Skip to content

If I Paste Customer Information Into AI, Where Does It Go?

November 16, 2026

An accountant asked a version of this question online:

“I want to use ChatGPT to analyze client data for my accounting business, but I’m worried about possible security risks. Will people be able to access the data I put into ChatGPT?”

That’s a good question, and it deserves a real answer instead of reassurance.

The short version: it depends on two things. Which account you’re signed into, and what you paste. Both are under your control, and sorting them out takes about fifteen minutes.

You’re not being paranoid

This is the single most common worry owners have about AI.

A 2025 Paychex survey of about 1,100 small business owners and HR professionals found data privacy and security was the top concern, named by 54%. For 31%, it was the main reason they hadn’t adopted AI at all. (Paychex, March 2025)

So a third of owners are sitting out entirely over a question nobody answered for them.

What actually happens when you paste something in

Four things, roughly in this order.

It leaves your computer. The AI doesn’t run on your laptop. Your text travels to the company’s servers and gets processed there.

It gets stored. Your conversation sits in your chat history until you delete it. When you do delete it, OpenAI says it’s removed from their systems within about 30 days.

A person might read it. Not routinely. But conversations can be reviewed by staff for safety checks, and on personal plans, clicking thumbs up or thumbs down can send that whole conversation for review.

It might be used to train the model. This is the part people actually mean when they ask the question. It’s also the only part that changes depending on your account.

One fear worth putting to rest: your client’s numbers are not going to pop out in a stranger’s chat window tomorrow. That’s not how these systems work. The real concerns are plainer. Your customer’s information is sitting on a company’s server, it may be read by someone, and it may become part of what the model learns from.

The part that depends on your account

There are two doors into the same tools, and they have different defaults.

Personal accounts are opted in to training by default. That covers ChatGPT Free, Go, Plus and Pro, and the equivalent Claude plans. Your chats help improve the model unless you turn that off.

Business accounts are not. ChatGPT Business and Enterprise, the OpenAI API, Claude for Work, and Gemini inside Google Workspace all exclude your content from training by default. There’s no toggle to remember and no setting for a new employee to get wrong.

If you handle customer records, use a business account. ChatGPT Business runs about $25 per person per month, less if you pay annually, with a two-seat minimum. That is not what’s stopping people. Not knowing the difference is what’s stopping people.

Three settings worth fifteen minutes

1. Turn off model training. In ChatGPT, go to Settings, then Data Controls, then switch off “Improve the model for everyone”. In Claude, it’s in your privacy settings. On a business account this is already the default, but it costs nothing to check.

2. Decide how long things stick around. Business plans let an admin set how long conversations are kept for the whole workspace. On a personal account, deleting old chats is the version of this you have.

3. Look at what’s already in there. Both ChatGPT and Claude let you export your full history. Download it, read through it, and delete anything that shouldn’t be there. If your staff have been using AI for a while, this is worth an hour.

Two honest caveats. Turning off training doesn’t cover feedback you submit on a personal plan. And conversations flagged for safety review can still be looked at. Opting out reduces your exposure. It doesn’t take it to zero.

The list that matters more than the settings

Most real exposure isn’t technical. It’s a tired owner pasting a whole client file at 9pm because it was faster than trimming it.

So write down a rule and stick it somewhere you’ll see it.

Keep out of the chat window:

  • Full client files and account records
  • Social security numbers, tax IDs, and bank or card details
  • Medical records, and anything else covered by a privacy law in your trade
  • Signed contracts and anything under an NDA
  • Employee records and payroll detail
  • Passwords, API keys, and logins
  • Anything a client specifically told you not to share

Fine to work with:

  • The numbers without the names
  • A description of the situation instead of the document itself
  • A paragraph with the identifying details swapped out
  • Your own templates, policies, pricing, and marketing copy

The trick is to strip the identity and keep the substance. “A client owes $48,000 and is 94 days overdue, what are my options?” gets you the same answer as pasting the real ledger. It just doesn’t hand over your client’s name to do it.

What settings don’t fix

A toggle doesn’t override an agreement you signed.

If your client contract says their information doesn’t go to a third party without consent, that still applies. Professional rules for accountants, lawyers, healthcare, and financial advice still apply. If you have customers in the EU, their names, emails, and order histories are personal data under GDPR, and how you handle them is your responsibility, not the AI company’s.

None of that means you can’t use AI. It means the contract is a separate question from the settings, and it’s worth ten minutes with the actual document before you start.

The honest answer: it isn’t “AI is fine” and it isn’t “never paste anything.” It’s this. Use a business account, turn off training, and don’t paste what you don’t need to paste.

Keep reading

Want to talk it through?

Tell me about the task that takes up your team's time.

Newsletter

Plain-language notes on software and AI for small business

One short email when there is something worth sharing. No spam. Unsubscribe any time.