PCMag reviewed and compared the privacy policies behind today’s most popular AI apps, and the findings show clear differences in how aggressively they collect data. One platform in particular stands out for gathering the least.
ALSO READ: Trump got a priceless gold clock from the Rolex CEO, and then he cut Switzerland’s tariffs to 15%
AI chatbots are now embedded in daily workflows, powering everything from school assignments and work meetings to personal projects and routine searches. But while the convenience is undeniable, the privacy trade-offs remain far less clear. After reviewing the privacy policies and data-collection disclosures for ChatGPT, Copilot, DeepSeek, Gemini, and Qwen, PCMag found that the gap between these platforms is wider than most users may realize.
The reality is simple that many AI chatbot apps collect sizable amounts of user information, sometimes more than users knowingly give. And, in several cases, that data can end up shared with advertisers, third-party partners, or even governments, depending on the jurisdiction.
As PCMag notes in its review, “If the service is free, you’re probably the product, not the customer.” That dynamic holds especially true in the emerging world of AI apps, where entire business models depend on massive data flows.
How invasive are today’s most popular AI chatbots?
While comparing the privacy policies, PCMag found a pattern similar to its earlier examination of TikTok alternatives Lemon8 and RedNote: apps that appear harmless on the surface often gather surprisingly broad sets of data. That’s why its is routinely recommended to use browser-based versions of major services whenever possible. Installed apps tend to collect far more information, and AI chatbots are no exception.
ALSO READ: Spotify not working? Users report widespread outage, as they “can’t even open it”
After reviewing the App Store privacy labels for ChatGPT, Copilot, DeepSeek, Gemini, and Qwen, PCMag identified meaningful differences. Some chatbots only collect basic identifiers and app-usage data. Others, such as Google’s Gemini, pull in an expansive catalog of user information.
Gemini’s reported data intake includes browsing history, contact lists, emails, photos, precise location, search history, texts, and videos. In PCMag’s words, “It seems a bit much,” especially when compared with Qwen, which claims to gather only device IDs and interactions. But even that comes with caveats. Qwen’s privacy report doesn’t fully align with its written privacy policy, and because app-store privacy labels are self-reported, users have no independent verification.
This lack of oversight runs across every chatbot reviewed. Neither Apple nor Google audits the accuracy of these privacy disclosures, meaning users must rely on the honesty of the companies behind the apps.
Which AI chatbot has the strongest privacy protections?
To get a more accurate read on what data is actually collected, PCMag turned to the companies’ own published privacy policies. Here again, the differences were striking.
DeepSeek and OpenAI offer more concise, readable policies. Qwen’s policy is short as well, though PCMag notes several typos, not a reassuring sign when dealing with something as critical as data protection.
ALSO READ: New poll delivers big blow to Trump as approval rating takes sharp dive
One policy stood out for a different reason: Google’s Gemini. Its privacy documentation is extensive, clear, and upfront about the scope of its data collection. The company openly states that it processes user audio and allows people to delete or disable Gemini activity history — something PCMag recognized as a helpful step. Google also acknowledges that human reviewers may read Gemini chats to evaluate quality and safety.
The policy states that human reviewers “assess if Gemini Apps’ responses were low-quality, inaccurate, or harmful. They also suggest better responses.” Users who want to opt out must turn off chat history entirely.
Even so, PCMag report reiterates an important point, “Given that all the apps above record and store your prompts, it’s best to refrain from sharing secrets with any of these AI chatbots.”
Among all the platforms evaluated, Microsoft’s Copilot emerged as the strongest in terms of privacy. While its policy requires navigating several linked documents — because Microsoft 365 Copilot and Bing function differently, the outcome is clear. Copilot collects minimal customer data, does not share user information with advertisers, and does not feed user prompts into training datasets for large foundation models.
Copilot’s architecture is built around using a customer’s existing Microsoft 365 data to generate contextual responses. ChatGPT, by contrast, draws from and adds to a much broader public dataset, meaning user prompts may be used to train the model. Microsoft emphasizes that Copilot Studio meets federal and industry standards, including FedRAMP, HIPAA, and SOC certification. From a privacy standpoint, that places it well ahead of competing AI chatbot apps, as per a report by PCMag.
Who actually gets access to your data?
Concerns about data access aren’t limited to U.S. companies. As PCMag highlights, Chinese AI companies operate under rules that give the government broad rights over user data and intellectual property. This became a national conversation earlier in the year when TikTok faced a temporary U.S. ban over its data practices.
DeepSeek, which is based in China, also collects extensive data from U.S. users. Cliff Steinhauer of the National Cybersecurity Alliance told PCMag that DeepSeek’s rise should encourage broader global discussions around AI privacy. He stated that, “Chinese AI companies operate under distinct requirements that give their government broad access to user data and intellectual property.”
ALSO READ: What does 67 mean, who made the 67 meme and why is it so popular?
PCMag notes that this issue doesn’t only apply to China. Many U.S.-based AI apps gather large amounts of customer information as well, reflecting an industry-wide problem rather than a country-specific one.
How can users stop chatbots from collecting their information?
For users hoping to reduce data exposure, PCMag offers a straightforward solution: avoid mobile apps whenever possible. Instead, run AI models locally on a personal computer. It requires capable hardware, but tools like Ollama allow users to operate language models entirely offline. Local models such as DeepSeek R1 provide more privacy because no personal information leaves the device.
Another privacy-friendly option is using AI tools already embedded in new PCs, many of which process data directly on-device without the need for cloud connections. PCMag notes that one editor even created a personalized chatbot using Nvidia’s Chat With RTX tool, demonstrating that private AI use is possible without relying on data-hungry external platforms.
FAQs
Which chatbot collects the least data??
Microsoft Copilot currently collects the least, based on its own privacy policy.
Are app store privacy labels reliable?
Not fully. App Store and Play Store labels are self-reported and unverified.