How to manage personal data across AI services (2026)
At a glance
- Training preferences, stored-content deletion and response requests are separate actions.
- Controls depend on the product, account and region; check exceptions before sharing personal data.
- Keep request confirmations and provider replies. Submission alone does not establish removal.
What you need to know first
Training preferences, deleting saved content, and requesting action on an AI response are separate tasks. Product, account, region and exceptions determine what each control covers. An answer about you does not prove your data was used in training.
Choose your task
Each task below uses a different route. Pick the one that matches what you want; you can do more than one.
Stop future use of your chats for training
Use the product’s training or model-improvement setting, where your product and account offer one. It covers future training under that provider’s rules. It doesn’t delete saved chats or undo completed training. Use steps 1 and 2 in the step-by-step list below.
Check the outcome: Reopen the setting and confirm it’s saved. Note the date and the scope the provider states, including exceptions such as feedback.
Delete saved chats, files or memories
Use the product’s own chat, file, memory and account deletion controls. Each can have its own retention period and exceptions. Use step 3 in the step-by-step list below.
Check the outcome: Confirm the item no longer appears in your account and keep any deletion confirmation. Deletion from the provider’s systems follows its stated timeline and exceptions, which your account view doesn’t show.
Report a response that contains personal data about you
Use the provider’s privacy or personal-data request route. It covers what the service says, not the pages a response cites. Use steps 4 and 5 in the step-by-step list below.
Check the outcome: Keep the provider’s reply and recheck the response later, recording the prompt and date each time because answers can vary. A request or a reply alone doesn’t establish that the response changed.
Remove a public page or listing
Ask the site that publishes it, through its own removal or privacy route. For people-search sites and data brokers, the privacy request kit helps you choose a route, and the blank tracker is a file you fill in yourself to record each request. Neither covers AI providers, and the tracker doesn’t send anything. Use step 6 in the step-by-step list below.
Check the outcome: Check the page itself after the site replies. Removing a page reduces exposure there; it doesn’t show that a model changed.
Worked example: one response, three routes
Alex is a fictional person. Alex asks an AI service what it knows about Alex. The response cites a people-search page with an old home address, repeats a project Alex described in an earlier chat in their own account, and names an employer Alex never worked for, with no source.
| Part of the response | What it is | Evidence to keep | Next step |
|---|---|---|---|
| The cited people-search page | A public page the response links to | The page URL, the response showing the citation, and the date | Ask the people-search site to remove the listing, then recheck the response. Removing the page doesn’t show that a model changed. |
| The project from an earlier chat | Saved content in Alex’s own account, such as an earlier chat or a memory | Which chat or memory it came from | Delete that chat and any memory made from it, and review the training setting for future chats. |
| The wrong employer | An inaccurate response with no cited source | The prompt, response, date and product | Report it through the provider’s privacy request route and say what’s wrong. Some providers say a correction may not always be possible. |
The three routes don’t depend on each other. Alex can use any of them, in any order.
Step-by-step: manage your AI service data
- Step 1: Identify the AI product, account type and region you use; read its current privacy documentation
- Step 2: Review the saved training preference and any separate rules for feedback, safety review or temporary use
- Step 3: If you want to remove stored content, check chat, file, memory and account deletion controls separately; retain confirmations
- Step 4: For a response containing personal data, save the prompt, response, date, product and cited URLs, then follow the provider’s privacy request process when available
- Step 5: Provide only the evidence and verification requested through the official channel; record the request scope and provider reply
- Step 6: Address any public source separately and recheck the specific source or response; submission alone is not a verified removal
Opt out of AI training
Some products offer training preferences for eligible account content. Check availability and scope for your product and account. Feedback, safety review, public content and commercial products may have separate rules; there is no universal toggle across providers.
Retention and deletion limits
There is no single deletion deadline across AI services. Retention depends on the product, data type and exceptions. Review the relevant provider guide before deleting content.
A training preference does not delete stored chats, files or memories. Stored-content deletion follows provider policies and exceptions. Neither action alone establishes a change to previously completed training, and an output concern needs a separate response from the provider.
Separate data request
Use the provider’s official privacy documentation to find the route for your concern. A privacy portal, account setting, regional objection and form about third-party personal data in a response can cover different information. Keep the confirmation and provider reply.
Key gotchas
- Treating a training preference as deletion of saved content
- Treating a submitted request as a verified removal
- Assuming a paid account, feedback or temporary chat follows the same rules as another product
- Assuming an answer about you proves training-data inclusion
- Using an account or regional form for a different product without checking its scope
What this does not cover
Provider controls do not remove unrelated broker listings or other public pages. If an AI response cites a page, check that source and seek removal there separately. The response alone is not proof that a model trained on that page.
Broker removal can reduce personal data available from those public listings. It does not undo completed model training, establish that a model used a particular listing, or replace a provider's output-removal process.
See what personal data is exposed about you beyond AI Products (General).
Run a free scan →Frequently asked questions
Can I opt out of AI training?
Some services offer training preferences for eligible account content. Availability and exceptions depend on the product, account and region. A conversation setting does not necessarily cover public web data or feedback.
Does a training opt-out delete my chats?
Training permission and stored content are separate controls. Check the provider’s chat, file, memory and account deletion options, including retention exceptions.
What if an AI response contains personal data about me?
Keep the prompt, response, date, product and any cited URL. Follow the provider’s personal-data or privacy request process when available. Requests are reviewed under its policy and applicable law; submission does not establish acceptance or removal.
Does removing a broker listing remove it from AI models?
Removing a listing can reduce exposure at that source. It does not prove that a model trained on the listing, undo completed training or remove other copies. Follow up separately on a specific AI response.
Sources
- OpenAI consumer choices, feedback and commercial scope
- OpenAI development sources and personal-data requests
- Claude consumer choice, feedback and safety exceptions
- Gemini activity, feedback and deletion distinctions
- Perplexity training preference and continued processing
- Copilot app-version and account boundaries
- Meta’s documented EU objection route
This checklist reflects the linked provider documentation; signed-in interfaces and individual outcomes were not tested. Platforms may change their settings layout at any time. This guide is informational and not affiliated with the AI providers listed here.