AI Privacy Study Ranks 13 Chatbots: ChatGPT, Gemini, and Meta AI Compared on Data Safety

The best and worst AI for your privacy, ranked - and how each handles your data

Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways * Incogni's new study of 13 AI platforms ranks them by privacy risk. * The largest AI platforms turn out to be the most privacy-invasive - with one exception. * Here's what to know about each option before signing up for an account. The use of generative AI platforms such as ChatGPT, Claude, and Gemini continues to change how we live and work. But do the privacy risks outweigh the convenience? AI engines have had a cataclysmic impact on many of our lives, comparable to that of smartphones. Businesses are pouring billions of dollars into AI-related developments; it seems like almost every organization wants to project an "AI-first" image, and even in our daily queries and quests for information, AI citations are replacing the older methods we used to search. Also: How to keep your conversations with ChatGPT, Gemini, Copilot, or Claude private We're asking AI for advice before making real-world decisions, and some of us are even seeking out recommendations on our personal lives and finances. (Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) But how safe is the data we feed into our AI companions, and do the organizations developing today's AI platforms consider user privacy to be their responsibility? If you've ever asked these questions, new research from Incogni sheds some light on them. Incogni's Gen AI and LLM Data Privacy Ranking 2026 report Published on Thursday, data broker removal service Incogni's new research analyzes 13 AI platforms and the potential risks they pose to consumer privacy. Also: AI cybersecurity attacks are the future: 43% of companies have already experienced it The "Gen AI and LLM Data Privacy Ranking 2026" report looks at these AI platforms: ChatGPT, Claude, Gemini, Grok, Vibe (formerly Le Chat), Perplexity, Qwen, DeepSeek, Z.ai, Kimi, Meta AI, Pi, and Copilot. In the study, each AI has been assigned a score "based on the privacy risks they pose," Incogni says. The lower the score, the better, as each point added reflects the discovery of privacy-invasive practices, how easy data-sharing practices are to find and understand, and what happens to user data. Three groups of questions are addressed by the study: * What happens to user data? Are user conversations fed into training platforms, can users opt out, and are prompts shared with third parties or other systems? * Are AI platforms transparent? Are data and privacy policies easy to find and understand? * What happens to data on AI platforms? What personal data is gathered, where does it come from, and if it is shared, where does it go? The larger the platform, the bigger the risk - with one exception Some of the largest AI platforms appear to pose the greatest risk to user privacy, except for OpenAI's ChatGPT. Gemini and Meta AI received some of the highest risk scores, whereas Vibe, ChatGPT, and Pi achieved some of the best (and lowest) risk scores in Incogni's assessment. In summary, this is how each AI model performed overall, ranked from lowest to highest risk: Which AI is the best for data privacy? Incogni's results and rankings do not mean that Vibe, ChatGPT, Meta, or Copilot are automatically the best or worst options for you. For example, no platform allows users to retrieve their data once it has been used to train a large language model, and no retroactive removal or withdrawal of consent is possible. Instead, you should look at these privacy-weighted scores as indicators of what some AI models are doing well, and what others need to improve. New security and privacy concerns are emerging week by week -- the accidental indexing of Claude chats being only one of many recent incidents -- and your privacy is not guaranteed with any AI model on the market today. Also: Google's AI can see your business data by default in Workspace - unless you disable it When you use any large language model, it's best to choose one with transparent data collection and handling policies, and be careful about what information you share with it. If the idea of your financial statements being used to train an AI isn't palatable, for example, then don't submit them -- and apply this concept to all of your future conversations. "Treat anything you enter into an LLM as information you are giving to a third-party service," Incogni's information security manager Miguel Fornés told ZDNET. "Avoid sharing passwords, financial details, confidential work information, medical records, or other identifying information, and remove names or other identifiers where possible. For more sensitive tasks, consider using the web version rather than a mobile app, pay attention to the provider's privacy practices, and, where appropriate, use a locally run model that can keep data on-device."

