Think back to the first time you opened ChatGPT or Claude. Before you'd typed a single word, the app had already made a few privacy choices on your behalf.
Those choices cover what it remembers about you, how long it keeps your conversations, and whether it learns from them. Most of us never revisit them, and that's rarely because we've weighed the options and agreed. The settings are simply tucked away where you'd only find them if you went looking. ChatGPT added a Privacy Center in late September that explains its controls, which is a good sign that this is worth your attention. It's a map, though, not a switchboard. The switches themselves still live in Settings, and that's where this tour takes you. This post covers the three privacy settings that matter most, in ChatGPT and Claude, the two AI apps you're most likely to use. It takes about twenty minutes, and afterwards your AI setup reflects what you decided instead of what the company decided for you.
The trade-off nobody mentions
Every privacy setting costs you some convenience. People who work in security and privacy know this, and app designers count on you not thinking about it. The more an AI remembers about you, the more useful it gets, and the more of your life is stored on the company's servers.
Memory is where this trade-off is easiest to see, partly because ChatGPT and Claude have made opposite choices about it. Say you're helping your father manage his care. With memory on, ChatGPT can remember his medications, his diagnosis, and the name of his specialist, so next week you can simply ask "what questions should I bring to Dad's appointment?" and get a specific, useful answer. Claude, by default, won't keep those health details at all. It has chosen privacy over convenience on your behalf, and if you'd like it to remember, you switch on a setting called "Include sensitive topics in memory." Neither company is wrong. They've drawn the line in different places, and you get to draw your own: remembering saves you re-explaining the whole situation every time, while forgetting means his medical history isn't stored in your account.
Here's my own confession. I've spent my career in security and fraud prevention, and my AI privacy settings are fairly relaxed. I value the convenience, and I value how much more useful these tools become when they know me. I'm comfortable with that because I've thought about what I'm actually trading. Privacy settings decide what the company does with what I share: whether it remembers, how long it keeps it, and whether it learns from it. Security is about keeping everyone else out. The two are connected, because the more an app remembers about me, the more valuable my account becomes to someone who shouldn't be in it. So I protect the account itself: two-factor authentication on everything, long and unique passwords, firewalls on my devices and home network, and a clear plan for what to do, and how quickly, if something goes wrong. With that in place, the trade left over is one I'm happy to make. The company holds more of my information, and in return the tools work better for me.
So neither choice is wrong. The right one for you depends on how comfortable you are with what the company keeps, and on how well your account is protected. If your security basics aren't in place yet, meaning two-factor authentication and strong, unique passwords, start there. They protect everything stored in your account, whichever way you set the toggles.
The three settings that matter
A quick note before we start: the steps in this post follow the ChatGPT and Claude phone apps, as of late September 2026. On a computer, tablet or desktop app, the menus may be arranged a little differently, and both companies rearrange them often. If something isn't where I've described, look for the same setting name nearby.
"Privacy settings" actually covers three separate controls. It helps to look at them one at a time, because each one answers a different question:
1. Training: does your input teach future models? When you chat with an AI, the company may use your conversations to improve its models.
- In ChatGPT, the toggle is called "Improve the model for everyone" in Settings, under Data controls (on some versions it's grouped under Security). It's on by default for personal plans.
- In Claude, it's "Help improve our AI models" under Settings → Privacy (on some versions it's labelled "Help improve Claude"). Anthropic asked existing users to make a choice in 2025, and plenty of people clicked through that pop-up with the toggle left on. Check yours.
In both apps, turning training off applies to new conversations. More on that in the honest boundary section.
2. Memory: does the app remember you across conversations? This is separate from training. Both apps can now carry details about you from one chat to the next: your work, the projects you're juggling, how you like answers written, the people you mention often. With memory on, you spend less time setting the scene, and the answers fit your life rather than someone else's. You might turn it off, or prune it, if you share a device, if you'd rather certain subjects didn't follow you from chat to chat, or if you notice it carrying an outdated detail (an old job, a finished project) into new conversations.
- In ChatGPT, go to Settings → Personalization → Memory. What you see there depends on which version of memory your account has. Some accounts show two switches, Reference saved memories (specific facts it has stored about you) and Reference chat history (drawing on your past conversations), plus a Manage button that lists the saved facts. Others show a single Enable Memory switch and a written summary of what ChatGPT knows about you.
- In Claude, go to Settings → Capabilities → scroll down to Memory. Since late August, Claude lists what it remembers as Topics you can open, edit, or delete. It leaves out health and other sensitive subjects unless you turn on "Include sensitive topics in memory," and some things, like government ID and financial account numbers, it never saves.
3. Retention: how long is your data kept? Even with training off, companies keep data for a while for safety and legal reasons.
- ChatGPT keeps your regular chats until you delete them. Temporary Chats are deleted from OpenAI's systems after 30 days.
- Claude keeps data for 30 days after you delete it if training is off. If training is on, it can keep it for up to five years.
These windows are the fine print. They're worth knowing about, but not worth losing sleep over.
