AI Privacy Settings: the Cross-Platform Controls Checklist

Key Takeaway

• 🎛️ AI privacy settings decide training, not storage: turning the main toggles off stops future training use of your chats; the chats usually stay stored until you delete them separately.
• 🤖 ChatGPT shares by default: on Free, Plus and Pro accounts, “Improve the model for everyone” is on until you turn it off under Settings → Data controls.
• 🔍 Gemini keeps a mirror: the Keep Activity setting stores your prompts, and even after you switch it off Google holds your chats up to 72 hours for processing and feedback.
• 🧠 Claude retention swings 30 days to 5 years: leaving “Help Improve our AI models” on can keep your chats up to five years; opting out cuts the window to about 30 days.
• ⏱️ The tonight-action: run the 15-minute sweep below across all three chatbots plus the assistants on your phone, then re-check your AI privacy settings after every major app update.

Default-on data sharing is the single fact most people miss when they first use an AI chatbot. On ChatGPT Free, Plus and Pro, OpenAI states that data sharing for model training is “enabled by default,” and the switch sits one menu deep under Settings → Data controls. Gemini stores every signed-in prompt in your Google Account while its Keep Activity setting stays on. Claude may keep consumer conversations for years if its model-improvement setting stays on. These are the AI privacy settings that actually move your data out of the training pipeline, and all three take minutes to change. This walkthrough shows the exact clicks for each platform, what each control does and does not do, and the one ordered sweep that covers the whole job.

We verified every step against OpenAI’s data-controls help page, Google’s Gemini Apps activity guide and Anthropic’s privacy center on November 10, 2026. Menu names change as the apps update, so treat the wording as current-but-re-checking: the settings’ locations are stable, the labels are not.

What Your AI Privacy Settings Can and Cannot Do

One split explains most confusion in this area: training controls and storage controls are separate. OpenAI’s help center says this directly — turning off “Improve the model for everyone” does not delete or hide saved chats. Deleting a chat, deleting your account, or clearing your history are different actions at different places. The same split runs through Google and Anthropic: Keep Activity and Claude’s toggle decide what feeds the model builders and how long copies live; the delete commands decide what you can still see.

The second split is training controls and ad controls are separate. Where ads exist, ad personalization has its own switch — OpenAI documents separate ads controls, and Google’s ad personalization lives in your account, not in Gemini’s activity page. If your worry is a marketer, check ads settings after the training pass.

Where you use the app matters too. OpenAI applies your choice across all signed-in devices at once, but for signed-out use the choice lives in that browser only. Work and school accounts sit under an administrator: Google states that Workspace admins control Gemini Apps Activity settings, and enterprise AI products (OpenAI Business/Enterprise, Anthropic’s Claude for Work, the API tier) do not use your content for training by default. The AI privacy settings below are for the personal side of the fence.

ChatGPT: the Data Controls Pass

Your AI privacy settings on ChatGPT live in one place — the Data controls panel — and one toggle does the heavy lifting.

Turn off training sharing (web)

1. Open your account menu (top-right profile icon).
2. Select Settings.
3. Select Data controls.
4. Select Improve the model for everyone, turn it off, and select Done.

Turn off training sharing (mobile)

1. Open the sidebar and tap your profile icon to open Settings.
2. Tap Data controls.
3. Toggle off Improve the model for everyone.

The Data controls panel carries the rest of your AI privacy settings: chat history and training, archived chat handling, temporary chats, memory, data export and account deletion. OpenAI keeps the full panel documented on its Data controls in ChatGPT help page.

What the off position really buys you

After the toggle, new conversations stop feeding the training pipeline. Three caveats from OpenAI’s own documentation are worth your three seconds each:

• Feedback re-opens the door. If you hit thumbs-up or thumbs-down after opting out, “the entire conversation associated with that feedback may be used to train OpenAI models.” Avoid rating chats that carry sensitive content.
• Temporary chat is the one-tap privacy mode. Temporary chats do not appear in history, do not create or update memories, and are not used to improve the models — though they may be retained up to 30 days for safety. Start one from the chat composer when you paste anything sensitive.
• Archiving changes nothing. Archived chats follow your training setting as before; archiving is a tidy-up action, not a privacy action.

Two smaller switches sit in the same neighborhood. Voice mode has a separate data-sharing control for training. And if you run coding tasks through Codex, that tool has its own “Include environments” switch inside Codex settings — the ChatGPT toggle does not reach it.

The memory question

OpenAI treats memory and training as different settings. Memory is what lets ChatGPT personalize answers across chats; if you want a blank slate per conversation, turn memory off in Data controls or use temporary chat. Both fit inside the same AI privacy settings pass, and we map the memory internals separately in our ChatGPT memory setting explainer.

