How to opt out of AI training: the setting in 7 apps

To opt out of AI training, you have to find a separate switch in each app, and some of them are on until you turn them off. Below is the exact setting for seven services, taken from each company's own help page on November 11, 2026, with what the switch does not cover.

There is no single "opt out of AI" button for the internet. Each company decides whether to train on your data, what counts as your data, and where the switch lives. Some switches are for you, some only for an admin, and some just send AI companies a polite request.

Quick table

Service	Setting name	Who can change it	Default
Google Gemini app	Keep Activity	You	Check yours
YouTube (outside AI companies)	Third-party training	Channel owner	Off
Slack	Email request for global models	Workspace owner	Not used for generative AI
Substack	Tell AI tools not to train their models on your content	Publication owner	Off
Medium	Third-party AI and your stories	You	Prioritize maximum reach
Grammarly	Product Improvement and Training	You, or an admin on team plans	On
Figma	Content training	Team admin	On for Starter and Professional

Google Gemini: turn off Keep Activity

Open Gemini Apps Activity and turn off Keep Activity.

According to the Gemini Apps Privacy Hub (last updated October 26, 2026), when Keep Activity is on, Google uses your chats, uploads and related activity to improve its services, "including training generative AI models," and some of it is reviewed by people. With it off, your future chats "won't be used to train our AI models, unless you choose to send Google feedback."

What it does not do:

Chats are still kept with your account for 72 hours so Gemini can answer and for safety.
Chats that were already reviewed by people stay for up to three years, even if you delete your activity.
Sending feedback (the thumbs buttons) shares that conversation again.
With Keep Activity off, connected apps like Google Workspace stop working in Gemini.

If you only want one conversation kept out, Google says temporary chats are not used to train its AI models. For what Google's AI does with Gmail itself, see does Google AI read my Gmail.

YouTube: leave third-party training off

In YouTube Studio, open your channel settings and check the third-party training setting. Rights holders using Content Manager find it on the Settings page.

YouTube's help page on third-party training says this setting is off by default, so if you never turned it on, you don't need to do anything. If you did, you can switch it off or narrow it to "Selected companies only." Changes can take up to seven days to show in the YouTube Data API, which is where outside companies check permissions.

Two limits: the setting is about outside companies building AI models, and the page does not describe a switch for Google's own use. YouTube also says it "can't control what a third-party company does" once it has your videos. Only public videos are eligible in the first place.

Slack: ask your workspace owner to email Slack

There is no toggle. Slack's privacy principles say: "We do not develop generative AI models using Customer Data," and it will not train generative AI on your workspace without the customer opting in.

Slack does use workspace data to train non-generative "global" models, for things like emoji and channel suggestions. To opt out of those, an Org Owner, Workspace Owner or Primary Owner emails feedback@slack.com with the workspace or org URL and the subject line "Slack Global model opt-out request." Regular members cannot do it themselves, so ask whoever runs your workspace.

Substack: the privacy toggle in publication settings

Go to your publication's Settings, scroll to Privacy, and turn on Tell AI tools not to train their models on your content.

Substack's help article is clear that this is a signal, not a lock: it "will only apply to AI tools that respect this setting." It also warns that blocking training may make your publication harder to find in tools and search engines that show AI-generated answers.

Medium: Third-party AI and your stories

On the web, click your profile picture, then Settings, then Account, and scroll to Third-party AI and your stories. In the iOS or Android app, tap your avatar, the settings icon, then Account.

Per Medium's help center, the default is Prioritize maximum reach, which asks AI companies not to train on your stories unless that would stop them crediting and linking to you. Choose Minimize third-party training to ask all AI companies not to train on your stories, even if some AI products then stop linking to you. The setting does not change how your stories appear in search engines or in AI-assisted search features.

Grammarly: Product Improvement and Training

Grammarly's help article says this control is on by default for individual users. To switch it off:

Free and Premium: go to account.grammarly.com/security/privacy and click the Product Improvement and Training toggle.
Pro with one user: go to account.grammarly.com/admin/data_settings and click the same toggle.
Pro with more than one user: only the account's administrators can change it.

With it off, Grammarly says the content you type "will not be used for training or product improvement for other users." It may still use non-content data, such as word counts and which suggestions you accept.

Figma: Content training (admins only)

Only a team admin can change this. Figma's help article says content training is on by default for Starter and Professional teams, and off on Organization and Enterprise plans, where it currently cannot be turned on.

Starter: in the file browser, open your team, click All Folders, open the menu next to the team name and choose View settings. On the Settings tab, scroll to AI, select Manage AI settings and turn off Content training.
Professional: in the file browser, open your team, click Admin, open the Settings tab, scroll to AI and turn off Content training.

Turning it off stops new content and edits from being used. It does not remove access to Figma's AI features, which is a separate setting.

Meta, ChatGPT and Claude

Meta (Facebook, Instagram, WhatsApp): has its own rules; covered in a separate guide.
ChatGPT: has its own data controls; covered in a separate guide.
Claude: has its own privacy setting; covered in a separate guide.

What an opt-out never covers

Three patterns repeat across almost every help page above:

It works going forward. Gemini, Figma and Grammarly all describe the switch as stopping future use. Data already used or already reviewed is not pulled back.
Feedback is separate. Rating an AI answer or reporting a problem usually sends that content in, even with training off.
Public posts can be scraped anyway. Substack and Medium can only ask other AI companies not to train. Whether a company listens is up to that company.

So the most reliable way to keep something out of a training set is to not hand it to a service that can read it. Private messages, drafts and email are the obvious cases.

Frequently asked questions

Is there one setting that opts you out of AI training everywhere?

No. Each company has its own switch, and some, like Slack's, are not a switch at all but a request your workspace owner sends by email. Go through the apps you actually use, starting with the ones that hold private writing.

Does opting out delete data that was already used for training?

Not according to the companies above. Google says chats already reviewed by people are kept for up to three years even after you delete your activity, and the other help pages describe their switches as applying to new content.

Can I stop Google from training AI on my YouTube videos?

The setting YouTube documents covers third-party companies only, and it is off by default. YouTube's help page on third-party training does not describe a switch for Google's own models.

Does turning off Keep Activity stop Gemini from working?

No, Gemini still answers you. Your chats just stop appearing in your activity and, unless you send feedback, are not used to train Google's models. Connected apps like Google Workspace stop working until you turn the setting back on.

Where Secria fits

Email is one place where you don't have to hunt for a switch. Secria Mail has no ads and Secria never mines your mail. Your mailbox is locked with a key made from your password, on your device, and every plan, the free one included, uses hybrid post-quantum encryption. Sign-up is a name, the address you want and a password, with no phone number.

If you want an inbox that is not mined for anything, you can create a free Secria account with no phone number and no card.