How to Review AI Training and Data-Use Controls (2026)

AI privacy settings are not one universal switch. A service may treat prompts, uploads, feedback, account history, diagnostic logs, personalization data, and public web content differently. The applicable rule can also depend on the product, account tier, region, administrator, and whether you submitted feedback.

This guide was reviewed on September 9, 2026. It points to current provider documentation and explains what each control does and does not establish. Settings and names can change, so confirm the live control before entering sensitive information.

The important distinctions

Before changing a setting, identify which activity you are trying to limit:

• Model training or fine-tuning: using data to change or improve a model. An account opt-out normally applies going forward; it does not by itself prove that previously trained model behavior has been reversed.
• Service operation and retention: storing chats, uploads, prompts, logs, or files so the product can respond, provide history, prevent abuse, or meet legal obligations. Turning off training is not the same as deleting stored data.
• Feedback: a thumbs-up, thumbs-down, written comment, bug report, or support ticket may be handled under a separate policy. Do not include personal information in feedback unless necessary.
• Personalization and safety: a service may use activity to personalize results, detect abuse, or improve safety even when a setting excludes data from a particular training program.
• Web crawling: a website owner can publish crawler instructions, such as `robots.txt`. Those instructions concern access by a named crawler; they are not a deletion request, a security control, or a guarantee that every scraper will comply.

Quick provider map

OpenAI: ChatGPT and the API

For personal ChatGPT use, OpenAI's current data-controls guidance says you can open Settings → Data Controls and turn off Improve the model for everyone. Check the live account setting rather than assuming that a Free, Plus, or other personal plan has the same behavior everywhere.

OpenAI says that, by default, it does not use inputs and outputs from ChatGPT Business, Enterprise, Edu, or the API platform to train or improve its models, subject to the product terms and any explicit opt-in. The OpenAI business-data page and API data-controls documentation are the appropriate sources for a business or developer workflow.

Deleting a chat, turning off training, and submitting a formal privacy request are different actions. For deletion or other personal-data requests, use the OpenAI Privacy Portal. Do not paste a private prompt into a public support forum to ask whether it was used.

Anthropic: Claude

Anthropic documents consumer and commercial data use separately. Its Privacy Center explanation says consumer users can review the model-improvement choice, while Incognito chats are not used to improve Claude even when the setting is enabled. The same page points commercial-product users to separate terms.

Review the current Claude privacy controls in the account settings and read the explanation shown for your plan. Do not assume that an API, Claude for Work, education, or other commercial agreement has the same retention or training policy as a personal account. If you submit feedback, treat that submission as a separate disclosure and avoid including unnecessary personal or confidential information.

Google Gemini

Google calls the relevant account control Keep Activity in current guidance, while some help pages and older interfaces use Gemini Apps Activity. Google's activity-management guidance says that when the setting is off, future chats may not appear in activity and, when you do not submit feedback, are not used to improve Google's AI models; Google may still retain them for a limited period and use them for service and safety purposes. The page also describes a possible delay before the setting takes effect.

Google's Gemini Apps Privacy Hub explains that some saved interactions may be reviewed by human reviewers and that reviewed data can have a longer retention period. It also distinguishes connected apps and other product features. Temporary Chat is another product control, not a universal deletion mechanism. Read the current description for the account and feature you are using before treating a chat as ephemeral.

To reduce exposure, turn off the relevant activity setting, delete activity through the official control, review connected apps separately, and avoid uploading information that you would not want retained for the stated service and safety purposes.

Meta AI and public social content

Meta's generative-AI policies and controls vary by country, product, and type of content. Start with the Meta Privacy Center's generative-AI information and use the current option shown for your account. Do not rely on a form or deadline described for a different region.

Public posts, comments, images, and profile information can have different visibility and processing rules from private messages or direct interactions with Meta AI. Changing a post from public to private, deleting content, or changing an AI setting may affect future processing in different ways; none should be presented as proof that a prior model has been retrained. If you submit an objection or privacy request, keep the confirmation and read the response about scope and exceptions.

Grok and X

X's official Grok help page says users can control whether public data and interactions, inputs, and results with Grok are used to train or fine-tune xAI models. The documented path is Privacy & Safety → Data sharing and personalization → Grok & Third-party Collaborators → Data Sharing. The page also says that making an account private can prevent posts from being used for that public-data training route.

The same guidance warns that opting out does not prevent a deployed model from learning through normal use of an X feature powered by Grok, and voluntary feedback can be separate. Treat the toggle as a control with a defined scope, not as a promise that all prior content or model behavior has been erased.

