A Surfshark study reveals startling privacy concerns as leading AI apps, including Meta AI and Google Gemini, collect vast amounts of user data.
Artificial Intelligence chatbots have quickly become indispensable companions in everything from writing assistance to personal productivity. But at what cost? A newly published Surfshark report has sounded the alarm bells for privacy advocates, revealing that some of the most popular AI-powered chatbot applications are harvesting up to 90% of possible user data types.
Who’s Watching You? A Data Dive into Popular AI Apps
Conducted by Surfshark’s leading system engineer Karolis Kaciulis, the analysis dissected the privacy practices of 10 major AI chatbot apps, including the recently launched Meta AI. By analysing Apple App Store disclosures and privacy policies, the study compared how much personal data each app collects, what it’s linked to, and whether the data is used for advertising or sold to third-party brokers.
“We identified major disparities in how these apps treat user data, and it’s shocking how deep some go,” said Kaciulis.
Here’s a breakdown of what was discovered:
- Meta AI, which was added to the study on May 20, 2025, topped the list by collecting 32 out of 35 possible data types — that’s over 90%.
- Google Gemini came in second, gathering 22 unique data types, including precise location — a data point only collected by Gemini, Meta AI, Copilot, and Perplexity.
- ChatGPT appears more privacy-conscious, collecting only 10 types of data, such as contact information and usage data. Crucially, it does not track users or rely on third-party ads.
- Copilot, Poe, and Jasper are the most aggressive in tracking users for advertising and data resale purposes.
Location, Identity, and Everything In Between
The data collection wasn’t just technical. Surfshark’s report highlights that 45% of the examined AI chatbots collect location data, while 30% track users for advertising and profiling. These stats bring renewed attention to the silent trade-off between convenience and control.
Apps that collect “data linked to you” may also process identifiers, search history, user-generated content, and more — allowing them to build comprehensive user profiles that could potentially be sold or leaked.
Learning from Your Social Media? Yes, Really.
The report also warned that Meta AI’s learning model is fed by content from Facebook and Instagram, including users’ posts and images. This makes the app particularly invasive as it bridges private conversations with broader social media activity.
“These chatbots might be fast, but the output is often mediocre because they’re learning from a sea of both accurate and wildly inaccurate information,” added Kaciulis. “And unlike humans, they face no accountability for the content they produce.”
A Flawed System with No Legal Oversight
The study further referenced an incident involving Grok, the chatbot developed by X (formerly Twitter), which responded to unrelated queries by echoing white nationalist themes. This, Surfshark warns, underscores the unpredictability and lack of ethical oversight in current AI models.
In a time where AI is becoming deeply embedded in everyday work, education, and personal interactions, the lack of transparency and legal accountability poses a growing concern.
What Can You Do?
If you’re using AI tools, consider the following steps to protect your digital footprint:
- Review app permissions and adjust data sharing settings.
- Use privacy-oriented chatbots like ChatGPT, which limit tracking and ad usage.
- Be cautious about what you share, even in seemingly private AI chat sessions.
- Consider VPNs and other digital tools to anonymise your online behaviour.
Final Word: The Price of Convenience
AI-powered tools are here to stay — but the Surfshark report is a sobering reminder that the free help you receive might be anything but free. Behind the curtain of seamless answers and auto-generated content lies a vast system of user profiling, behavioural tracking, and data monetisation.
The more aware users become, the better equipped they’ll be to demand privacy-first AI.
🔒 Play smart. Stay private. Choose your AI tools wisely.
Also read: 3 AI Tools to Boost Your Online Income in 2025 – No Tech Skills Required
