A VPN company has analyzed the data collection practices of the ten most popular AI chatbots in the App Store, ranking them from worst to best. Using Apple's required privacy disclosures, Surfshark evaluated how many types of personal data each app collects. The findings show that AI chatbots collect an average of 14 out of 35 possible data types, with 70% tracking users' locations.

Meta AI emerged as the worst offender, collecting 33 out of 35 possible data types - nearly 95% of all available personal information. This includes financial data, making it the only app in that category. Meta AI also gathers sensitive information like racial or ethnic data, sexual orientation, pregnancy information, disability status, religious beliefs, political opinions, genetic data, and biometric information. Google Gemini ranks second-worst, collecting 23 unique data types including precise location data shared only with Meta AI, Copilot, and Perplexity.

The analysis reveals that free AI chatbot services treat users as products rather than customers. Even paid subscriptions don't reduce data collection, and many apps use chat interactions as training data. The safest option is using Siri's fallback to ChatGPT, where Apple has negotiated anonymized queries that cannot be used for training. This same protection is expected when the new Siri integrates with Google's Gemini.