How to Remove Yourself from OpenAI and Other AI Training Data

The AI Training Data Problem You Did Not Consent To

Large language models like ChatGPT, Gemini, and Meta AI were trained on massive datasets scraped from the internet. That includes personal information: your name, your social media posts, your forum comments, your published work, and in many cases, your private data purchased or scraped from data brokers.

A 2023 study by researchers at ETH Zurich found that GPT-3.5 could accurately recall personal email addresses for over 80% of tested subjects. A separate study by UC Berkeley demonstrated that language models can be prompted to reveal training data including phone numbers, addresses, and social media handles.

You did not consent to this. But you do have options.

Your Legal Rights Regarding AI Training Data

GDPR Article 17: Right to Erasure

If you are an EU/EEA resident or if an AI company processes your data in the EU, GDPR Article 17 gives you the right to have your personal data erased. This applies to AI training data, though enforcement is still evolving.

Key provisions:

• Article 17(1)(b): You can withdraw consent and request erasure if there is no other legal basis for processing
• Article 17(1)(d): Data that has been unlawfully processed must be erased
• Article 6(1)(f): Companies often claim "legitimate interest" as a basis for training, but this can be challenged

The Italian data protection authority (Garante) temporarily banned ChatGPT in 2023 over GDPR violations, and OpenAI subsequently added opt-out mechanisms for European users.

CCPA Section 1798.105: Right to Delete

California residents can request deletion of personal information held by businesses, including AI companies. Under the California Consumer Privacy Act:

• You have the right to request deletion of your personal information
• Businesses must comply within 45 calendar days
• This right extends to information used for AI model training, though the practical implementation is complex

The "Unlearning" Problem

A critical technical challenge exists: once personal data has been used to train a model, it is extremely difficult to truly "remove" that data from the model's weights. Companies typically address deletion requests by:

• Removing the data from future training datasets
• Filtering the data from model outputs
• Fine-tuning models to reduce the likelihood of reproducing the data

True "machine unlearning" is an active area of research but is not yet practical at scale. This means that exercising your opt-out rights today primarily prevents your data from being used in future training runs and reduces its appearance in model outputs.

How to Opt Out: Company by Company

OpenAI (ChatGPT, DALL-E, GPT-4)

Step 1: Disable Chat History Training

• Log in to ChatGPT at chat.openai.com
• Click your profile icon in the bottom left
• Go to Settings > Data Controls
• Toggle off "Improve the model for everyone"
• Note: This prevents future conversations from being used for training. Past conversations may have already been used.

Step 2: Submit a Data Deletion Request

• Visit privacy.openai.com
• Click "Make a Privacy Request"
• Select "Delete my personal information from OpenAI's training data"
• Provide your information so they can locate your data
• Submit the request

Step 3: Opt Out of Web Scraping (for website owners)

Add the following to your site's robots.txt file:

```

User-agent: GPTBot

Disallow: /

User-agent: ChatGPT-User

Disallow: /

```

Response time: OpenAI states they will respond within 30 days (45 days for CCPA requests).

Google (Gemini, Bard, Search Generative Experience)

Step 1: Manage Gemini Activity

• Go to myactivity.google.com
• Select "Gemini Apps" from the left sidebar
• Click "Turn off" to stop saving Gemini activity
• Delete existing activity by clicking "Delete" and selecting "All time"

Step 2: Opt Out of AI Training

• Visit Google's Privacy Hub at myaccount.google.com/privacy
• Navigate to "Data & Privacy" > "History Settings"
• Turn off "Web & App Activity" for Gemini
• For broader AI training opt-out, use Google's data deletion form at support.google.com/accounts/troubleshooter/6358155

Step 3: Block Google-Extended (for website owners)

Add to your robots.txt:

```

User-agent: Google-Extended

Disallow: /

```

This blocks Google from using your website content for AI training while preserving regular search indexing.

Meta AI (LLaMA, Meta AI Assistant)

Step 1: Opt Out of AI Training on Facebook/Instagram

• Open Facebook or Instagram
• Go to Settings > Privacy > "AI at Meta"
• Look for "Generative AI Data Subject Rights"
• Select your country/region and submit an objection form
• Explain that you object to your data being used for AI training

Step 2: Submit a Formal Objection (EU/UK residents)

• Visit Meta's "Your Information and AI at Meta" page
• Submit a Right to Object form under GDPR Article 21
• Meta is legally required to respond within 30 days

Important note: Meta's opt-out process has been criticized by European regulators for being unnecessarily complex. If your initial request is denied, escalate by filing a complaint with your national data protection authority.

Anthropic (Claude)

• Visit anthropic.com/privacy
• Anthropic states that conversations with Claude are not used to train models by default for API customers
• For consumer Claude.ai users, prompts may be used for safety research but can be opted out via Settings
• To request deletion of personal data, email privacy@anthropic.com with your request

Microsoft (Copilot, Bing AI)

Step 1: Manage Copilot Data

• Go to account.microsoft.com/privacy
• Navigate to "Activity History"
• Clear your Copilot conversation history
• Under Privacy settings, review and limit data sharing

Step 2: Submit a GDPR/CCPA Request

• Visit microsoft.com/en-us/concern/privacy
• Select "Privacy" as your concern category
• Submit a data deletion request specifying AI training data

Step 3: Block AI Crawling (for website owners)

```

User-agent: Bingbot

Disallow: /ai-training/

```

Note: Blocking Bingbot entirely will remove your site from Bing search results. Microsoft does not currently offer a separate AI-specific crawler user agent.

