How to audit what ChatGPT knows about you – and reclaim your data privacy


gettyimages-2268540971

Bloomberg / Contributor/ Bloomberg via Getty

Follow ZDNET: Add us as a preferred source on Google.


If you’re one of the 900 million people who reportedly use ChatGPT every week, the chatbot might be a staple of life. Maybe it helps you get work done or come up with meal plans. You might even consult it whenever you have a scuffle with a friend or family member. 

But as you turn to ChatGPT for increasingly more in your life, you may want to re-evaluate how much personal information you’re disclosing along the way. Ideally, you know not to disclose sensitive financial information — but other details about you could also be worth shielding. 

Also: I put GPT-5.5 through a 10-round test: It got a near-perfect score

(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Privacy experts are already sounding the alarm about the potential harms of saying too much to your chatbot. The underlying concern is that no one is entirely sure how your personal information, whether sensitive or seemingly innocuous, could be used in the future. Some fear personal data could end up in a mass surveillance system or be used in other unforeseen ways that will ultimately harm or disadvantage you. 

That ambiguity, they argue, is reason enough for caution. 

Here are five ways you can better manage the amount of personal information ChatGPT has about you, if you’re using a consumer account. 

1. Opt out of training data

One step you can take to make your ChatGPT experience more secure is to stop OpenAI from using your information to train its models. Security experts are voicing concern that if your data ends up in a model, it could one day be used in a way we can’t even anticipate right now.

Go to Settings > Data controls > Improve the model for everyone. Then toggle the switches off and click “Done.”

Also: OpenAI is training models to ‘confess’ when they lie – what it means for future AI

You can also use OpenAI’s privacy portal to “Make a Privacy Request.” Select “I have a consumer ChatGPT account,” and then “Do not train on my content.” From there, you might be prompted to sign in. After that step, you’ll see a button to “Submit Request.” 

This technique only applies to your data in ChatGPT moving forward.

2. Delete old chats 

Another action you can take to clean up the information you’ve given to ChatGPT is to delete old chats. There are two ways to do this. One is to go to Settings > Data controls > Delete all chats.

You can also delete individual chats from the left-hand sidebar by clicking the three dots next to the name of the chat. 

Also: How to clean up your digital footprint – and why it matters more than you think

Although the conversation will disappear from your chat history immediately, it can take up to 30 days to be permanently deleted from OpenAI’s systems, according to the company’s website

OpenAI also stipulates two exceptions where this rule might not apply: circumstances where the company has to hang on to data for “security or legal obligations” or because the data has been “de-identified and disassociated from your account.” 

3. Use temporary chats

If you don’t want to keep up with deleting chats as you go, you can use ChatGPT’s temporary chats. A temporary chat will not appear in your history, nor will it reference anything from previous conversations or memories. It will not be used for training data, either. 

Similar to the retention policy for deleted chats, OpenAI might hold a copy of your temporary chat for up to 30 days, according to an FAQ page.

To start a temporary chat, click the button labeled “Temporary” in the bottom-right of a new chat. 

For some ChatGPT users, this might lead to a less personalized experience, as ChatGPT won’t be able to learn anything new about you that might inform future responses. 

4. Manage memories

The idea behind memories is for ChatGPT to retain certain chat details that could, in theory, make the chatbot more useful over time. For example, you might ask it to remember that you have a dog or are vegan.

According to ChatGPT’s FAQ on memories, there are two main settings you can use to control memory.

Also: 11 ways to delete or hide yourself from the internet

You can go to Settings > Personalization and then click the “Manage” button next to “Memory.” That step will pull up a list of saved memories that you can either delete altogether or individually. You can also toggle off the switches for “Reference saved memories” and “Reference chat history.”

OpenAI might also keep a log of saved memories for up to 30 days.

5. Delete your account

It’s a more extreme measure, but you can always delete your account altogether. It’s a permanent move, though, so be sure it’s what you want. One way to do this is to go to OpenAI’s privacy portal and “Make a Privacy Request.” Select “I have a consumer ChatGPT account,” and then “Delete my ChatGPT account.” 

You can also go to Settings > Account and then click “Delete” under “Delete account. According to OpenAI, you can only do this if you logged on within the last 10 minutes; otherwise, you’ll have to sign in again. From there, you will have to type in your email to confirm and “DELETE,” which will unlock the “Permanently delete my account” button. Finally, click that button.

How to find out what ChatGPT knows 

If you’re unsure of how much information you’ve given ChatGPT — and whether you should take any of the steps above — there’s a way you can get a fairly comprehensive idea of what the chatbot knows about you. Just ask ChatGPT.

My editor, Aly Windsor, asked ChatGPT directly how much it knew about her. It replied with a thorough list of personal details she’d shared with it. She then asked it to produce a prompt that could succinctly spell out everything it knew about her in a scannable profile, and fed the prompt back to ChatGPT. You can try the same approach. You might be surprised by what and how much it returns. 





Source link

Leave a Reply

Subscribe to Our Newsletter

Get our latest articles delivered straight to your inbox. No spam, we promise.

Recent Reviews


The battle between AMD and NVIDIA rages on eternally, it seems, though it’s rather a one-sided battle in the desktop PC market, where NVIDIA holds something like 95%, and AMD most of what’s left apart from Intel’s (almost) 1%.

