How to Stop AI From Training on Your Conversations: A Beginner’s Guide to ChatGPT, Claude, and Gemini Privacy Settings

Three weeks ago a reader emailed me after finishing our risks guide with a question I couldn’t answer off the top of my head: “Okay, but how do I actually turn this off?” She’d pasted a client contract into ChatGPT to get a plain-English summary, then panicked, because somewhere in the back of her mind she remembered reading that AI companies “use your data.” She didn’t know which setting controlled that, whether it was already too late, or if the same problem existed in Claude and Gemini too.

I didn’t have a clean answer for her, so I went and found one. This is that answer, expanded into the guide I wish existed before she had to ask.

To be clear upfront: none of this is about being paranoid. Using ChatGPT, Claude, or Gemini is fine, and for most everyday tasks the privacy stakes are low. But there’s a real difference between “I don’t care if this becomes training data” and “I didn’t know I had a choice.” This guide exists to make sure it’s a choice, not a default you never looked at.

Why This Actually Matters

Here’s the plain version, without the scare tactics. When you chat with an AI tool, that conversation can end up in one of two places: it either just answers your question and gets discarded (aside from basic logs), or it gets folded into the pool of data used to train the next version of the model. The second one means a human reviewer could, in some circumstances, read what you wrote, and fragments of your conversation could theoretically influence how the model responds to other people down the line, in the same way it learned from books and articles.

For a question like “what’s a good side dish for salmon,” that distinction barely matters. For a therapy-adjacent conversation, a client’s financial details, an unpublished manuscript, or your company’s internal roadmap, it matters a lot. The problem, as I found out helping that reader, is that every major AI tool handles this differently, defaults to a different setting, and buries the toggle in a slightly different menu. So let’s go through each one, one at a time, using the actual current menu paths rather than vague advice to “check your settings.”

If you haven’t yet read What Is AI, Really?, it’s worth doing first — this guide assumes you already know roughly how these tools work, and builds directly on it.

The One Concept You Need First

Every platform below has some version of the same idea: a toggle that controls whether new conversations can be used to train future models. Understanding two things about that toggle will save you a lot of confusion:

1. Turning it off is not retroactive. If training was on for the last six months and you switch it off today, that history doesn’t get pulled back out of any training run that has already started. The toggle only protects conversations that happen after you flip it.
2. Turning it off doesn’t mean “nobody ever sees this.” Every company still keeps a copy of your chats for a limited time for security, abuse prevention, and legal reasons, even with training disabled. What changes is whether that data becomes part of the model’s permanent knowledge.

Keep those two facts in mind and the rest of this is just mechanics.

How to Turn Off Training in ChatGPT

ChatGPT’s data-sharing setting is called “Improve the model for everyone,” and on Free, Plus, and Pro personal accounts, it’s on by default. That means unless you’ve already gone in and changed it, your conversations are currently eligible to be used for training.

Steps (web and mobile):

1. Click your profile icon (top-left or top-right, depending on your device).
2. Select Settings.
3. Go to Data Controls.
4. Find “Improve the model for everyone” and toggle it off.

A few details worth knowing:

• Turning this off also stops saving that conversation to your visible chat history, so the two are linked, not separate switches.
• If you want to keep your chat history and stop training, OpenAI’s help documentation confirms there’s a way to do both — the option lives in the same Data Controls menu, so check the sub-settings rather than assuming it’s all-or-nothing.
• Temporary Chat is your fastest option for a one-off sensitive conversation: it doesn’t save to history and isn’t used for training, regardless of your main setting. Look for it in the menu at the top of a new chat.
• One easy trap: even with training turned off, tapping the thumbs-up or thumbs-down button on a specific response can flag that individual exchange for review. If you don’t want a conversation looked at, don’t rate it.
• ChatGPT Business, Enterprise, and Edu accounts don’t train on your data by default — this mainly affects personal Free, Plus, and Pro accounts.

How to Turn Off Training in Claude

Claude’s setting is called “Help improve Claude,” found under Settings > Privacy. As of Anthropic’s 2025 policy update, every consumer account (Free, Pro, and Max) has to actively choose a setting rather than being defaulted into one silently — but it’s easy to click through that choice quickly during signup without registering what you agreed to, so it’s worth double-checking even if you think you already decided.

Steps (web and mobile):

1. Sign in and click your profile name (bottom-left on desktop).
2. Select Settings, then Privacy.
3. Find “Help improve Claude” and confirm it’s toggled off.

A few extra controls worth knowing about while you’re in there:

• Incognito chat — start a new conversation and tap the ghost icon to open a chat that isn’t used for training and has a shorter retention window, similar to ChatGPT’s Temporary Chat.
• Location metadata — Claude can use your IP address to estimate your general location for things like local search results. There’s a separate toggle for this in the same Privacy menu if you’d rather it didn’t.
• Memory — under Settings > Capabilities, you can turn off “Generate memory from chat history” if you don’t want Claude recalling details from past conversations.
• If your data was already opted in before you changed the setting, Anthropic’s documentation is explicit that new conversations going forward are protected, but anything already inside a training run can’t be pulled back out.

If you’re weighing Claude against the other two tools for reasons beyond privacy — speed, writing quality, document handling — our Claude vs ChatGPT vs Gemini comparison covers the practical differences.

How to Turn Off Training in Gemini

Gemini’s version of this setting lives inside Gemini Apps Activity, sometimes labeled “Keep Activity.” Because Gemini is tied to your broader Google account, this one is a little more spread out than the other two.

