What you can (and can’t) put into AI at work

You’ve got access to ChatGPT, Copilot or whatever your company rolled out last quarter, and you can probably name a dozen tasks a week it could take off your plate. What nobody gave you is a clear answer to the question that actually stops you: is it okay to paste this in? So you do one of two things. You avoid AI at work, and watch colleagues who worry less get faster, or you guess and quietly hope nothing you pasted comes back to bite you. Neither is a great place to work from.

A quick note before we get into it: I’m not a lawyer, and your employer’s policy outranks anything you read here, this post included. What I can give you is a better question. “Is this okay?” is too vague to answer, which is exactly why it keeps you stuck. The useful question is which bucket this belongs in.

Three buckets

Green is for anything that doesn’t identify anyone or reveal anything private. Public information, your own writing, generic structures and templates all live here. If you want help outlining a presentation, tightening a paragraph you wrote, brainstorming names for a workshop or turning messy notes into a clean agenda, paste away.

Yellow is for material that becomes sensitive because of its context. Internal strategy, a draft that names a client, notes from a meeting that wasn’t public. None of it is a secret on its own, but together it tells a story you wouldn’t want floating around. Before it goes in, strip it down: names become roles, exact figures become ranges, and the details that make it recognizable come out. Or use a tool your company has specifically approved for this kind of work.

Say you want help with the wording of a tricky performance conversation. You don’t need to hand over the whole file to get good advice.

The assistant still has everything it needs to help you with tone and structure, and nothing it doesn’t.

Red means it doesn’t go in. Personal data about customers or colleagues, passwords and access credentials, anything covered by an NDA or by regulation, such as health, financial or legal records, and anything your policy names specifically. When you can’t decide whether something is yellow or red, treat it as red. Being too careful costs you a few minutes, and being careless can cost a lot more.

Three habits that make it routine

Read the policy once, all the way through. It’s usually shorter than you fear. If your company doesn’t have one, that’s useful information in itself, so ask for one in writing. That way the decision isn’t resting on your shoulders alone.

Look at the data settings of the tool you’re using. Whether your conversations are used to train future models depends on the product and the plan, and defaults change. Checking takes five minutes, which is a small price for not having to assume.

Make redacting a reflex. The first few times it feels slow. After a week or two you’ll catch yourself typing “the client” instead of their name without thinking about it, and that’s the goal: a habit you don’t have to remember is a habit that holds when you’re busy.

Why it’s worth getting right

Women are often asked to lead the AI conversation at work without ever getting a private place to ask the basic questions first. These are the questions you won’t ask out loud in a meeting, and they deserve straight answers, without an audience.