Vibe and ChatGPT rank as the most privacy-friendly platforms, while Copilot, Meta AI, and Kimi finish at the bottom
Incogni, a leading personal data removal service and data privacy company, today released its Gen AI and LLM Data Privacy Ranking 2026, a comprehensive analysis of how 13 popular generative AI platforms collect, use, and share personal information. The research found that some of the biggest names in AI also carry some of the greatest privacy risks, with Microsoft Copilot, Meta AI, and Kimi receiving the three highest overall risk scores.
At the other end of the ranking, Mistral AI’s Vibe received the lowest overall privacy-risk score, followed closely by ChatGPT. The results suggest that popularity and resources do not necessarily translate into stronger privacy practices for users.
Incogni researchers evaluated ChatGPT, Claude, Gemini, Grok, Vibe, Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi, and Copilot across 11 weighted criteria. Researchers examined what happens to conversations and other user data, whether people can opt out of having chats used for AI training, how clearly companies explain their privacy practices, and the amount and types of personal information they collect and share.
Also Read: How Conduent’s Google Cloud Partnership Signifies a Turning Point for Generative AI
Other key findings include:
- ChatGPT had the most transparent privacy practices in the study. OpenAI provided the clearest information about how user data is handled and had the most approachable privacy policy and supporting resources among the companies reviewed.
- No platform allows users to remove data once it has already been used to train a model. Opting out can prevent future conversations from being used, but it does not remove information already incorporated into training.
- Nine of the 13 platforms provide a simple toggle that can stop conversations from being used for training, although the ease of finding and using these controls varies. Other platforms require users to submit requests, send emails, or navigate less clearly defined processes.
- None of the providers disclosed the exact datasets used to train their models. Most rely on broad descriptions such as publicly available information, private datasets, or social media data, making it difficult for people to know whether their personal information has been included.
- AI apps can collect information well beyond what users type into a chatbot. Among the 12 mobile apps analyzed, seven collected email addresses and five collected phone numbers. The iOS apps for Gemini and Meta AI also disclose the collection of sensitive data categories.
- Privacy policies remain difficult for the average person to understand. Every company studied had a privacy policy estimated to require a college-graduate reading level.
“People are putting more of their lives into AI tools, whether they’re asking a quick question, working through a personal issue, or using them on the job, but it can still be incredibly difficult to figure out where that information goes,” said Darius Belejevas, CEO of Incogni. “A privacy setting can help, but it only addresses part of the picture. These companies may still be collecting information through their apps, other products, and external sources. Users deserve a much clearer view of what is being collected and what control they actually have over it.”
The level of privacy protection users receive can also depend on where they live. Nine of the 13 platforms maintain separate privacy disclosures for European Union users, which generally provide stronger controls or greater transparency. Kimi, Z.ai, Qwen, and Meta AI do not offer an equivalent EU-specific framework.
Compared with Incogni’s 2025 analysis, Mistral AI and OpenAI again finished near the top of the ranking. The results also emphasize a persistent problem among large technology companies: privacy policies spanning sprawling ecosystems of products and services can make it difficult for users to determine exactly which practices apply to the AI tool they are using.
For the 2026 study, Incogni researchers scored each platform from 0 (most privacy-friendly practices) to 1 (least privacy-friendly practices) across individual criteria. Those criteria were then weighted according to their importance to user privacy. Researchers reviewed privacy policies, supporting documentation, and disclosures from the Apple App Store and Google Play Store. Data for the study was collected from June 15 through July 6, 2026.
About Incogni
Incogni helps people take control of their data by removing their personal information from various sources, including data brokers and people search sites. Its simple, user-friendly service helps prevent personal information from being sold, reducing exposure to cybercrime, scams, identity theft, and unwanted marketing.
