ChatGPT privacy settings - what they actually do
i never checked my ChatGPT privacy settings until i exported my data. 847 conversations over 14 months. legal strategy questions, business plans, personal stuff, code debugging, even some journaling. all sitting on a server in San Francisco. tagged to my email, my IP, my identity.
tl;dr: ChatGPT's privacy settings let you opt out of training and delete conversations, but OpenAI still stores your data for 30 days and logs your IP permanently. the real fix is switching to a tool that does not collect data in the first place.
key takeaways:
- disabling training only stops future use of your data, existing conversations may already be in the training set
- OpenAI retains IP addresses, payment info, and device data permanently even after you delete conversations
- NanoGPT and local Ollama models eliminate the data collection problem entirely
that export was the moment i started looking for alternatives. but before i switched, i locked down everything i could. here's how - and why it's still not enough.
step 1: disable training on your data
short answer: turn off 'improve the model for everyone' in ChatGPT settings to stop OpenAI from using your new conversations for training, though past data is already captured.
this is the big one. by default, OpenAI uses your conversations to train future models. most people never touch this setting.
how to disable it:
- click your name (bottom-left)
- go to Settings
- click Data Controls
- turn OFF "Improve the model for everyone"
- click Done
what this does: OpenAI stops using your new conversations for training. existing conversations may already be in their training data. there's no undo button for that.
what this doesn't do: OpenAI still stores your conversations for 30 days for "abuse monitoring." they still log your IP address. they still know who you are. disabling training is damage control, not privacy.
step 2: delete your conversation history
short answer: delete your ChatGPT conversations weekly to limit exposure, though OpenAI may retain data for legal purposes even after deletion.
even with training disabled, conversations are stored for 30 days. delete them regularly.
how to delete:
- hover over a conversation in the sidebar
- click the trash icon
- confirm deletion
pro tip: delete conversations weekly. set a calendar reminder. i do mine every sunday night - takes 2 minutes.
the catch: even after deletion, OpenAI may retain data for legal and safety purposes. their privacy policy says they can keep data if they believe it's needed for "legal obligations." so "delete" really means "mark for eventual deletion, maybe."
step 3: export your data (then delete everything)
want to see what OpenAI has on you? export first. you might be surprised.
short answer: export your ChatGPT data to see exactly what OpenAI has on you, then delete everything from their servers.
how to export:
- Settings → Data Controls
- click "Export Data"
- confirm
- check email for the download link
the export includes all conversations, account info, and usage data. download it, review it, then delete everything you don't want sitting on their servers.
my export: 847 conversations spanning 14 months. personal questions, work strategy, code debugging, journaling. all on a server i don't control. that was the moment i switched to NanoGPT.
step 4: use a separate email
don't use your main email for ChatGPT. create a ProtonMail alias.
short answer: create a ProtonMail alias specifically for AI tools to break the link between your AI usage and your real identity.
why: your email is linked to all conversations. if OpenAI gets hacked or shares data, your real email is exposed. ProtonMail aliases are free and take 30 seconds to set up.
how:
- create a free ProtonMail account
- go to Settings → Addresses
- create an alias specifically for AI tools
- use only that alias for ChatGPT, Claude, Gemini, etc.
this breaks the link between your AI usage and your real identity. easiest win on this list.
step 5: use a VPN
your IP address reveals your location and identifies you to your ISP. use a VPN when accessing ChatGPT.
recommended: Mullvad ($5/month, accepts Monero, no email needed) or ProtonVPN (free tier available).
short answer: use Mullvad or ProtonVPN when accessing ChatGPT to hide your IP address from OpenAI's logs.
avoid: free VPNs, NordVPN (data issues), any VPN claiming "military-grade encryption" - that's marketing nonsense.
your IP address alone can identify you to anyone with a subpoena. a VPN adds one layer of separation between you and OpenAI's logs.
step 6: consider switching
even with all these settings, ChatGPT still stores data for 30 days. they still know your IP (unless you VPN). they still have your payment info. and their privacy policy gives them broad rights to use your data.
if privacy actually matters to you - not "i toggled a setting" privacy, but real privacy - you need a tool that doesn't collect data in the first place.
short answer: even with all settings locked down, ChatGPT stores data for 30 days and logs your IP. switching to NanoGPT or local models eliminates the problem.
| feature | ChatGPT | NanoGPT | local AI (Ollama) |
|---|---|---|---|
| data stored | 30 days minimum | none after processing | never leaves device |
| account required | yes | no | no |
| payment traceable | yes (credit card) | no (Monero) | free |
| models available | GPT-4 only | 125+ models | Llama, Mistral |
| setup difficulty | easy | easy | hard |
| cost | $20/month | ~$2-8/month | free (hardware costs) |
NanoGPT is the middle ground. pay-per-prompt, crypto payments, no conversation storage. you get GPT-4o, Claude, Gemini, and 120+ other models without an account. i've been using it for 4 months - see my full review.
Local AI (Ollama + Llama 3) is the ultimate option, but requires decent hardware and technical setup. see my self-hosted AI guide for when it makes sense.
what OpenAI actually stores
let's be specific. i pulled this from their privacy policy and my data export:
| data type | retention | used for training? | can you delete? |
|---|---|---|---|
| prompts | 30 days | yes (unless opted out) | yes (after 30 days) |
short answer: OpenAI keeps your prompts for 30 days, your IP and payment info permanently, and your usage patterns forever. deleting conversations does not erase everything.
| conversations | until deleted | yes (unless opted out) | yes | | IP address | stored | no | no | | account email | permanent | no | no (must delete account) | | usage patterns | permanent | yes | no | | payment info | permanent | no | no (must delete account) | | device info | permanent | no | no | | location (from IP) | permanent | no | no |
that's a lot of data. and "30 days" for prompts doesn't mean it's gone after 30 days - it means it's moved to a different database. OpenAI's policy is clear: they can retain data for legal, safety, and research purposes indefinitely.
the uncomfortable truth
ChatGPT's privacy settings give you the feeling of privacy without the reality of it.
what you can control:
- training opt-out (new conversations only)
- conversation deletion (after 30-day retention)
- export and review your data
short answer: ChatGPT's privacy settings give you the feeling of privacy without the reality. they reduce data collection but never eliminate it.
what you can't control:
- IP logging
- payment info storage
- device fingerprinting
- server-side data retention for "safety"
- third-party data sharing (check their policy - it's broad)
OpenAI employees can access your conversations for safety and abuse monitoring. this is in their terms of service. if you're sending sensitive information, assume a human could read it.
what i actually did
- flipped all the settings above (5 minutes)
- exported my data (eye-opening)
- deleted all old conversations
- created a ProtonMail alias for AI tools
- switched to NanoGPT for daily use
- set up Ollama for sensitive queries
- deleted my ChatGPT account
short answer: after exporting 847 conversations over 14 months, i switched to NanoGPT for daily use and Ollama for sensitive queries. total time: about an hour.
took about an hour total. the hardest part was deciding to actually do it.
the bottom line
ChatGPT's privacy settings are a band-aid. they reduce data collection but don't eliminate it. OpenAI is an advertising-adjacent company that monetizes data - their privacy settings are designed to make you feel comfortable, not to actually protect you.
the real fix: use a tool that doesn't collect your data in the first place. NanoGPT for daily AI, Ollama for sensitive stuff. that's what i do. that's what works.
last updated: july 2026
short answer: ChatGPT's privacy settings are damage control, not privacy. the real fix is using a tool that does not collect your data in the first place.