Is It Safe to Paste Client Data Into ChatGPT? I Read the Fine Print for ChatGPT, Claude and Gemini

Short answer: it depends on your plan and one setting. On personal plans, ChatGPT and Gemini can train on what you paste unless you switch it off. Claude only trains on your chats if you said yes. Business plans of all three don't train on your chats by default. And none of these settings stop the company from storing what you pasted.

So before a client list goes anywhere, check three things: which plan you're on, how one switch is set, and what's in the list.

I'll show you all three. I read the privacy and help pages for ChatGPT, Claude and Gemini in full on October 22, 2026, then opened my own settings. One of them surprised me.

Who trains on your chats, plan by plan

This is the part most people get wrong, because the answer changes with the plan you pay for.

"Personal" here means Free, Plus, Pro and Max plans. "Business" means ChatGPT Business and Enterprise, Claude Team and Enterprise, and paid Google Workspace accounts.

If you run a small business on a personal ChatGPT Plus account, you're on the left side of that table. Paying $20 a month doesn't move you to the right side.

How to turn off training in ChatGPT

ChatGPT has one switch, and it's easy to find once you know its name.

Go to Settings, then Data Controls, then "Improve the model for everyone", and turn it off.

Off means your new chats stop going into training. Mine was already off. OpenAI also offers a "do not train on my content" request in its Privacy Portal, and either one is enough.

Two details from OpenAI's own help page on how your data is used matter here:

• Temporary chats aren't used for training, but OpenAI keeps them for up to 30 days for safety checks.
• ChatGPT Business, Enterprise, Edu and the API aren't used for training by default.

How to check Claude's training setting

Claude works the other way round. Since Anthropic's privacy policy change took effect on September 28, 2025, personal Claude accounts (Free, Pro and Max) are only used for training if you choose to allow it.

The problem is how most of us made that choice. It came up as a pop-up, we clicked something, and we never looked again.

Go to Settings, then Privacy, then "Help improve our AI models" to see what you picked.

Here's why it's worth the 20 seconds. According to Anthropic's training page, if the setting is on, your chats can be kept de-identified for up to 5 years in training pipelines. If it's off, a chat you delete is removed from Anthropic's back-end storage within 30 days (retention details here).

Incognito chats in Claude are never used for training. And content Claude reads through a connector, like Google Drive, isn't used for training either, unless you copy it into the chat yourself.

Gemini's Keep Activity setting, and why mine was on for 3 years

This is the one that got me.

On a personal Google account, Gemini's Keep activity setting is on by default. With it on, Google uses your chats to improve its AI, including training. Google's Gemini Apps Privacy Hub also says a subset of chats is read by human reviewers, and it tells you plainly not to enter confidential information you wouldn't want a reviewer to see.

I checked mine. Keep activity had been on since May 2023. I had no idea.

To turn it off, go to myactivity.google.com/product/gemini and set Keep activity to off.

What off actually changes:

• New chats aren't used for training (unless you send feedback, more on that below).
• Google still keeps your chats for 72 hours.
• Chats a reviewer has already read are disconnected from your account and kept for up to 3 years, even if you delete your activity.

If your company pays for Google Workspace, the rules are different. Google's Workspace generative AI privacy page says there's no human review and no training outside your domain without permission.

The thumbs up button can undo your settings

This is the clause I'd never have found without reading the whole page.

When you click thumbs up or thumbs down on a reply, you're sending feedback. In all three tools, feedback has its own rules, and they override what you set above.

So if a chat has client details in it, don't rate it. Liked the reply? Copy it and move on.

5 things I never paste into ChatGPT

Settings protect you from training. They don't make a chatbot the right place for everything. These five stay out of every AI tool I use, whatever the plan.

1. Client names next to their emails or phone numbers. That's their personal data, and it isn't mine to share.
2. Deal values next to company names. One leak and my pipeline is public.
3. Contracts or anything under NDA. Pasting it can break the NDA by itself.
4. Passwords, API keys and logins. Chats are stored, and some get read by people.
5. Work I want credit for. Ask the mathematician in the next paragraph.

That last one isn't hypothetical. In September 2026, NYU mathematician Tristan Buckmaster said OpenAI's claimed solution on the Navier-Stokes problem followed his own approach, after he'd checked his work in Codex, OpenAI's coding tool. Futurism covered the dispute and Nature followed up. OpenAI denies its people saw his work, but says it "cannot rule out" that de-identified usage data helped its models. Nobody has settled it. I'd rather not be the next example.

How I paste a client list now

I still use AI on client work. I just strip out who it's about first. Here's what I do every time.

1. Swap the names for Client A, Client B, Client C.
2. Delete the emails, phone numbers and addresses. Usually the whole column.
3. Round deal values into ranges, like $10k to $25k.
4. Use a temporary chat in ChatGPT or Gemini, or an incognito chat in Claude.
5. Skip the thumbs up on that chat.

You still get the analysis. The AI never learns who it's about.

If the list is long, I don't paste it to get it cleaned, because that would defeat the point. I give the AI the column headers only and ask for spreadsheet formulas:

I have a spreadsheet with these columns: [paste the header row only]. Write me a Google Sheets formula, one per column, that:
1. replaces each company or person name with Client A, Client B, Client C
2. clears every email, phone number and street address
3. rounds any deal value to a range like $10k to $25k
Don't ask me for the data. I'll run the formulas myself.

The real list never leaves my sheet.

Turning training off doesn't stop storage

Here's the catch that sits under all of this. Every setting above controls training. None of them stops the company from keeping what you pasted for a while.

• ChatGPT keeps temporary chats for up to 30 days.
• Gemini keeps chats for 72 hours even with Keep activity off.
• Gemini chats a reviewer already read stay for up to 3 years, even after you delete them.

And there's one more check that has nothing to do with AI settings. Read your client contract. Some say no AI tools at all, on any plan. If yours does, no setting fixes that.

Get the next one

I write about building smarter systems with AI. Subscribe to Flow Blueprints to get the next test in your inbox.

Facts checked October 22, 2026. These policies change, so check them again before you rely on them.