But as dominant and popular as NVIDIA is, AMD proponents could always raise the value argument. On a per-dollar basis, you get more value with an AMD card, and even better, you have the benefit of AMD “FineWine” which ensures your card will become even better with time.

What “FineWine” meant—and why it mattered

FineWine was something that AMD fans began to notice during the GCN (Graphics Core Next) architecture. Incidentally, the last AMD dedicated GPU I bought was the R9 390, which was of that lineage. Since then, all my AMD GPUs have been embedded in consoles or handheld PCs, but I digress.

The R9 390 is actually a good example of FineWine. Launched in 2015, like many AMD cards, the R9 390 had a rough start, and I sold mine in exchange for a stopgap card in the form of the RTX 2060, because I wanted to play Cyberpunk 2077 on PC, where it wasn’t broken the way it was on consoles. Even though, on paper, the raw power of the RTX 2060 wasn’t much more than a 390, the AMD card’s performance on my (then) 1080p monitor was a stuttery mess, whereas everything suddenly ran great on my 2060 the minute the AMD GPU was expunged from the system.

But, a decade later, that same game is perfectly playable on this card, as you can see in this TechLabUK video.

A lot of it is because the developers have kept patching and improving the game, but this is something you see across the board for AMD cards on various games. This is FineWine. Years later, with continued driver updates from AMD, the cards go from being a little worse than their NVIDIA equivalent at launch to being as good or even a little better in the long run.

Of course, that’s not super helpful to customers who buy hardware at launch, but it has given some AMD users computers with longer lifespans than you’d think, and made many used AMD cards an even better bargain.

Why AMD’s FineWine era worked

A bit of smoke and mirrors

The PULSE AMD Radeon RX 6800 XT next to an AMD RX 6600 XT Phantom Gaming D. Credit: Ismar Hrnjicevic / How-To Geek

FineWine wasn’t magic, of course. The phenomenon was the result of a mix of factors. AMD’s architectures were in some cases a little too forward-thinking for the APIs of the day. Massively parallel with a focus on compute, they’d only come into their own with DirectX 12 and more modern games. NVIDIA’s cards at the time were better optimized to run current games well. Over time, NVIDIA cards would make similar architectural changes, but with better timing.

The other reason FineWine was a thing came down to driver maturity. As a much smaller company with fewer resources, it seems that AMD had some trouble releasing cards with optimized drivers. So, over time, the card would start performing as intended.

In both cases, you could frame FineWine not as the card getting better, but rather getting “less worse” over time. If you set the bar low at launch, the only way is up. However, there’s a third factor to take into account as well. AMD dominates console gaming. The two major home console series have now run on AMD GPUs for two generations, and so games are developed with that hardware in mind. This also gives newer titles a bit of a leg up, though it’s hard to know exactly by how much.

How AMD moved on from FineWine

It seems worse, but it’s actually better

An AMD RX 9070 XT Gigabyte gaming graphics card. Credit: Ismar Hrnjicevic / How-To Geek

With the shift to RDNA architecture, AMD made a deliberate change in philosophy. Modern Radeon GPUs are designed to perform well right out of the gate. Reviews on day one are much closer to what you could expect years later. There are still decent gains to be had on RDNA cards with game-specific optimizations (Spider-Man on PC is a great example), but the golden age of FineWine seems to be in the past now.

That’s a good thing! Products should put their best foot forward on day one, so let’s not shed a tear for FineWine in that regard. So it’s not so much that AMD doesn’t care about improving the performance and stability of older cards over the years, it’s that the company is now better at its job, and so there’s less room for improvement.

Sapphire NITRO+ AMD Radeon RX 9070 XT GPU

Cooling Method

Air

GPU Speed

2520Mhz

The AMD Radeon RX 9070 XT from Sapphire features 16GB of DDR6 memory, two HDMI and two DisplayPorts, and an overengineered cooling setup that will keep the card cool and whisper quiet no matter the workload.


NVIDIA kept the idea—but changed the formula

It’s all about AI

It’s funny, but these days I think of NVIDIA cards as the ones with major longevity. Take the venerable GTX 1080 and 1080 Ti cards. These cards only lost game-ready driver support in 2025, which doesn’t immediately make them useless, it just means no more optimization for those chips. What an incredible run, getting a decade of relevant game performance from a GPU!

But, that’s not really NVIDIA’s take on FineWine. Instead, the company has taken to adding new and better features to its cards long after they’ve been launched. Starting with the 20-series, the presence of machine-learning hardware means that by improving the AI algorithms for technologies like DLSS, these cards have become more performant with better image quality over time.

While NVIDIA has made some features of its AI technology exclusive to each generation, so far all post 10-series GPUs benefit from every new generation of DLSS. Compare that to AMD which not only offers inferior versions of this new upscaling technology, but has locked the better, more usable versions to later cards, such as the case with FSR Redstone.


FineWine is an ethos, not a brand

In the case of my humble RTX 4060 laptop, the release of DLSS 4.5 has opened new possibilities, notably the ability to target a 4K output resolution, which was certainly not on the table when I first took this computer out of the box. We might not call it “FineWine,” but it sure smells like it to me!



Source link