Gemini: Keep Activity, the Delete Commands and the 72-Hour Rule

Your AI privacy settings on Gemini center on one page — Gemini Apps Activity — which Google recently renamed around “Keep Activity.”

Turn it off (web)

1. Go to myactivity.google.com/product/gemini (or click your profile in Gemini → Settings & help → Activity).
2. At the top, the switch reads On. Click it.
3. Choose Turn off — or Turn off and delete activity if you want the stored prompts deleted in the same move.

In the mobile app

1. Open the Gemini app → Menu → your profile → Gemini Apps Activity.
2. Tap On at the top, then Turn off or Turn off and delete activity.

The part almost every explainer misses is the retention gap behind the toggle. Google’s help page states plainly: “Even when Keep Activity is off, your conversations will be saved with your account for up to 72 hours to allow Google to provide the service and process any feedback.” Those hidden copies never appear in your activity list, and they are gone only on Google’s clock. For routine chats the 72-hour pool is harmless; for anything you would hate to see leave your hands, the honest lesson is that your AI privacy settings narrow future exposure, but no consumer AI toggle erases data the way your bank statement erases a payment.

The audio and Live-video box

Inside the same Activity page there is a separate subscription: “Improve Google services with your audio and Gemini Live videos and screenshares.” It is off by default and requires Keep Activity to stay on. If you never intend to let Google train on your voice recordings and camera streams with the AI assistant, leave Keep Activity off after your sweep — or if you need the history back, keep Keep Activity on but confirm that box stays unchecked.

Deleting what is already stored

On the same page, click Delete → All time to clear your whole Gemini prompt history, or use the ranges (last hour, last day, custom range) for a scrape. Google also honors auto-delete timers you can set on that page — a middle path if you want history kept briefly for handoff and then shredded. Workspace accounts, again, follow your administrator, not this menu.

Claude: the Model-Improvement Toggle

Your AI privacy settings on Claude live under Settings → Privacy, and the control that matters is one switch.

1. Sign in at claude.ai.
2. Open the menu on your profile icon → Settings.
3. Select Privacy.
4. Under Help Improve our AI models, click the button to toggle it off.

Anthropic documents that exact path in its privacy center guide to model-improvement settings. The retention stakes around that one toggle are the sharpest of the three platforms. Anthropic’s consumer policy lets chats from users who leave training on be retained for up to five years (de-identified); turning the setting off stops the future training runs and shortens retention to roughly 30 days. Either way the change is forward-looking: Anthropic states that chats already inside “a training run that has started” are not pulled back, and models already trained keep what they learned. The lesson for your AI privacy settings is to not wait for a breach headline — the toggle works from the next chat, not the last one.

Two Claude-specific safety valves round out the picture. Incognito chats are never used for training even when the model-improvement setting is on — Anthropic’s own sensitive-data help page names that exemption. And Anthropic’s guidance to consumers is blunt about what never belongs in a consumer chatbot input at all: financial details like card or account numbers, health records, passwords, and confidential documents. The trained-on-the-mistakes version of that advice sits in our AI data privacy ledger, which tracks which AI services actually train on what you paste and why.

The Phone Layer: Assistants That Were Watching Before Chatbots

Your AI privacy settings are not only in the chat apps. Phones ship with two always-listening surfaces of their own — the voice assistant and the keyboard’s cloud suggestions — and both have their own switches.

iPhone

1. Settings → Apple Intelligence & Siri → check the ChatGPT page (the extension you may have enabled) and its personal-data options.
2. Settings → Privacy & Security → Analytics & Improvements → turn Improve Siri & Dictation off.
3. While you are there, review Improve Apple Intelligence-type reporting toggles in the same menu.

Android

1. Open myactivity.google.com → Web & App Activity → set it off, or leave it on with an auto-delete timer (3, 18 or 36 months).
2. Settings → Google → Settings for Google apps → the digital-assistant (Gemini) page — confirm your activity choice matches the web pass.
3. Keyboard: Gboard’s smart features pull from your Google Account’s controls, so the Web & App Activity choice already covers the main leak.

One pass covers the assistant surfaces: the Gemini activity control you already set flows into the assistant and the voice recorder boxes; the iPhone pass is a two-toggle job. Your shared-family-device layer deserves its own look too — a partner’s or child’s sign-in keeps separate AI privacy settings, and the same sweep applies per account. If the phone side is about where recordings get processed and which household apps train on meeting notes, our AI note-taker privacy ledger maps the meeting-recording apps specifically.

