How To Stop ChatGPT From Using Your Ideas To Train Itself

There's a specific kind of professional nightmare that doesn't involve public humiliation, exactly, but comes uncomfortably close. You're sitting in a room, waiting for your moment, and you realize that the moment has already been stolen. Quietly. By an algorithm. And the person holding your work doesn't even know they took it.

This happened. A professor of occupational therapy spent weeks writing a speech for her students' white coat ceremony. She revised it carefully, got it to a place she was proud of, and then — for the first time — fed it to an AI assistant for some light feedback. She didn't change any settings. She didn't think about what she was handing over. Why would she? She was just asking for help with sentence clarity.

On the day of the ceremony, she sat onstage and watched the dean and the provost deliver her speech. Not a version of it. Her speech. The same words, the same structure, the same emotional beats she had built across weeks of drafting. Her best guess: the administrators typed in some keywords — "occupational therapy," "doctoral students," "white coat ceremony" — asked the AI to write something, and the AI obliged by drawing heavily on her work, which it had absorbed because she hadn't told it not to.

She gave her speech anyway. Her students responded with real enthusiasm, which she says felt entirely different from how they'd responded to the administrators' performance of the same words, because they knew her. The material meant something because of who was delivering it.

But the material wasn't only hers anymore. And she hadn't meant to give it away.

The Default Setting Nobody Told You About

Here's the thing most professionals don't know, and that the platforms have little incentive to advertise prominently: the default setting on almost every major AI tool is that your input can be used to improve the model. This is in the terms of service. It is documented. It is also written in the kind of language — "help improve and train our models," "enhance your experience across our services" — that sounds innocuous until you understand what it actually means.

What it means is that the text you put into the chat window may be used to shape future outputs. Not necessarily in the direct, verbatim way that happened to the professor — that's an unusually visible version of a usually invisible process. Normally, your input gets absorbed into a diffuse statistical representation of patterns, and it re-emerges in modified form across millions of other users' queries. But the underlying principle is the same: you gave them the raw material, and they used it.

The professionals most exposed to this are the ones who use AI tools most thoughtfully and most seriously. The person who feeds in a rough draft of a strategy document to stress-test the logic. The manager who pastes in a difficult employee situation to help draft a performance review. The consultant who uploads client research to identify patterns. The entrepreneur who shares a business concept to sharpen the positioning. These are exactly the people generating the most original, high-value intellectual input — and they're doing it inside platforms that, by default, keep what they bring.

Opting out takes two minutes. Here's how, on the three most commonly used platforms.

On ChatGPT, if you're signed in on the web: click your profile icon, go to Settings, select Data Controls, and turn off "Improve the model for everyone." If you're using it without signing in: click the question mark icon in the bottom right, select Settings, and turn off the same toggle.

On Claude: click your profile icon or name, go to Settings, navigate to Privacy, find the toggle labeled "Help improve Claude," and turn it off.

On Perplexity: log in via a browser, click your profile avatar in the top right, go to Settings or Preferences, find the AI data usage section, and toggle off the data retention or AI training option.

That's it. It takes longer to read those instructions than to execute them. The question is why most professionals haven't done it — and the answer is that nobody made the cost feel real until someone like the professor sat onstage watching strangers deliver her work.

What Your Professional Output Is Actually Worth

The privacy conversation around AI tools tends to get framed around sensitive data — client information, financial projections, personal details about employees. That framing is correct and important, but it undersells the risk by focusing on the obvious categories.

The subtler and in some ways more significant exposure is around original professional thinking. The frameworks you've developed over years of working in a specific domain. The way you've learned to diagnose a particular kind of organizational problem. The voice you've built as a communicator that makes people want to read what you write. The specific insight you've developed about your industry that nobody else has quite articulated in the same way.

These things have value precisely because they're scarce. They represent accumulated cognitive labor that is specific to you — to your particular history, your particular relationships, your particular pattern of successes and failures. They are your professional IP, even if you've never thought of them that way.

When you feed them into an AI tool without thinking about the data settings, you're giving that IP away. Not dramatically, not all at once, not in a way that will show up as a clean line of theft. But you're contributing your specific thinking to a system that will use it to produce generic thinking for anyone who asks the right question.

In an economy that is actively devaluing generic output — because AI can now produce a passable version of almost anything generic for nearly nothing — original professional thinking is more valuable than it has ever been. The irony is that many professionals are inadvertently diluting their most valuable asset by using the very tools that made that asset valuable in the first place.

Building The Right Habits Around These Tools

The opt-out is the starting point, not the whole answer. There's a broader set of professional hygiene practices worth building, particularly as AI tools become more embedded in daily work.

Know what you're putting in before you put it in. This sounds obvious, but in practice most people treat AI tools the way they treat search engines — typing queries reflexively without pausing to consider what information the query reveals or transfers. A search query is ephemeral. Input to a generative AI tool potentially isn't.

Strategic documents under active development, performance-related information about specific employees, proprietary client data, financial projections, legal matters — these should not enter third-party AI tools without a clear understanding of how the platform handles data. Many organizations now have formal policies on exactly this. Most employees don't know the policies exist. If you're a manager, the right move is to ask your IT or legal team for the current guidance, then communicate it to your team before the situation where it matters arises.

Be especially protective of the materials that represent the most concentrated version of your professional thinking: original research, frameworks you've developed, writing with genuine voice, strategic analysis that reflects years of pattern recognition in your domain. The risk of feeding these into an AI tool isn't that they'll be stolen in any dramatic sense. It's that they'll become part of the substrate from which future generic outputs are built. Your specificity will be averaged out.

What The Professor Got Right

The professor gave her speech anyway. She stood up in front of students who had already heard her words delivered by two other people, and she delivered them again — and it was different. Not because the words were different, but because the relationship was different. Her students knew her. They'd sat in her classes, brought her their anxieties about placement, watched her care about their professional development in ways that a dean delivering a keyword-generated speech could not replicate.

The words were the same. The meaning wasn't.

This is the most important thing to understand about the current moment in professional life: AI can produce your words, your structure, your approximate argument. It cannot produce your relationships, your credibility, your accumulated trust, or the specific knowing that comes from being the particular person you are in the particular context you've built. Those things are yours.

Protect them with the same care you'd bring to anything else you've spent years building. Start with the opt-out settings — two minutes, done. Then develop the habit of pausing before you share anything that represents your real thinking, and asking yourself what you're handing over and to whom.

The professor's speech is still hers. Her students knew it. Just make sure you're the one controlling who else does.