AI chatbots are now being questioned to summarize a document, prepare an email, brainstorm an idea, or solve an everyday problemThis practice sometimes gives the impression of a private exchange. However, not all AI applications report the same data collection practices. A study by Surfshark compares the main tools available on the App Store, with a clear observation. They all collect personal data, but not in the same proportions.
Meta AI and Gemini, champions of personal data collection
Before looking at the rankings, one point needs to be addressed. The data gathering It does not only concern messages sent to a chatbot. It also encompasses information related to the account, device, location, shared content, or service usage.
To create its comparison, Surfshark relied on the App Store privacy policies of the most popular AI chatbots. The study was then supplemented with additional privacy policies, notably those of ChatGPT and DeepSeek. In total, 35 categories of personal data were examined.
In this ranking, Meta AI comes out on top by a wide margin. The tool claims to collect 33 out of 35 types of data, covering almost the entire scope of the study. It is also the only one in the panel to mention financial data. At this level, we're no longer talking about just an assistant that processes your queries, but a service capable of compiling a much broader profile.
Google Gemini It occupies second place, with 23 types of data collected. Its case stands out due to the variety of information involved. Surfshark specifically mentions precise location, contact details, user content, search history, and browsing history. For a tool linked to the Google ecosystem, this diversity is noteworthy, especially if you use Gemini for personal or professional purposes.
The most sensitive point concerns the nature of the data. Meta AI and Gemini are among the services declaring that they collect sensitive information, such as political opinions, religious beliefs, sexual orientation, certain biometric data or health-related information.
ChatGPT, Claude, DeepSeek and others display more nuanced profiles.
Le OpenAI chatbot, which has largely popularized conversational AI among the general public, nevertheless ranks third. After the latest updates to its data collection practices, ChatGPT It now reports 17 types of personal data out of 35, compared to 10 in the previous analysis. The additions include approximate location, health and fitness data, search history, audio data, advertising data, and customer support.
Behind it, Claude and DeepSeek are further down the rankings, with 13 types of data each. The former maintains a more stable profile, primarily related to service operation, security, authentication, and support. DeepSeek raises another issue, since Surfshark indicates that information is stored for as long as necessary on servers located in China. A data breach has already exposed more than a million records, including... conversation history and access keys.
Since the study does not stop at the most well-known names, it is worth seeing where the others stand.
- Poe collects, among other things, contact information, identifiers, usage data, and user content.
- Microsoft Copilot adds a sensitive point with precise location tracking.
- Perplexity collects fewer categories, but also remains associated with precise location.
- Grok appears primarily in the context of contact information, diagnoses, and identifiers.
- Pi remains in the lower part, with a more limited collection
- Jasper AI closes the ranking with a profile focused on diagnostics, identifiers, and usage data.
How to choose an AI chatbot without exposing too much of your personal data
You absolutely can integrate artificial intelligence into your routine without sacrificing your privacy. Protecting your information simply requires adopting strict digital hygiene practices. Follow these recommendations to secure your daily communications.
- Remove any mention of medical data, financial results, or trade secrets from the body of your texts.
- Take the time to replace real names, physical addresses, and phone numbers with pseudonyms before sending an instruction.
- Control the permissions granted to each application from your smartphone's settings to revoke access to the microphone, contacts, and GPS module.
- Disable history saving in the chatbot's internal settings to prevent developers from reusing your requests
- Choose your tool based on the complexity of the task and prioritize the secure web version rather than the mobile app often more indiscreet
source: Surfshark
