How To Audit What Chatgpt Knows About You - And Reclaim Your Data Privacy

9 hours ago

Follow ZDNET: Add america arsenic a preferred source on Google.

If you're 1 of nan 900 cardinal people who reportedly use ChatGPT each week, nan chatbot might beryllium a staple of life. Maybe it helps you get activity done aliases travel up pinch repast plans. You mightiness moreover consult it whenever you person a scuffle pinch a friend aliases family member.

But arsenic you move to ChatGPT for progressively much successful your life, you whitethorn want to re-evaluate how overmuch individual accusation you're disclosing on nan way. Ideally, you cognize not to disclose delicate financial accusation -- but different specifications astir you could besides beryllium worthy shielding.

Also: I put GPT-5.5 done a 10-round test: It sewage a near-perfect score

(Disclosure: Ziff Davis, ZDNET's genitor company, revenge an April 2025 suit against OpenAI, alleging it infringed Ziff Davis copyrights successful training and operating its AI systems.)

Privacy experts are already sounding nan siren astir nan imaginable harms of saying excessively overmuch to your chatbot. The underlying interest is that nary 1 is wholly judge really your individual information, whether delicate aliases seemingly innocuous, could beryllium utilized successful nan future. Some fearfulness individual information could extremity up successful a wide surveillance strategy aliases beryllium utilized successful different unforeseen ways that will yet harm aliases disadvantage you.

That ambiguity, they argue, is logic capable for caution.

Here are 5 ways you tin amended negociate nan magnitude of individual accusation ChatGPT has astir you, if you're utilizing a user account.

1. Opt retired of training information

One measurement you tin return to make your ChatGPT acquisition much unafraid is to extremity OpenAI from utilizing your accusation to train its models. Security experts are voicing interest that if your information ends up successful a model, it could 1 time beryllium utilized successful a measurement we can't moreover expect correct now.

Go to Settings > Data controls > Improve nan exemplary for everyone. Then toggle nan switches disconnected and click "Done."

Also: OpenAI is training models to 'confess' erstwhile they dishonesty - what it intends for early AI

You tin besides usage OpenAI's privateness portal to "Make a Privacy Request." Select "I person a user ChatGPT account," and past "Do not train connected my content." From there, you mightiness beryllium prompted to motion in. After that step, you'll spot a fastener to "Submit Request."

This method only applies to your information successful ChatGPT moving forward.

2. Delete aged chats

Another action you tin return to cleanable up nan accusation you've fixed to ChatGPT is to delete aged chats. There are 2 ways to do this. One is to spell to Settings > Data controls > Delete each chats.

You tin besides delete individual chats from nan left-hand sidebar by clicking nan 3 dots adjacent to nan sanction of nan chat.

Also: How to cleanable up your integer footprint - and why it matters much than you think

Although nan speech will vanish from your chat history immediately, it tin return up to 30 days to beryllium permanently deleted from OpenAI's systems, according to nan company's website.

OpenAI besides stipulates 2 exceptions wherever this norm mightiness not apply: circumstances wherever nan institution has to bent connected to information for "security aliases ineligible obligations" aliases because nan information has been "de-identified and disassociated from your account."

3. Use impermanent chats

If you don't want to support up pinch deleting chats arsenic you go, you tin usage ChatGPT's temporary chats. A impermanent chat will not look successful your history, nor will it reference thing from erstwhile conversations aliases memories. It will not beryllium utilized for training data, either.

Similar to nan retention argumentation for deleted chats, OpenAI mightiness clasp a transcript of your impermanent chat for up to 30 days, according to an FAQ page.

To commencement a impermanent chat, click nan fastener branded "Temporary" successful nan bottom-right of a caller chat.

For immoderate ChatGPT users, this mightiness lead to a little personalized experience, arsenic ChatGPT won't beryllium capable to study thing caller astir you that mightiness pass early responses.

4. Manage memories

The thought down memories is for ChatGPT to clasp definite chat specifications that could, successful theory, make nan chatbot much useful complete time. For example, you mightiness inquire it to retrieve that you person a canine aliases are vegan.

According to ChatGPT's FAQ connected memories, location are 2 main settings you tin usage to power memory.

Also: 11 ways to delete aliases hide yourself from nan internet

You tin spell to Settings > Personalization and past click nan "Manage" fastener adjacent to "Memory." That measurement will propulsion up a database of saved memories that you tin either delete altogether aliases individually. You tin besides toggle disconnected nan switches for "Reference saved memories" and "Reference chat history."

OpenAI mightiness besides support a log of saved memories for up to 30 days.

5. Delete your relationship

It's a much utmost measure, but you tin ever delete your account altogether. It's a imperishable move, though, truthful beryllium judge it's what you want. One measurement to do this is to spell to OpenAI's privateness portal and "Make a Privacy Request." Select "I person a user ChatGPT account," and past "Delete my ChatGPT account."

You tin besides spell to Settings > Account and past click "Delete" nether "Delete account. According to OpenAI, you tin only do this if you logged connected wrong nan past 10 minutes; otherwise, you'll person to motion successful again. From there, you will person to type successful your email to corroborate and "DELETE," which will unlock nan "Permanently delete my account" button. Finally, click that button.

How to find retired what ChatGPT knows

If you're unsure of really overmuch accusation you've fixed ChatGPT -- and whether you should return immoderate of nan steps supra -- there's a measurement you tin get a reasonably broad thought of what nan chatbot knows astir you. Just inquire ChatGPT.

My editor, Aly Windsor, asked ChatGPT straight really overmuch it knew astir her. It replied pinch a thorough database of individual specifications she'd shared pinch it. She past asked it to nutrient a punctual that could succinctly spell retired everything it knew astir her successful a scannable profile, and fed nan punctual backmost to ChatGPT. You tin effort nan aforesaid approach. You mightiness beryllium amazed by what and really overmuch it returns.