GitHub Copilot

GitHub's individual-subscriber policy guidance says that beginning April 24, 2026, GitHub may use interactions from Copilot Free, Pro, Pro+, and Max plans—including inputs, outputs, code snippets, and context—to train and improve AI models unless the individual disables the setting.

For a personal account, open Copilot settings and set Allow GitHub to use my data for AI model training to Disabled. GitHub separately says Copilot Business and Enterprise customer data is not used to train AI models without customer authorization. An organization-managed seat, repository policy, Copilot cloud agent, session syncing, and model-training control are separate settings; check all that apply to your workflow.

Microsoft Copilot

Do not combine every Microsoft Copilot product into one privacy claim. Microsoft 365 Copilot and Copilot Chat have enterprise data-protection commitments. Microsoft's official documentation says prompts and responses in that service are not used to train the underlying foundation models, while also describing logging, organizational data, and service-boundary controls.

Personal Copilot, consumer Microsoft accounts, Windows features, and other products can have different settings and terms. Open the privacy controls for the exact product and account you use. A statement about Microsoft 365 enterprise data is not evidence about a personal Copilot session.

Perplexity

Perplexity's current data-collection guidance says Free, Pro, and Max users can control AI training data collection through the account settings' AI Data Usage or equivalent control. The same page says the setting applies to data collected after the opt-out date and that Enterprise data is not used for AI training under its enterprise policy.

Check the live label and scope in your account. An opt-out may not delete old searches, feedback, account records, or data processed for security and service operation.

LinkedIn

LinkedIn's Data for Generative AI Improvement guidance says the setting controls use of member data and content to train content-generating AI models going forward. LinkedIn also says the setting does not control models used for personalization, security, safety, or anti-abuse purposes, and that feedback may be handled separately.

The LinkedIn generative-AI FAQ describes regional differences and additional objection routes. Check your profile's current setting, region notice, feedback practices, and public-visibility choices. Do not summarize this as “LinkedIn trains everything by default” or “one toggle stops all AI processing”; both are too broad.

Slack and Zoom

Slack says in its AI privacy principles that it does not develop generative-AI models using Customer Data, while also explaining that some predictive models and service improvements use data under its customer terms. Slack offers a workspace-level opt-out from global models through the current owner/admin contact route. This is different from turning off an individual AI feature.

Zoom's AI privacy guidance says Zoom does not use communications-like customer content to train Zoom's or third-party AI models. It also explains that data is processed to provide features and that feedback may be used to improve services. Administrators can configure AI features; that feature switch is not the same as a deletion request.

Website owners: `robots.txt` is a request to crawlers

If you publish a website, you can use `robots.txt` to express which documented crawlers may access which paths. Google explains that `robots.txt` manages crawling and is not a way to remove a URL from search or erase data already copied. Google's Google-Extended documentation says the token can control whether Google uses crawled site content for certain Gemini training and grounding purposes without affecting Google Search ranking.

For example, a site owner could publish a rule for the Google token:

User-agent: Google-Extended
Disallow: /

Only add other vendor tokens when that vendor documents the token and its scope. A robots rule is not a technical barrier, does not govern user-controlled downloads or copied datasets, and cannot guarantee compliance by an unknown scraper. It also does not remove old training data. For removal from a specific service, use that service's current privacy or copyright route where applicable.

What data brokers can and cannot prove

People-search sites and data brokers may publish names, addresses, phone numbers, relatives, or other profile fields. That does not prove that every profile is collected for AI training, that a named AI company obtained it from a particular broker, or that removal will change a model's output. Public visibility creates exposure, but a causal chain requires evidence about the source and the service.

You can review the exact provider listing and submit a provider-specific request where an appropriate route exists. OfflistMe creates browser-local, user-reviewed drafts for supported workflows; you submit each request yourself, and provider coverage and requirements vary. Treat this as one exposure-reduction measure, not an AI untraining guarantee.

A practical review checklist

1. Identify the exact service, account tier, region, workspace, and feature.
2. Read the provider's current data-use page and the explanation beside the setting.
3. Turn off the narrow training or improvement control that applies to your account.
4. Delete stored activity separately when the provider offers a deletion control.
5. Review feedback, connected apps, public posts, uploaded files, and administrator policies.
6. Save the setting confirmation and the date you changed it.
7. Recheck the provider after material policy or product changes.

Frequently asked questions

Does opting out remove data already used to train a model?

