# What Employees Should (and Shouldn't) Paste Into AI

A plain-language guide to AI security for employees: what's safe to paste into Claude, ChatGPT, and Gemini, what isn't, and why the plan your company is on matters more than the tool itself.

Most companies now have someone pasting client details, financial numbers, or draft contracts into an AI chat window without a second thought. It's not recklessness, it's that nobody's told them what actually happens to that text once it's sent, or that the answer changes depending on which login they're using. Here's the version of that conversation that's short enough to actually remember.

## The plan matters more than the tool

The single biggest factor in whether your pasted text becomes training data isn't which AI you use, it's which plan you're logged into. Across Claude, ChatGPT, and Gemini, the pattern is the same: free, personal accounts train on your conversations by default, and you have to go find the setting to opt out. Paid business, team, and enterprise accounts do not train on your data by default, full stop.

That means the intern using the free version of an AI tool on their personal account is operating under completely different rules than the same person logged into a company-paid seat. If your team is using free accounts for work, you don't have an AI security problem so much as a "we're on the wrong plan" problem.

## A simple rule, not a policy document

Nobody reads a 12-page AI acceptable use policy. What sticks is a rule they can apply in three seconds, before they hit paste.

**Never paste:**
- Social insurance numbers, financial account numbers, health records
- Anything covered by an NDA
- Unreleased legal matters, HR investigations, termination details

**Pause first:**
- Client names attached to specifics (a contract value, a complaint, a diagnosis)
- Internal financials, strategy documents, anything you'd wince at seeing in a competitor's inbox
- Full documents, when a summary or a redacted version would do

**Generally fine:**
- Your own drafts and writing
- Publicly available information
- Anonymized versions of the above: swap real names for "Client A," round the numbers, strip the letterhead

The test that covers most of this in one line: if you wouldn't paste it into a public forum post, don't paste it into a free AI account either.

## Where Claude, ChatGPT, and Gemini actually differ

All three follow the same shape, free and personal tiers lean toward training by default, paid business tiers lean away from it, but they're not identical, and the differences are worth knowing in plain terms:

- **Claude (Anthropic):** Free and Pro accounts can be used to train future models unless you opt out in settings. Business, Enterprise, and API usage do not train on your data.
- **ChatGPT (OpenAI):** Free and Plus accounts are used for training by default unless you turn it off. Team, Enterprise, and API usage are excluded from training by default, no opt-out needed.
- **Gemini (Google):** Free, personal Gemini activity feeds training unless you turn off activity history or use temporary chat. Google Workspace and Enterprise usage isn't used for training and isn't reviewed by humans.

None of this is a reason to avoid a particular tool. It's a reason to check which account your team is actually logged into before assuming any of them are "safe."

## The real fix isn't vigilance

Telling employees to be careful only works until the fourth time that day they're moving fast and forget. The more durable fix is removing the free-tier trap in the first place: put the team on paid business accounts where training is off by default, and write the three-tier rule above into something you hand out once instead of repeating from memory.

If your business is already having the "what are we allowed to do with AI" conversation internally, that's usually the sign it's worth twenty minutes with someone who's done it before, not a bigger document.