Before You Share Health Data With AI

More people are turning to AI tools like ChatGPT, Claude, and Gemini to help make sense of health information, a lab result that's hard to decode, a symptom pattern worth understanding, a question they didn't get time to ask a provider. Before you paste anything personal into a chat window, it's worth knowing a few things about what happens to it on the other end.

One thing this is not: a HIPAA question. HIPAA governs how healthcare providers and their systems handle your records. When you personally choose to share your own information with an AI tool, that choice isn't covered by patient privacy law one way or the other. What matters instead is simpler and more personal: does this company keep what you tell it, and what do they do with it. That's a decision you get to make for yourself, not a legal protection you're relying on.

Don't have an account yet?

Pick one to start, you can always add another later.

• ChatGPT: go to chatgpt.com, click Sign Up, use your email or a Google/Apple sign-in
• Claude: go to claude.ai, click Sign Up, same options
• Gemini: go to gemini.google.com, sign in with any Google account

All three are free to create and take under two minutes. Once you're in, keep reading, the next part is about what happens after you sign up.

What the free (and even paid) tiers actually do

As of 2026, the default settings across all three major platforms lean toward using your conversations to improve their models, not excluding them.

ChatGPT (OpenAI)

Free tier conversations may be used for training unless you turn that off, and free accounts in the US now show ads. Paying for Plus doesn't change the training default by itself, you still have to opt out manually. Setting to change: Settings → Data Controls → turn off "Improve the model for everyone." For anything you don't want saved to your history at all, use Temporary Chat.

Claude (Anthropic)

Claude built an early reputation as the privacy-forward option, and for a long time that was true by default. That changed with a policy update, consumer conversations (Free, Pro, and Max) may now be used for training unless you actively opt out. Setting to change: Settings → Privacy → turn off "Help improve Claude." Claude also has an Incognito mode for conversations you don't want to appear in your history at all.

Gemini (Google)

Free-tier conversations can be reviewed by humans and used to improve the product. Paying for the personal Google AI Pro tier does not change this, it's treated the same as the free tier. Setting to change: turn off "Gemini Apps Activity."

Across all three, the pattern is the same: paying for a personal subscription buys you a better model, not better privacy. Only business, team, and enterprise tiers exclude your data from training by contract instead of a toggle you have to remember to flip.

Don't want to hunt through settings menus? Ask the AI directly.

Once you're signed in, paste this into a new chat. It works no matter which of the three you picked, since you're asking it to explain its own settings.

I just created an account with you. Before I start using you for anything health related, walk me through two things. First, where do I find the setting that controls whether my conversations are used to train your models, and how do I turn it off. Second, explain your version of a private or temporary chat mode, what it actually does and doesn't protect, and when I'd want to use it instead of a regular saved conversation. Keep it simple, I'm not technical. I'll probably have follow up questions after, so leave room for that.

So which one is best for health data?

Honestly, that's the wrong question. The platform you choose matters less than what you do once you're in it. Whichever tool you're already comfortable with, ChatGPT, Claude, or Gemini, the move that actually protects you is the same: turn off the training setting, and use the temporary or incognito mode for anything you'd rather not have saved anywhere at all.

So why wouldn't you want to share it?

None of this is about avoiding AI. It's about knowing what's actually at stake before you decide.

• Once something you share trains a model, it can't be pulled back out, even if you delete your account later.
• An AI account doesn't come with the kind of confidentiality your relationship with a doctor or therapist has. Companies can and do change their policies, sometimes without much notice. Claude is a real example, it used to exclude your conversations from training by default, then Anthropic changed that default and made staying opted out something you now have to actively choose.
• Genetic data is the sharpest version of this. People who shared their DNA with 23andMe felt exactly the way you might feel about your own health information, fine with it, no big deal. Then the company was breached, went bankrupt, and its genetic database became a sellable asset under new owners, on terms nobody who spit in that tube ever agreed to. GINA, the federal law against genetic discrimination, exists because insurers and employers had already done this to people, used genetic results against them, before it became illegal.
• The riskiest moment usually isn't a calculated decision to share, it's the reflex. You just got a diagnosis or a test result back, and the first thing you do is open a chat and ask AI to explain it. That's the most human response there is, and it's also the exact moment you're least likely to be thinking about who could use that later, an insurer, an employer, someone deciding whether you're a risk worth covering.
• This isn't hypothetical, and it isn't limited to health data either. Property insurers are already using AI on satellite and drone imagery to scan roofs and cancel or deny coverage, sometimes without the homeowner ever knowing a photo was taken or having a way to see what got flagged. Same mechanism, a company using AI to find a reason to say no, already running today, just on a different kind of data than yours.
• Unlike a single conversation, an account is tied to your identity over time. The more you share, the more it adds up into a pattern connected to you specifically, not an anonymous one-off.
• There's an important exception to hold alongside all of this: your relationship with your own healthcare provider. Providers are held to a higher standard under HIPAA than any consumer AI account will ever be, you can ask exactly how your data is being used, and you can opt out. But you can also opt in, and that's not automatically a risk. An AI scribe handling notes during your visit can mean your provider spends more time actually talking with you and less time typing. The goal was never caution for its own sake, it's understanding the benefit and the risk well enough to navigate a technology that's only going to keep expanding, not disappear.

What this means: knowing which relationship you're actually in when you hit send, one bound to protect you, or one where you're the one setting the terms, and sharing accordingly.

Before you share health data, do this

• Pick your platform and find the training toggle (see above)
• Turn it off
• For anything especially sensitive, a diagnosis, a family history detail, a mental health concern, use temporary or incognito mode instead of a regular saved chat
• Remember that turning a setting off going forward doesn't erase conversations you've already had

Information current as of August 2026. AI companies change their privacy settings and defaults often, sometimes without much notice. Check your own account settings before relying on anything above.