10 Things You Should Never Type Into ChatGPT or Any AI Chatbot

The 10 Things, Specifically

What Actually Happens to This Information

None of this assumes a hidden conspiracy reading your messages in real time. The realistic picture, based on how major AI companies have described their own practices, is more mundane: conversations get stored, may be used to improve future models unless you opt out, and are occasionally reviewed by human staff as part of safety and quality processes rather than continuously monitored. That's a real privacy consideration, not a dramatic one — but it's exactly why "this probably doesn't matter" isn't the same as "this is actually fine."

The professional-confidentiality items on this list (client data, patient information, active legal matters) carry a second layer of risk beyond the AI company itself: pasting that information into a general-purpose chatbot can violate your own professional or legal obligations, regardless of what happens to the data afterward. That's a risk you're taking on personally, separate from the AI platform's own privacy practices.

How to Get the Same Help Without the Risk

Why This List Matters Even More on a Work Account

Several items on this list carry extra weight in a professional context specifically because the confidentiality obligation isn't yours alone to waive. Client data, patient information, and proprietary business details are often covered by contracts, professional licensing rules, or company policy that exists independent of whatever the AI provider does with the input. A well-meaning employee pasting a client's contract into a general chatbot to get a quick summary can create a real compliance problem, even if the AI company itself handles the data responsibly, simply because the client never consented to their information being processed by a third-party tool at all.

This is one of the clearest arguments for using a company-approved AI tool with the right data agreements in place, rather than whichever chatbot happens to be open in another browser tab. If your workplace hasn't set clear guidelines on which AI tools are approved for which categories of data, that's worth raising directly with IT or legal — the ambiguity itself is a risk, independent of whether anything has gone wrong yet.

Frequently Asked Questions

Is it fine to share this kind of information with a paid or enterprise AI account?

Generally lower-risk than a free consumer account, since business and enterprise tiers typically exclude conversations from model training by default and come with stronger contractual data protections. It's still worth checking the specific provider's terms rather than assuming, since protections vary between companies and tiers.

What if I already accidentally shared something sensitive?

Delete the conversation as soon as you notice, and check your account's data controls for any option to request removal from training data specifically. Deletion isn't always instant on the backend — most platforms retain deleted conversations briefly for safety review — but it still meaningfully reduces long-term exposure.

Does this apply to voice assistants and AI features built into other apps too?

Yes, and often people forget this — AI features embedded in email, browsers, and productivity tools follow the same general caution, even though they don't look like a standalone chatbot window. Check the specific privacy settings for each tool rather than assuming your chatbot privacy settings automatically extend everywhere.

Is there a real difference between "training data" risk and "human review" risk?

Yes — training data risk is about your specific phrasing potentially influencing future model behavior in ways that are hard to reverse, while human review risk is about a staff member potentially seeing the literal content for safety or quality purposes. Both are reasons for caution, but they're different mechanisms and different settings often control each one separately.

Does it matter if I'm using a "temporary" or "incognito" chat mode?

It generally reduces retention and training risk compared to a standard conversation, since most platforms exclude temporary chats from long-term history and training by default. It doesn't necessarily eliminate the human-review possibility entirely if content gets flagged for safety reasons, so it's a meaningful risk reducer rather than a full guarantee.