Where I Draw the Line on AI and Privacy

TL;DR: I don’t mind targeted ads, and my Instagram feed works for me, so I don’t treat privacy as an on/off switch. I sort what I share with AI into three tiers based on how bad the worst case would be, keep the stuff I can’t undo out of my vault, and use a few basic safeguards. Here’s where I draw my line and how you can draw yours.

Confession: I love my Instagram feed

Let me start with a hot take that probably puts me in the minority: I don’t mind targeted ads. Whether Meta is listening to my conversations or just reading the trail I leave online, I’m not upset about it. I get it, I’m paying them for their services with my data.

Here is what my Instagram feed looks like: family friendly travel ideas for Taiwan, Singapore and Hong Kong, delicious food, new restaurants to try, kids extracurricular ads and travel hacking tips. That is exactly what I want to see. My experiences with ads have been overall positive, and I’d rather get a feed full of things I care about than a random one.

Sometimes I feel like there are expected norms in society where we’re all supposed to have the same reactions to money (“I want more!”), math (“yuck, who wants to do that?”), and advertising (“ugh, they suck!”). I see myself as more of a realist who doesn’t enjoy complaining (see my personality results), so I enjoy being a contrarian when it makes sense to be.

I understand the effects of advertising are real. Subliminal by Leonard Mlodinow was a super interesting book. I was very impressed by his research on how much of our behavior is shaped below our awareness, and I believe it. But I also still believe I’m always in control of myself and my brain.

Maybe that’s just how I’m wired, maybe I’m an outlier, and maybe I’m a little naive. I operate with a strong regret minimization model (I share more in this post). This is just what works for me.

How I think about AI and privacy

While I consider myself as someone who’s not very into conspiracy theories, I admittedly do not read the full terms & conditions, so I can’t tell you what Claude does with my data. I just assume they are not actively selling it.

Instead of asking myself if something is private before typing it into AI, this is generally how I think of it:

If this got out, what’s the worst that could happen, and could I live with it?

Ads clear that bar easily. The worst case is that someone knows I’m planning a trip to Hong Kong, and the cost is seeing more Hong Kong travel content. I can live with that.

This is also what I took away from my personality test experiment. Gemini described me closely from only four sets of results and a few sentences. That didn’t scare me, but it did make me want to choose what goes where on purpose and I try to be as clear as possible when I chat with AI to make sure the results are helpful for me.

My three tiers

I sort everything by the worst case, not by how personal it feels. Here is how it breaks down for me:

Tier	What goes here	Examples	Worst case if it got out
1. Share freely	Things where the worst case is an ad	Travel plans, food I like, my kids’ activity interests, general questions	I see more content about it
2. Share deliberately	Things I only give to a tool I’ve chosen to trust	My personality results, my goals, how I make decisions, background about my family	Someone could model how I think, which is a real cost but one I can live with
3. Never share	Things I can’t undo	Passwords and keys, my SSN, sensitive details about my kids	Someone could get into my accounts or my identity

The tiers aren’t about secrecy. They’re about how hard it is to recover. I can recover from a stranger knowing I like egg tarts. I can’t easily recover from someone holding the keys to my accounts.

What stays out of my vault and AI chats

My vault and my AI chats are where I’m most deliberate, because that’s where I put the most context about my life. Three things stay out completely:

• Passwords and keys. Logins, API keys and recovery codes never go in a note or a chat.
• My Social Security number. I don’t give it to AI. That’s just a no brainer.
• Sensitive details about my kids. Everyday family context is fine because it helps AI give better answers. Anything sensitive stays out.

Here is the part where my line may differ from yours: I’m okay sharing my credit cards (name and last four digits) and passport numbers when I have a good reason to. AI helps me keep track of my credit cards so I know what to apply for next and when. I log my passport number in my Obsidian vault because it’s helpful to have it handy in an organized note.

I’m not saying you should copy that. It’s my own call, and it’s a good example of why I’d rather you draw your own line than borrow mine.

My safeguards

Being relaxed about ads works for me because I follow standard safety protocol for the things that matter. None of it is fancy:

• I monitor my credit reports. You can get your reports for free at annualcreditreport.com. Double check the URL, because lookalike sites exist.
• I use a password manager and two-factor authentication. Unique passwords everywhere, and a second step on the accounts that matter most.
• I check the privacy settings of my AI tools. I look at what a tool does with my chats, whether it has memory, and who or what can read the folder my vault lives in.

That last one takes about ten minutes per tool, and it’s the habit I’d start with. It’s also why I’m comfortable giving AI so much context in tier 2: I decided where the data goes before I shared it.

Where I might be wrong

I might be naive in thinking it’s not a big deal for people to know things about me for advertising purposes. I’m aware of that, and I’d rather say it out loud than pretend I have it all figured out.

The strongest argument against my view is that data doesn’t always stay where you put it. Information collected for ads can be combined with other data, sold or exposed in a breach. That is exactly why I keep tier 3 locked down and keep an eye on my credit reports. I’m not ignoring the risk, I’m putting my caution where the damage would be hardest to undo.

It’s also possible that my personality makes me an outlier. Some people feel real discomfort at the idea of being profiled, and that discomfort is a legitimate input. If that’s you, your tier 1 should probably be smaller than mine.

I’d describe my approach as a healthy balance of paranoia and freedom. AI amplifies whoever is using it, so I want enough caution to protect myself and enough freedom to actually benefit from what these tools can do with my context.

Draw your own line

You don’t need my tiers. You need your own answer to the worst-case question, written down once so you aren’t deciding on the fly. Here is a short way to do it:

• List the kinds of information you share with AI, such as travel, work, family, health, money and logins
• For each one, write the worst case if it got out
• Sort them into three groups: fine to share freely, share deliberately, never share
• Check the privacy settings of the AI tools you actually use
• Put the “never share” list somewhere you’ll see it, like the top of your notes

If you’d like AI to help, paste this prompt into a chat. It works best if you answer honestly instead of what sounds responsible:

I want to decide what personal information I'm comfortable sharing with AI tools. Interview me one question at a time about the kinds of information I share, such as travel, work, family, health, money and logins. For each one, ask me what the worst case would be if it got out and whether I could live with it. After about 10 questions, sort my answers into three groups: fine to share freely, share deliberately with a tool I trust, and never share. Only use what I told you, and flag anything you're unsure about.

There’s no right answer here. The goal is to make the decision on purpose, which is more than most of us do by default.

---

Frequently asked questions