The 15-Minute AI Privacy Sweep (Tonight)

Run the steps in this order; the whole pass lands around, at most, a quarter of an hour. This is the artifact your AI privacy settings work reduces to:

1. ChatGPT (3 min): account menu → Settings → Data controls → “Improve the model for everyone” off → Done. Repeat on the mobile app if you use it (the signed-in setting syncs, but confirm on your device).
2. Gemini (3 min): myactivity.google.com/product/gemini → toggle Turn off and delete activity → confirm. Then Delete → All time on any leftover history. Then the Web & App Activity choice for the assistant layer.
3. Claude (2 min): claude.ai → Settings → Privacy → toggle “Help Improve our AI models” off. Scan for incognito on anything sensitive going forward.
4. Phone assistant (4 min): iPhone: Improve Siri & Dictation off. Android: Web & App Activity off or on a timer.
5. Verify (3 min): reopen each settings page and confirm the toggles read off; confirm the Keep Activity page says off; confirm Claude’s toggle row shows the off state. Screenshot all three for your own record.

A worked example of the shape this takes in one app: you open myactivity.google.com/product/gemini, the switch at top reads On; you click it, choose Turn off and delete activity, confirm the disclosure, and the page flips to Off with your stored prompts removed. Total elapsed: under a minute — that is the whole pattern across the other two apps, and the reason the sweep finishes tonight rather than “one day.”

What Your AI Privacy Settings Still Will Not Fix

Set the toggles, then keep the boundaries honest:

• Already-trained stays trained. Anthropic’s rule — opt-out changes the pipeline from the next chat; ChatGPT’s archived chats follow your setting; Gemini’s 72-hour pool rides its own clock. Retention windows apply to copies already stored.
• Ads live outside this pass. OpenAI documents separate ads controls; Google ad personalization sits in your account, not in Gemini. If ad noise is your worry, close that loop on its own page.
• The pasted-content leak needs judgment, not settings. Consumer chatbots with training off still process your paste on their servers. Contracts and non-disclosure terms beat toggles — and the account side has its own hardening list in the nine account-security settings worth locking on every AI login you keep.
• Enterprise accounts need none of this. Business, education and API tiers of all three providers do not train on your content by default — nothing to switch on those.

Household backups are the last mile. Family photos and voice notes you sync to self-held storage dodge the model builders entirely; our self-hosted photo backup guide is the zero-cloud path. And none of the settings above touch your legal rights — export, deletion and complaint routes sit in law, not in app menus; our data privacy rights guide walks those routes for Philippine readers.

FAQ

Do AI privacy settings delete my old chats?

No. OpenAI states that turning off “Improve the model for everyone” does not delete or hide saved chats; deletion is its own action in your chat list or account settings. The same separation holds on Gemini and Claude: history lives until you delete it, training stops when you toggle.

Is one toggle enough on ChatGPT?

One toggle covers training for ordinary chats. The exceptions to know: rated feedback may send the whole associated conversation to training even after you opt out, voice mode has a separate data-sharing control, and Codex has its own environment switch. Temporary chat is your clean-room mode for one-off sensitive asks.

What happens to my Gemini chats after I turn Keep Activity off?

Google saves them up to 72 hours to run the service and process feedback, then those copies expire. They will not appear in your activity list while they are held. Future conversations stop being stored — and stop being trainable — once the switch is off.

Can I pull my past Claude chats out of training?

Not retroactively. Anthropic states that chats already in a training run that started are not removed, and models already trained keep what they learned. The toggle protects everything from your next conversation onward.

Do AI privacy settings stop the apps from using my content for ads?

No. Ad personalization is a separate control set on every platform — OpenAI documents dedicated ads controls, and Google’s ad settings live in your account rather than the Gemini activity page. Turn your AI privacy settings first, then set ad controls if that is also your concern.

What about signed-out or incognito use?

ChatGPT saves your signed-out choice in that browser only, so re-set it per browser. Claude’s incognito chats are exempt from training even with the main setting on. Gemini requires a signed-in account to manage its activity controls at all, so there is no incognito tier to govern.

How often should I re-check these settings?

Re-run the sweep after every major app update and at each quarterly check-in. Menu labels move — Gemini’s page was renamed around “Keep Activity” only recently — and providers occasionally add new toggles (voice, ads, agents) that arrive switched to their default, not to yours.

@online{agron2026aiprivacy,
  author       = {Agron, Edmon},
  title        = {{AI Privacy Settings: the Cross-Platform Controls Checklist}},
  year         = {2026},
  date         = {2026-11-10},
  organization = {WorldNgayon},
  url          = {https://worldngayon.com/ai-privacy-settings/},
  urldate      = {}
}

TY  - ELEC
AU  - Agron, Edmon
TI  - AI Privacy Settings: the Cross-Platform Controls Checklist
T2  - WorldNgayon
PY  - 2026
DA  - 2026/11/10/
UR  - https://worldngayon.com/ai-privacy-settings/
Y2  - 
ER  -