Both apps also have an "off the record" mode. ChatGPT calls it Temporary Chat and Claude calls it Incognito. The conversation doesn't show up in your history, doesn't create memories, and isn't used for training. (In ChatGPT, you can now choose at the start whether a temporary chat draws on what it already knows about you. Non-personalized, the default, gives you a clean slate.) Use it when you want to ask something without it following you around.
Three moments when these settings matter
1. Everyday drafting in ChatGPT. You paste in a draft email or a personal letter. Nothing scandalous, just nothing you want feeding a tech company's training data. That's the case for turning off "Improve the model for everyone." (If the draft includes your employer's client or internal information, a toggle isn't enough. Use whatever AI tools your workplace has approved and in their environment.)
2. Health questions in Claude. You ask about a medication, a symptom, or a test result. Health questions are among the most revealing things people type into AI, because together they sketch your medical history. Claude's memory skips health topics by default, but the conversation itself still sits in your history. Use Incognito for these, or at minimum make sure "Help improve our AI models" is off, so the questions and data aren't kept.
3. The shared device. The family iPad, or the laptop your grandkids borrow. Anyone who opens the app can see your history. Temporary Chat and Incognito help here: you can ask a sensitive question and leave nothing in the sidebar.
The honest boundary
Opting out only works going forward. If you remember one thing from this section, make it this one. Turning off training in either app stops future conversations from being used. It doesn't pull back conversations that were already used, and both companies' help pages make this clear. Flip the toggle today, but know that it can't undo yesterday.
Feedback is a separate door. If you tap thumbs-up or thumbs-down on a response, you're sending the whole conversation along with your rating. OpenAI and Anthropic both say feedback conversations may be used for training. Anthropic keeps them for up to five years. If you're being careful, be careful with the feedback buttons too.
Deleting a chat doesn't always delete the memory. In ChatGPT, a memory saved from a conversation stays behind even after you delete that conversation. Check your saved memories separately.
Deletion takes time. Temporary and Incognito chats are deleted after 30 days, not the moment you close them. These are standard practices, and the companies document them openly. Still, "delete" rarely means "gone this second," and it's better to know that going in.
Don't type what you wouldn't want read back. This rule holds no matter how you set things up. Account numbers, precise health details, anything you'd shred if it were on paper: keep it out of the chat box entirely. Settings reduce your exposure, but they can't remove it entirely.
And the standing boundary for this whole publication: this is information, not advice. My aim is to show you where each setting lives and what it does, so that however you set them, the choice is an informed one and it's yours.
Try this in ten minutes
ChatGPT: turn off training (2 minutes)
- Open the sidebar (tap the menu icon at the top left), then tap the gear icon to open Settings.
- Tap Data controls. On some versions it sits under a heading such as Security. If you can't find it, ChatGPT's new Privacy Center can point you to it.
- Decide if you want to turn off "Improve the model for everyone." You only need to do this once. When you're signed in, it applies to your account on every device.
Claude: check your training setting (2 minutes)
- Tap the menu icon at the top left, then tap your name at the bottom to open Settings, and choose Privacy.
- Look at "Help improve our AI models" (on some versions, "Help improve Claude"). If it's on and you didn't mean to turn it on, switch it off. It takes effect immediately and applies to your account everywhere.
Review what each app remembers (3 minutes)
- In ChatGPT: open the sidebar, tap your name, then tap Memory. Depending on your account, you'll see a list of saved memories under Manage, or a memory summary. Read through it and delete anything you'd rather it forgot. You can also just type "please forget that..." in a chat.
- In Claude: open Settings, tap Capabilities, then scroll down to Memory, and look through this section. Open any topic to edit or delete it, and check whether "Include sensitive topics in memory" is set the way you want.
Try off-the-record mode (1 minute each)
- ChatGPT: make sure you're in Chat mode (not Work), start a new chat, and tap the icon at the top right that looks like a speech bubble drawn with a dashed outline. If you're asked, choose Unpersonalized for a clean slate, and don't save the chat if you want it to stay temporary.
- Claude: start a new chat and tap the ghost icon at the top of the screen.
- Ask something. Notice that nothing appears in your sidebar.
The two-minute audit (do this one today)
Open ChatGPT and Claude and ask yourself one question about each: is there anything in my history here I'd be uncomfortable with a stranger reading? If yes, delete those conversations (you can even tell ChatGPT or Claude to find them and delete them), then flip the toggles above. If you no longer need a conversation, there's no reason to keep it.
The bigger lesson
Here's what I want you to take from the tour: checking defaults is a habit, not a task. Companies update these apps constantly, and settings drift. A redesign moves a toggle. A new feature arrives already switched on. Put a recurring reminder in your calendar twice a year: "AI privacy checkup, 20 minutes." Open both settings pages, confirm your choices are still your choices, and delete what has piled up.
You already do this with smoke detectors and credit card statements. Add the AI apps to the rotation. The companies set the defaults, and the calendar reminder makes sure you keep choosing for yourself.
Consider it decoded.
Written by Amanda · Midlife TechCurious
Want these in your inbox? CLICK HERE for AI decoded for midlife women, straight to you.