Steps (web and mobile):

1. Go to gemini.google.com and open the Activity panel from the side menu (or visit myactivity.google.com/product/gemini directly).
2. Toggle Keep Activity to off.
3. From the same screen, you can also delete past activity individually or set an auto-delete period (3, 18, or 36 months) so old conversations clear themselves out automatically going forward.

A few Gemini-specific quirks to know:

• Turning Keep Activity off stops your chats from being used to improve Gemini and stops them from appearing in your saved history — but Google is explicit that it still processes your chats to respond to you and for safety/abuse monitoring, regardless of this setting.
• If you’re a Google Workspace user through school or work, your account administrator controls these settings, and you may not be able to change them yourself.
• Gemini also has a “Your past chats with Gemini” personalization setting under Personal Context in Settings — separate from Keep Activity — which controls whether Gemini references earlier conversations to personalize new answers. Worth turning off if you’d rather each conversation start fresh.
• Like the other two, Gemini offers Temporary Chats for one-off conversations you don’t want saved or factored into personalization at all.

A Side-by-Side Comparison

What These Settings Don’t Do

This is the part most guides skip, and it’s the part that actually matters if you’re making a real decision instead of just chasing a checklist.

• They don’t erase what’s already been used. If training was on for a while before you turned it off, that data may already be baked into a model that’s already been trained. There’s no “undo” for that.
• They don’t stop the company from storing your chats at all. Every platform keeps a copy for a limited window (commonly 30 days, though Gemini’s policy differs) for safety, fraud prevention, and legal compliance, independent of the training toggle.
• They don’t prevent a human from ever reviewing a specific conversation. All three companies reserve the right to review individual conversations if they’re flagged for policy violations, regardless of your training setting.
• They’re account-wide, but device settings can still leak data separately. Location permissions, connected apps, and browser extensions are controlled outside these menus and are worth a separate check.

None of this is a reason to panic. It’s the reason this guide exists instead of a single tweet saying “just flip the toggle.”

A Simple Rule for What Never to Paste In

Settings help, but the more reliable habit is deciding upfront what simply doesn’t belong in a chat window, privacy toggle or not. A good rule of thumb, borrowed from our risks guide: if you wouldn’t be comfortable with it appearing in a public forum post under your name, don’t paste it into any AI tool, even with training turned off. That covers passwords, full account numbers, medical record numbers, unpublished legal documents, and anything a client has explicitly asked you to keep confidential. For sensitive tasks like that, use Temporary or Incognito chat regardless of your main settings — it’s the closest thing to a safety net these tools offer.

A 10-Minute Privacy Checkup

If you use more than one of these tools, here’s the order I’d actually do this in:

1. Pick your primary tool first. Whichever one you use daily gets checked first — it has the most accumulated history.
2. Turn off the training toggle using the steps above.
3. Skim your existing chat history for anything genuinely sensitive and delete those specific conversations.
4. Set an auto-delete window if the tool offers one (Gemini makes this easiest), so you’re not relying on memory to clean up later.
5. Repeat for your second and third tools. Most people use at least two of these regularly, and the settings don’t carry over between them.
6. Put a reminder on your calendar for six months out. Privacy settings menus change more often than people expect, and a setting you turned off can occasionally reset after a major policy update, as happened with Claude’s 2025 change.

Frequently Asked Questions

Does turning off training make ChatGPT, Claude, or Gemini worse at answering me?
No. These settings only affect whether your conversations are used to train future versions of the model. They have no effect on how the current model responds to you.

Is it too late if I’ve already been using one of these tools for months with training on?
Not too late to protect what comes next. Turning the setting off going forward protects new conversations. Data already absorbed into a completed training run generally can’t be extracted, but that’s a reason to change the setting today rather than a reason not to bother.

Do paid plans (ChatGPT Plus, Claude Pro, Gemini Advanced) have better privacy by default?
Not necessarily. In all three cases, the training setting is something you choose yourself and applies the same way across Free and paid personal tiers. Business, Enterprise, and Education accounts are the ones with different, usually stricter, defaults.

Which of the three is the most private by default?
As of this writing, Claude requires users to actively choose a setting during signup rather than silently defaulting them into training, which is a meaningfully different approach from ChatGPT and Gemini, both of which start with data sharing on unless you turn it off. That said, “by default” isn’t the same as “forever” — check your own account rather than assuming based on general reputation.

What’s the fastest single thing I can do if I don’t want to read the whole guide?
Use each tool’s one-off private mode (Temporary Chat in ChatGPT and Gemini, Incognito chat in Claude) for anything sensitive, starting today. It’s less thorough than fixing your main settings, but it’s a 5-second habit that immediately reduces risk.

Final Thoughts

None of this requires you to distrust AI tools or stop using them. I use all three of these platforms weekly, and I’m not deleting my accounts over any of this. What changed, once I actually walked through each menu, is that I stopped assuming the defaults were doing what I wanted them to do. They usually weren’t.

Give yourself the ten minutes this week. Turn off the toggle in whichever tool you use most, glance through your recent history for anything you’d rather not have sitting there, and get in the habit of using the “temporary” or “incognito” mode for anything genuinely sensitive. That’s the whole system — no ongoing maintenance required beyond a check-in every few months.

If you’re still building your foundation with these tools more broadly, How to Write the Perfect AI Prompt and The Real Risks of Using AI are the two guides that pair best with this one.