention.
Claude Free, Pro or Max (Anthropic)Only if you allow it through the "Model Improvement" setting, or in other limited cases such as chats flagged for safety review. Incognito chats are never used.Check the setting. Use Incognito for sensitive questions.
Claude for Work and the Anthropic APINo by default. Anthropic says it will not use your inputs or outputs for training by default, though chats you rate with thumbs up or down may be used.Use for team work and custom tools.

Two catches apply to both vendors. First, feedback: on OpenAI's individual plans, if you click thumbs up or thumbs down, the whole related conversation may be used for training even if you opted out. Anthropic says feedback conversations can be stored for up to 5 years, and on its commercial products it may use feedback chats for training, so the same warning applies there. Tell staff not to rate chats that contain client data. Second, plans change. Vendors update these pages, so check the current wording before you rely on it. Tools other than ChatGPT and Claude have their own terms, so read those too. My guide on AI tools with no training on your data shows how to ask a vendor in writing.

Training is only one risk. Other risks are who in your company can see shared chats, how long the vendor keeps conversations, what a staff member's personal account does when they leave, and whether a client contract allows the data to go to a third party at all. An account that does not train on your data can still break a confidentiality promise.

The traffic-light table

Here is the rule sheet I would give a small team. Use the account type in the left column as a starting point, and add your own examples.

LightDataRule
GreenPublic information, your own marketing copy, generic questions, anonymous examples, your published price listFine in any approved account.
Amber: ask firstCustomer names and emails, internal prices and margins, draft contracts, staff details, supplier terms, meeting notes with named peopleOnly in a business plan or the API with training off, and only for a stated purpose. Remove names where you can.
Red: neverPasswords and login codes, card numbers, bank details, government ID numbers, health records, client files under a confidentiality duty, anything a law or contract says must stay insideDo not paste, in any tool, unless a lawyer or your client has cleared the exact tool in writing.

The US Federal Trade Commission (FTC) guide for business on protecting personal information gives five principles that fit here: take stock of what you hold, scale down to what you need, lock it, pitch it (dispose of what you do not need), and plan ahead. The best AI data rule is the second one. If the task does not need the name, leave it out.

How to remove the sensitive parts

  1. Replace names with labels, such as "Client A" and "Supplier B".
  2. Round or swap figures when the exact number does not matter to the task.
  3. Cut out account numbers, addresses and ID numbers entirely.
  4. Paste the part you need, not the whole file or email chain.
  5. Keep a private note of which label is which, outside the AI tool.

What to do if staff already pasted client data

It will happen, so have a calm plan.

  1. Find out what and where. Which account type was it, what data, which client, when? Ask without blame, or people will hide it next time.
  2. Check the account's settings. Was training on? If it was an individual plan, turn the training setting off now. Opting out applies to new conversations, so it does not undo what already happened.
  3. Delete the conversation and check whether the tool keeps deleted chats for a period. Deleting is good hygiene but does not prove the data is gone from the vendor's systems.
  4. Read the client contract and any privacy law that applies to the data. Some require you to tell the client or a regulator within a set time. Ask a lawyer. If you hold health, financial or children's data, do this the same day.
  5. Write down what happened and what you changed. This helps if a client asks later.
  6. Fix the cause. Give staff an approved account and a one-page rule, so there is a safe route that is as easy as the unsafe one.

Jobs with extra duties

Some work carries confidentiality duties by profession or by law. Law firms owe duties of confidentiality to clients, and accountants hold sensitive financial data. See my guides to AI for law firms and AI for accountants. Your professional body may have its own AI guidance, so check it. For general business use of the tool, see ChatGPT for business.

A setup for custom work that does not train on your data

If you want AI to work over your own documents, such as an internal answer bank or a customer-record assistant, the safer route is usually a custom tool that talks to the model through its API, not staff pasting into a chat window. Both vendors say that, by default, their API and commercial products do not use your inputs or outputs for training. A well-built tool adds the rest:

  • sends only the pieces of data the task needs;
  • keeps your documents in your own storage, with logins and access limits;
  • logs who asked what;
  • uses a vendor setting or agreement that matches your contracts, such as OpenAI's Zero Data Retention (which needs OpenAI's approval) if you need no stored content;
  • has a person check anything that leaves the building.

See my guide to an AI knowledge base for how this works with your own documents. If you are planning which AI work to try first, the AI Strategy Sprint is a paid 3-day first step where we find where AI could help and write down the exact steps, with a free 15-minute AI brainstorm call to start. Wider context is in AI implementation and AI for small business.

How I'd approach this

I would give every staff member the traffic-light sheet on one page this week, and I would buy one business-plan account that the whole team uses so nobody needs a personal one for work. I would turn training off on any individual account that has to stay in use, and I would tell people not to click the rating buttons on chats with client details. For anything red, I would use no AI tool at all until a lawyer or the client says yes in writing. And I would keep a short log of which tools staff use, because you cannot protect data going into a tool you do not know exists.

Questions people ask

Does ChatGPT train on my data?

On OpenAI's individual plans (Free, Plus, Pro) it may, unless you turn off "Improve the model for everyone" in Settings, Data controls. On ChatGPT Business, Enterprise, Edu and the API, OpenAI says it does not use inputs or outputs to improve its models by default. Check OpenAI's help page for the current wording.

Is it safe to put personal data into ChatGPT?

Treat personal data about customers or staff as amber or red. Use a business plan with training off, remove names where you can, and never paste ID numbers, card details or health information. Privacy laws and contracts may limit what you can send to a third party, so ask a lawyer.

Does Claude train on my chats?

On Claude Free, Pro and Max, only if you allow it through the Model Improvement setting, or in limited cases such as chats flagged for safety review. Incognito chats are never used. For Claude for Work and the API, Anthropic says it does not use inputs or outputs for training by default, but chats you rate with the thumbs buttons may be used.

Are business plans completely safe?

They are safer, not risk-free. They remove the training question on the vendor's side, but your contracts, who in your company can see shared chats, retention and staff habits still matter. See ChatGPT for business.

Can I ban AI tools at work?

You can, but staff often use them anyway on personal accounts, which is worse. A clear rule, an approved account and the traffic-light sheet usually work better than a ban.

What if a client forbids AI use on their files?

Follow the contract. Keep their data out of AI tools unless they agree in writing, and tell staff which clients are on the no-AI list. If their wording is unclear, ask them what they mean by it.

How do I get an AI setup that never trains on my data?

Use a business plan or the API, which are not used for training by default at OpenAI and Anthropic, and get the vendor's terms in writing. For custom work, a tool built on the API can also limit what is sent and log access. See AI tool with no training on my data.

More in this topic: Problem solving: AI strategy and adoption (24 guides) →

Sources

Ready for the next step?

A free 15-minute AI brainstorm. We find 3 broad areas where AI could help your business.

Ranjit Bhinge You talk to me, Ranjit. I reply within a few hours.
See the AI Strategy Sprint

Not sure what you need? Tell me what is going on. I will point you the right way, even if it is not me.

Blur Studio · 30 minute call

Book a 30 minute call

You talk to me, Ranjit. If you do not need me, I will tell you.

  • We talk about your business and what is going on.
  • You leave with a clear next step, even if it is not me.
  • I reply to every message within a few hours.
Loading the calendar…