AI Privacy Rankings Reveal ChatGPT Leads While Meta AI, Copilot Pose Highest Data Risks

Major AI Chatbots Ranked by Privacy Risks

A comprehensive study by data removal service Incogni has ranked 13 major AI platforms based on privacy risks they pose to users, revealing stark differences in how AI chatbots handle personal data2

2

. The "Gen AI and LLM Data Privacy Ranking 2026" report evaluated ChatGPT, Claude, Gemini, Grok, Vibe, Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi, and Copilot across three critical areas: data usage practices, transparency of privacy policies, and personal data collection methods2

2

.

The findings show that larger AI platforms generally pose greater privacy risks, with one notable exception. Vibe and ChatGPT emerged as the least invasive options, while Microsoft Copilot received the worst privacy score, followed closely by Meta AI and Kimi2

2

4

4

. ChatGPT particularly excelled in transparency and clarity of its privacy policies4

4

.

How AI Chatbots Use Your Personal Data for Training

By default, many AI chatbots use conversations and user prompts to train their AI models, creating detailed user profiles from every interaction1

1

. This data is stored on parent company servers like those operated by OpenAI and Google, potentially allowing sensitive information to be regurgitated in responses to other users asking similar questions1

1

.

The privacy risks of AI extend beyond explicit data sharing. AI platforms infer personal details from user prompts even when users never directly provide such information3

3

. These inferences can inform targeting or profiling activities without user knowledge or permission3

3

. A critical limitation exists across all platforms: no AI chatbots allow users to retrieve their data once it has been used to train a large language model, and no retroactive removal or withdrawal of consent is possible2

2

.

Steps to Stop AI Chatbots from Training on Your Data

Eight AI platforms offer straightforward opt-out mechanisms: ChatGPT, Claude, Copilot, DeepSeek, Grok, Perplexity AI, Pi, and Vibe all provide simple toggles to stop data collection for AI training4

4

. For ChatGPT, users must navigate to Settings, click Data controls, and toggle off "improve the model for everyone"1

1

. In Claude, the process involves accessing Privacy settings and disabling "Help improve our AI models"1

1

.

Gemini requires disabling multiple settings across both Google account preferences and the Gemini interface itself. Users must turn off Web & App Activity at myactivity.google.com and separately disable Gemini Activity settings1

1

. However, Gemini's opt-out comes with a significant trade-off: invoking it disables the option to have Gemini remember conversations4

4

.

Meta AI presents the most challenging case for users seeking to opt out of data collection. There is no guaranteed method to stop Meta from using your data for AI training, and users must submit a form that Meta may or may not act upon1

1

. Z.ai offers no opt-out option whatsoever, while Meta, Moonshot AI, and Qwen pay only lip service to the opt-out process without making it readily available4

4

.

What Lumo AI Paper Trail Reveals About Your Data

Proton's free Lumo AI Paper Trail tool exposes exactly what AI platforms know about users by analyzing exported data from ChatGPT and Claude accounts3

3

. The tool processes user data in minutes, generating detailed reports showing personal information across categories including location, education, interests, profession, age range, and political leanings3

3

.

Lumo identifies red flags in data collection and explains how AI models inferred personal details even when users never explicitly typed them in3

3

. The service is built on open-source components, doesn't rely on other LLMs, and doesn't share user data with third parties3

3

. Users can access basic audits for free, with paid subscriptions at $9.99 per month annually or $12.99 monthly removing limits on daily chats and file uploads3

3

.

Data Handling Practices and Transparency Issues

Incogni's research evaluated AI platforms across 11 criteria using a weighting system to emphasize the most important factors4

4

. The evaluation examined whether platforms use user prompts for training, how easy it is to opt out, whether they share information with third parties, accessibility of privacy policies, and what types of personal data mobile apps gather4

4

.

Copilot, Meta AI, and Kimi all failed at providing clear privacy policies specifically for LLM activities4

4

. Kimi maxed out on demerits for data handling practices related to AI training, while Copilot scored worst for direct handling of personalized data4

4

. A middle group consisting of Claude, DeepSeek, Gemini, Grok, Perplexity, Pi, Qwen, and Z.ai fell between the best and worst performers4

4

.

Deleted chats are not immediately removed from backend systems. Both Anthropic and OpenAI store deleted conversations for up to 30 days1

1

. Even after users disable data sharing settings, new and existing chats won't be used for future model training, but data will remain in training already in progress and won't be removed from models already trained1

1

.

Protecting Your Privacy When Using AI Platforms

Experts recommend treating any information entered into AI chatbots as data given to a third-party service2

2

. Avoid sharing passwords, financial details, confidential work information, medical records, or other identifying information2

2

. Where possible, remove names or other identifiers from prompts before submitting them2

2

.

For sensitive tasks, consider using web versions rather than mobile apps, pay attention to provider privacy practices, and where appropriate, use locally run models that keep data on-device2

2

. New security concerns emerge weekly, with recent incidents including accidental indexing of Claude chats2

2

. User privacy is not guaranteed with any AI model currently available, making vigilance essential when choosing which platforms to trust with personal information.