Hands-On Tech 285 Transcript
Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.
Mikah Sargent [00:00:00]:
Coming up on Hands-On Tech, is your AI chatbot keeping a diary on you? Stay tuned. Hello and welcome to Hands-On Tech. I am Mikah Sargent, and today we are answering your tech questions. Yes, we have a great tech question that has come in today from Wendell. I remind you all that if you have tech questions you would like answered, email me. hot@twit.tv is how you get in touch. But let's take a look at Wendell's question.
Mikah Sargent [00:00:40]:
All right. Wendell writes, hey Mikah, my wife has started talking to ChatGPT like it's a combination doctor, lawyer, and best friend. Last week she typed in her actual lab results and asked it what they meant. I told her all of that is going somewhere, and she says it's no different from a Google search and I'm being paranoid. So Who's right? When we type something into one of these AI chatbots, where does it go? Who can see it? And is it being used to train the thing? And is there a setting that turns any of that off, or is that wishful thinking? I use Gemini on my Android phone and she's on ChatGPT on her iPhone. So bonus points if you cover more than one. Wendell from Eau Claire, Wisconsin. First of all, Wendell, a split household.
Mikah Sargent [00:01:27]:
Ooh, very exciting. One of you on Android, one of you on iPhone. Very good question, and it's one that I think people don't— aren't taking the time to ask and aren't taking the time to think about. So I'm glad you are, and I'm glad we get to talk about it here on the show. Let's start with the short answer. Frankly, Wendell, you do win this one. A chatbot conversation, it's not a search. It's not the same thing as going to Google and doing a search.
Mikah Sargent [00:01:57]:
It's stored On the company's servers. It's tied to the account. It is open to human reviewers, and by default, it is used to train the model. So check, check, check, check, check, check on everything. The off switches, they are real. There's this— that's not wishful thinking. So Wendell, also accurate there, but they turn off the training of the models. They don't turn off the storing of the data.
Mikah Sargent [00:02:28]:
So if you're wondering where is it that the words go when you type it in instead of a Google search, every prompt, every upload, and even every reply is actually saved to the account on the company's servers. And so that is how you have chat history that follows you from your phone to your laptop. That's the only way to gain access to those in multiple places. And the default for personal accounts on both ChatGPT and Gemini is that conversations are used to train and improve the models. So you will be training the models if you decide to use these with the default permissions. So I was looking through both OpenAI and Google's support documents, and we'll include links to those in the show notes. But it's also important to understand that there are some cases where this is different. Work accounts, school accounts, and enterprise accounts do not have training on by default.
Mikah Sargent [00:03:26]:
And so those you have to opt into instead of opt out of them. And it's not just typed text. So photos, PDFs, and voice, especially voice, all count. And Gemini does also log info from the device that you're using and the apps that have connections as well. Interestingly, when it comes to OpenAI, it is a simple matter of going in and turning off the training of the models, and we'll talk about that in a moment. With Gemini, if you turn off model training, then you don't get access to your past chats at all. And so that makes Gemini a lot less handy to use when Each new chat is sort of bespoke. So who is able to see all of this information? Well, human beings.
Mikah Sargent [00:04:23]:
And this is the part that most people don't know about. Google, of course, says it outright in its privacy hub. Don't enter confidential information that you wouldn't want a reviewer to see. Gemini chats picked for review, they will get disconnected from the account and then kept for up to 3 years. Oh, and by the way, deleting activity later does not pull those back. OpenAI says that, quote, a limited number of authorized OpenAI personnel plus trusted service providers can access chats when it comes to abuse investigations, support, legal matters, and model improvement. Model improvement! That's sort of vague, and they need it to be that way. Oh, also courts, of course.
Mikah Sargent [00:05:06]:
There's no doctor-patient confidentiality, no attorney-client privilege that covers a chatbot conversation. In fact, Sam Altman said so in July 2025 and then called the situation very screwed up, meaning that he felt that it should be that there was legal protections there. TechCrunch had a whole document on that as well, which I will include in the show notes. And then of course, let's look to the real-world proof. In the New York Times lawsuit, a judge ordered OpenAI to preserve chats users had already deleted. The obligation ended September 26th, 2025. OpenAI says it's back to normal deletion, but data from April through September of 2025? Yeah, that's still sitting there under a legal hold. It's all there and accessible.
Mikah Sargent [00:05:55]:
What about advertisers? Well, yes, sort of. ChatGPT's free plan and Go plan do show ads. Plus and Pro don't. OpenAI says that advertisers never get the conversations and that ads aren't eligible to run next to its tools, but also just health chats in general, mental health chats in general, or political chats in general. But ad personalization, yeah, it can draw on chat history and memories unless you go in and turn that off. So when we talk about whether it's no different from a Google search, the truth is it is different from a Google search. But That's the thing is that there's a grain of truth in there. Google also saves searches to an account if the account is logged in.
Mikah Sargent [00:06:47]:
So frankly, there's no sort of clean situation here. All of it can be logging data. A search though is a handful of keywords, right? You're typing in, in theory, just keywords, not typing in full questions. But a chat, that's the entire lab report that you've sent in, the follow-up questions, the worry, the anxiety, the personal information. All of that is there. And the chatbot remembers. That's part of the whole point. ChatGPT has saved memories, plus has saved memories.
Mikah Sargent [00:07:18]:
That's what they're called. Plus there's also a setting that references past chats. And when you delete a chat, it doesn't necessarily delete the memory that's created from it. So memories can stick around even after chats have been deleted. I've had that happen before where I was one point talking about a newer Subaru that I was driving as a loaner device while my Subaru was in the shop. And when I was asking one of the chatbots about something to do with my Subaru, it said, you know, this, oh, your 2024 Subaru. And I thought, no, very much not that. Even though I had archived the past chats where that had happened, the memory was still there.