Not automatically. An opt-out normally controls future collection or use within the stated program. It does not by itself establish that stored logs were deleted or that existing model behavior was changed. Read the provider's deletion and unlearning disclosures separately.

Does deleting a chat delete every copy?

No universal answer applies. A provider may have activity, backup, abuse-monitoring, feedback, connected-service, or legal-retention systems with different rules. Use the provider's current deletion documentation and save the response.

Does `robots.txt` stop AI companies from using my website?

It can communicate a crawler preference to a documented bot that follows the protocol. It is not a guaranteed block, does not control every scraper, and does not erase content already collected.

Is a business or enterprise plan automatically private?

No plan label is a substitute for the current terms. Some business products have no-training commitments, while administrators, connected services, logs, feedback, and other product features can have separate rules. Verify the exact product and contract.

How do I opt out of AI training data on Reddit?

Reddit's official developer guidance says Reddit content cannot be used as an input for model training without Reddit's explicit consent. That is a platform and developer rule, not an account-deletion control. To remove your own Reddit content, use Reddit's current post, comment, or account-deletion tools separately and do not assume that deleting an account removes copies already retained elsewhere.

Source: Reddit's developer-platform guidance and account-deletion help.

How do I opt out of AI training in Gemini?

Google's current Gemini Apps guidance says to turn off Keep Activity in Gemini Apps Activity. Future chats then do not appear in that activity and are not used to train Google's AI models unless you choose to send feedback, although Google can retain them for up to 72 hours for service, feedback, and safety. Work or school accounts can have administrator-controlled settings.

Source: Google's Gemini Apps activity guidance.

How do I opt out of AI in Gmail?

There is no single Gmail-only switch that governs every Gemini or Workspace processing path. For personal Gemini Apps use, review connected apps, turn off Keep Activity, and delete Gemini Apps activity separately when appropriate. For a work or school account, your Workspace administrator may control the relevant setting and retention.

Source: Google's Connected Apps guidance and Workspace account guidance.

How do I opt out of AI in Google Drive?

First identify whether you are using Gemini in Drive under Google Workspace or connecting Drive to personal Gemini Apps. Workspace data is covered by Workspace-specific commitments, while a personal Connected Apps flow has separate Gemini activity and training controls. Deleting a Gemini in Drive conversation does not necessarily delete information saved in Gemini Apps Activity, so review both layers and any administrator policy.

Source: Google's Gemini in Workspace guidance and Connected Apps guidance.

How do I opt out of AI training data on GitHub?

For a personal GitHub Copilot plan, open GitHub Copilot settings and set Allow GitHub to use my data for AI model training to Disabled. GitHub documents different data-use terms for Business and Enterprise, and organization policies, cloud agents, extensions, and other GitHub features can have separate scopes. Check the setting and current policy for the account and product you actually use.

Source: GitHub's Copilot policy guidance.

How do I opt out of ChatGPT training?

For a personal ChatGPT account, open Settings → Data Controls and turn off Improve the model for everyone. OpenAI describes this as a control for new conversations within the stated model-improvement program; it does not delete old chats or remove personal information from other services.

Source: OpenAI's Data Controls guidance.

How do I opt out of Adobe AI training?

Adobe's current Content Analysis FAQ says Adobe does not analyze content to train generative AI models unless you choose to submit content to the Adobe Stock marketplace. If your account shows a Content Analysis control, review and change it to limit analysis for product improvement; that setting is not a universal control for every Adobe feature, service, or Adobe Stock licensing decision.

Source: Adobe's Content Analysis FAQ.

How do I disable Google AI training?

Google does not provide one control for every AI-enabled product. Personal Gemini users should review Keep Activity, connected apps, and separate deletion controls. Google Search services now also expose Search Services History for some accounts as the setting rolls out; Google says turning it off prevents future Search Services History activity from being used to train generative AI models unless you provide feedback, but it does not automatically delete past history. Gmail and Workspace smart-feature settings are separate product or administrator controls. Website owners can use the documented Google-Extended token in `robots.txt` for the crawler scope Google describes. None of these routes proves that copied content or prior model behavior was erased.

Source: Google's Gemini activity guidance, Search Services History guidance, Gmail and Workspace smart-feature controls, and Google-Extended crawler documentation.

Related guides

• Remove Personal Data from ChatGPT
• Manage and Delete Personal Data in Google Gemini
• How to Remove Your Data from Data Brokers
• What Is a Data Broker?
• OfflistMe Opt-Out Directory