The Hidden Pipeline: How Data Brokers Feed AI Training

While direct opt-outs from AI companies are important, they address only one part of the problem. A significant portion of AI training data comes not from direct web scraping but from data brokers and aggregated datasets.

How the Pipeline Works

• Data brokers collect your information from public records, social media, commercial transactions, and other sources
• Brokers sell aggregated datasets to AI companies, research institutions, and data resellers
• Datasets are incorporated into training corpora alongside web-scraped data
• AI models learn patterns from this data, including personal information
• Models can reproduce personal information when prompted in certain ways

The Common Crawl Connection

Most major AI models were trained at least partially on Common Crawl, a nonprofit that has been archiving the web since 2011. Common Crawl's dataset includes snapshots of data broker websites, people-search results, and other pages containing personal information.

When you search for your name on Spokeo or BeenVerified, that search results page may end up in Common Crawl's archive and subsequently in AI training datasets. This means your data broker profiles are likely embedded in multiple AI models.

Breaking the Pipeline at the Source

The most effective way to prevent your data from entering AI training pipelines is to remove it from data brokers. When your information is no longer published on data broker websites:

• Future web crawls will not capture it
• Future Common Crawl archives will not include it
• Future AI training datasets built from web data will not contain it

This is why data broker removal is not just about privacy from people-search sites. It is about cutting off the supply chain that feeds your personal data into AI systems, advertising networks, scam operations, and more.

GhostMyData removes your data from 1,500+ data broker sources, including the people-search sites and data aggregators most commonly scraped for AI training data.

What AI Companies Know About You (And How to Find Out)

Subject Access Requests

Under both GDPR (Article 15) and CCPA (Section 1798.110), you have the right to request a copy of all personal information a company holds about you. This includes information used for AI training.

How to submit a Subject Access Request:

• Identify the AI company's privacy contact (usually found in their privacy policy)
• Send a written request specifying that you want all personal information held, including training data
• Include enough identifying information for them to locate your data
• Reference the specific legal basis (GDPR Article 15 or CCPA Section 1798.110)
• The company must respond within 30 days (GDPR) or 45 days (CCPA)

What You Will Likely Receive

In practice, AI companies typically respond to subject access requests with:

• A copy of your account data and conversation history
• A statement about whether your data was included in training datasets
• A list of data sources used for training (in general terms)
• An explanation of how to opt out of future training

They will not provide a specific extract of "your data" from within the model, because that is not technically feasible with current architectures.

A Practical AI Privacy Action Plan

Priority 1: Stop the Bleeding

• [ ] Opt out of AI training on all platforms you use (OpenAI, Google, Meta, Microsoft)
• [ ] Disable chat history training on ChatGPT and similar tools
• [ ] Review and delete stored AI conversation histories

Priority 2: Block Future Collection

• [ ] Add robots.txt rules blocking AI crawlers on any websites you operate
• [ ] Remove personal information from data broker sites that feed AI training pipelines
• [ ] Start a free GhostMyData scan to identify and remove data broker exposures

Priority 3: Exercise Your Legal Rights

• [ ] Submit data deletion requests to AI companies holding your data
• [ ] File Subject Access Requests to understand what data is held
• [ ] If in the EU, consider filing complaints with your data protection authority for non-compliance

Priority 4: Ongoing Protection

• [ ] Monitor for re-listing on data broker sites (automated services handle this)
• [ ] Stay informed about AI privacy legislation in your jurisdiction
• [ ] Review AI companies' privacy policies when they update (they change frequently)
• [ ] Compare automated data removal services for continuous protection

Frequently Asked Questions

Can AI companies really delete my data from their models?

Not in the traditional sense. Once data has been used to train a model, it becomes embedded in the model's mathematical weights and cannot be surgically removed. However, companies can remove your data from future training datasets, filter outputs to prevent your information from appearing, and apply fine-tuning to reduce memorization of your data.

Is it legal for AI companies to use my personal data for training?

This is an active legal question. In the EU, several data protection authorities have found that AI training on personal data without consent violates GDPR. In the US, the legal landscape is less clear, but CCPA deletion rights apply to AI training data. Multiple class-action lawsuits are pending against major AI companies.

Does opting out of ChatGPT training affect past conversations?

No. Disabling "Improve the model for everyone" in ChatGPT settings only affects future conversations. Past conversations may have already been used for training. To address past data, you need to submit a separate data deletion request through OpenAI's privacy portal.

How do data brokers feed AI training?

Data brokers publish personal information on people-search websites that are indexed by web crawlers. These crawled pages end up in datasets like Common Crawl, which are used by AI companies for training. Additionally, some AI companies purchase datasets directly from data brokers or data aggregators.

Will removing my data from brokers remove it from AI models?

It will not remove data that has already been used for training, but it prevents your data from appearing in future training datasets built from web-scraped data. Over time, as models are retrained on newer data, the persistence of your personal information in AI outputs should decrease.

What about AI-generated images of me?

If AI models can generate images of you (typically this affects public figures), you can submit removal requests to the specific AI image generation service. GDPR Article 17 and CCPA Section 1798.105 apply to biometric data and likeness. Some states also have specific laws protecting likeness rights.

Related Reading

• What Is a Data Broker? Everything You Need to Know
• How Scammers Get Your Personal Information
• Compare Data Removal Services
• Start Your Free Privacy Scan