Mikah Sargent [00:07:56]:
And that can go even further. And also, HIPAA doesn't follow the lab— like, I need to stress this. HIPAA does not follow the lab results into the chat box. HIPAA binds doctors, it binds hospitals, it binds insurers, it binds their business partners. It does not bind an app that you choose to type into. That's not how that works. As far as those lab results, OpenAI did create a special separate space specifically for this, ChatGPT Health. That was announced on January 7th, 2026.
Mikah Sargent [00:08:32]:
So earlier this year, and then opened to US adults on, on the web and on iOS on July 23rd, 2026. In there, in there specifically, health conversations are not used to train the foundation models. Health gets its own memory, and that memory does not flow back into your regular chats. It also gets extra encryption. But say it with me, still not HIPAA. The protection, that's a company promise. It's not a federal law. If you are using a regular chat instead, well, let me give you some advice.
Mikah Sargent [00:09:10]:
If you decide not to use the health tool, if you are going to put your health data into these systems, crop out the name or mark out the name, remove the name entirely, not just mark out the date of birth. Get that out of there. The medical record number, please get that out of there. The address, Sure, get that out of there too, because the AI is using the numbers, but it's capturing everything. It doesn't need your identity to figure that out. And then use that explanation as a means to build better questions to ask your real live living human doctor. Don't let it replace the doctor. Let's talk about these settings.
Mikah Sargent [00:09:53]:
Uh, for the iPhone and for the web, in order to turn off training, You go into Settings, Data Controls. There's an option that says Improve the Model for Everyone. You turn that off. In it, chats will still show up in history. They're just no longer used for training, and this will sync across every device on the account. So when you're logged in, if you turn this off in one place, it's the same across the board. And the great thing is, yes, your chats still remain. unlike Gemini.
Mikah Sargent [00:10:26]:
There's also temporary chat. If you use the temporary chat feature, that stays out of history. It doesn't add to your memories. So that problem that we talked about earlier is also not involved, and it's not used for training. OpenAI still keeps a copy for up to 30 days for safety, but then it's gone. In memory for ChatGPT, go to Settings, Personalization, and Memory. You can see what ChatGPT has saved about you and make adjustments there. And then in ads on the free and go plan, if you're trying to make adjustments to those, especially the personalization feature, well, go to settings and then ad controls.
Mikah Sargent [00:11:05]:
You should not see those on a paid account that's past go. So free and go will have those, but, uh, up from there, no. And then deleted chats, if you just deleted chat across the board, those leave OpenAI systems within 30 I've got links to OpenAI's data controls FAQ and its temporary ChatGPT support as well. As for Gemini settings, things are a little bit different. In the Gemini app, you go to menu, you tap on your profile picture, you go to Gemini apps activity, and then you choose turn off. And you may want to choose turn off and delete activity. This is what's frustrating. In order to stop the model from training, you have to stop the app activity, which means, again, that every chat essentially becomes a temporary chat.
Mikah Sargent [00:12:02]:
So not as helpful from that point on. And part of the reason why I don't use Gemini as much as I used to, because I have model training turned off on my different AI chatbots. Keep activity, that's on by default if you're 18 years or older. It also has an auto-delete feature, and that defaults to 18 months. You can change that to as little as 3 months. You can change that to 36 months, or you can set it to never. And even if you have Keep Activity off, it's important to understand that this— to be clear, Keep Activity is separate from Gemini Apps Activity. So Keep Activity is a separate store.
Mikah Sargent [00:12:41]:
Even with Keep Activity off, chats will stay with the account for up to 72 hours. And that is so that it can kind of do the final processing and get everything as it needs to be. You can use a temporary chat in Gemini as well. So if you're going to use Gemini and you're using a more specific and kind of revealing set of chats, that is where a temporary chat might work. It's not used for training or human review. You get the same 72-hour window. But of course, the catch is that tapping thumbs up or thumbs down on one of those messages will send that conversation in for review regardless of the setting. So even if you have all this stuff turned off, if you give the, if you do a review on any of those, it will send that information in along with everything else in the chat or much of what else is in the chat.
Mikah Sargent [00:13:37]:
I've also got a link there in the show notes that talks about managing and deleting Gemini apps activity. So let's get to the bottom line here. Verdict. Dun dun. Wendell's right. A chatbot is less like a search box and far more like a diary that the company holds onto, not you. It's under their bed, not yours. Uh, still, you know, I, I wouldn't say that suddenly your, uh, wife is wrong to use it.
Mikah Sargent [00:14:06]:
You just need to use the right space. The space that's dedicated to it. And also be aware of how you're using it. ChatGPT Health is a place for those medical questions. If you're going to ask a chatbot medical questions and you're going to upload your personal health data, then ChatGPT Health is a great place for those medical questions. Temporary Chat is a place for anything that's sensitive. So if you're kind of looking for the quick fix, Well, here's what you do. It'll take you a couple of minutes.
Mikah Sargent [00:14:39]:
In ChatGPT, turn off improve the model for everyone, and in Gemini, set it to keep activity off. Turning off training though is not turning off storage. Don't forget that. And nothing, nothing typed into the chatbot is privileged or HIPAA protected. Protected. If you bear all that in mind, then I think you're going to be much better going forward. And Wendell, I want to thank you so much for taking the time to join me today on the show. Well, join me today, asking a question today on the show.
Mikah Sargent [00:15:16]:
Really appreciate it. This was a great question. Uh, very excited to see it pop up. Thank you, and I will catch you again next week for another episode of Hands-On Tech. Bye-bye.