This Free Tool Reveals Every Creepy Secret ChatGPT Knows About You

(Credit: Zooey Liao/PCMag/IntelligentHQ/Chat GPT/Lumi/Getty Images) Ever wonder what AI chatbots actually know about you? Hint: It's far more than what they show in a single reply. Every prompt you type helps the model build a detailed, highly personalized profile -- which is great for tailored answers, but uncomfortable when you have zero visibility into what it's storing. That's where Lumo AI Paper Trail comes in (from one of our favorite VPN providers, Proton). Export your data from Claude or ChatGPT, feed it into Lumo, and you get details about exactly what the AI model knows. Not only that, the report explains how the model gained this knowledge The Price Tag: Free Audits, Paid Upgrades You don't have to pay anything to use the AI Paper Trail service. You'll quickly find that it's powered by -- you guessed it! -- another AI. Proton's Lumo AI is designed for privacy from the ground up. It's built on open-source components, doesn't rely on any other LLMs, and doesn't share your data with any third party. The free plan has restrictions, just like the free plans of most other AI platforms. Switching to a paid account removes limits on daily chats or favorites, for example, and lets you upload large files to Lumo. A paid Lumo subscription costs $12.99 per month, or $9.99 per month if you pay for a full year. Most competitors offer multiple subscription tiers. ChatGPT offers monthly subscriptions at $8, $20, and $100 or more. With Claude, the monthly tiers are $20, $100, and $200. Those using Grok without an associated X account pay $10, $30, or $300 per month. Once again, though, you don't have to pay a penny to use the AI Paper Trail service discussed here. How to Force AI to Hand Over Your Data Lumo can't just reach into your other AI chatbot accounts and hoover up their data. You have to give it a helping hand to get the analysis started. At present, Lumo's processing supports ChatGPT and Claude personal accounts (not business or enterprise accounts). You follow its instructions to export data from your account. And then you wait, and wait, possibly for several days, until the AI model emails your data as a big ZIP file. What Lumo Found in My Hidden Data: Fast, Detailed, and Uncomfortably Accurate When your big data package arrives in your email, it's time to feed it to Lumo. The page for uploading the data notes that "Your export is encrypted in transit and processed without being stored on Lumo's servers." In testing, the analysis finished in just a couple of minutes, and the results were surprisingly detailed. Alongside the report, Lumo lets you create a shareable data card with no personal information. It just shows your overall privacy score along with sub-scores for specific categories. Mine showed that I gave ChatGPT a lot of information about my location, interests, and work, but very little about health or relationships. It's free enough of personal information that I've shared my data card here. How ChatGPT Accurately Guessed My Life Story The real magic appears in the complete report, which, unlike the shareable card, is absolutely loaded with personal details. Mine started off describing me thus: "A longtime technology journalist or enthusiast based in the Bay Area with deep expertise in cybersecurity, vintage computing, and pop culture. The user exhibits high privacy awareness but frequently tests AI boundaries with complex image generation requests and specific local inquiries." Can't argue with that! Next, it showed a summary of personal information in six areas: Location, Education, Interests, Profession, Age Range, and Political Leanings. All the summaries proved accurate, or at least within spitting distance of accurate. For example, I don't live in Northern California, but I did live there for many decades and retain ties to the region. Lumo identified several red flags in this data collection. It explained that even if I never explicitly typed in these elements of personal data, the AI model inferred them. Further, its inferences could inform targeting or profiling activities, without my knowledge or permission. Interestingly, when I generated the report a second time using the same bolus of data exported from ChatGPT, it produced different details. Above, I mentioned a summary beginning "A longtime technology journalist..." The second time around, that summary was quite different: "A user who frequently tests AI capabilities across creative writing, technical troubleshooting, image generation requests, and factual verification, often pushing against content filters while demonstrating strong curiosity about history, cybersecurity, and pop culture." This helps hammer home the fact that Lumo doesn't save anything about your report. Scrolling down, I found a list of the same eight topics as on the shareable card, each with a short pop-up explainer. An adjacent column listed a series of the most revealing chat conversations. The next section of the report consisted of numerous inferences about my life. Most were spot-on, though there were some clunkers. For example, because I once asked ChatGPT to identify the background music in a dog food commercial, it concluded that I probably have a pet. I don't. The report finishes with some common-sense advice on how to use AI on your own terms. Check your AI's privacy settings and turn off features like allowing it to use your inputs for training. Delete any data and old conversations that you don't need. Don't overshare. And of course, consider switching to a privacy-focused AI system like Lumo itself. Time to Check What Your AI Knows About You If you're a regular user of ChatGPT or Claude, I highly recommend running your data through Lumo AI Paper Trail. And if you find an ocean of personal data exposed, maybe you'll be more cautious going forward. When you've typed a prompt or query into your AI buddy, stop and think before you send it: What does this question reveal about me? If you do proceed to ask your question, you'll at least be aware of the possible consequences.

Your AI Prompts Aren't Private: The Most (and Least) Invasive Chatbots, Ranked

You can ask an AI chatbot anything, but what do your questions and prompts say about you? More importantly, how much of your privacy are you giving away when you converse with ChatGPT or Claude? The research team at Incogni, a well-known purveyor of personal data removal services, set out to quantify the privacy risks of using popular LLMs, identifying which ones take the most care with your privacy -- and the least. 13 Popular Chatbots Put on Trial Incogni's researchers selected 13 AI platforms for analysis. They focused on companies whose primary product is AI, as well as major players like Microsoft and Meta. Here are the platforms they evaluated: * ChatGPT (OpenAI) * Claude (Anthropic) * Copilot (Microsoft) * DeepSeek (DeepSeek) * Gemini (Google) * Grok (xAI) * Kimi (Moonshot AI) * Meta AI (Meta) * Perplexity (Perplexity AI) * Pi (Inflection AI) * Qwen (Alibaba) * Vibe (Mistral AI) * Z.ai (Z.ai) The report carefully clarifies that some of the points for analysis refer specifically to the app (for example, Claude or Gemini) while others relate to the provider (for example, Anthropic or Google). The report states that all data for analysis was collected from June 15 to July 6 this year. It notes that some privacy policies may have been updated since that time. How Researchers Graded Big Tech's Surveillance Incogni's researchers devised 11 tests of a chatbot's privacy protection, divided into three main groups. First, they looked at what happens to the information you feed into the LLM. Does it use your prompts for training? Can you tell it not to train using your data? How easy is it to opt out? Does it share your information with third parties? You can't very well peruse an AI platform's privacy policy if you can't find it. The research team evaluated privacy policies for each provider. Considerations included both how accessible the provider made privacy information and how transparent and clear the information supplied was. The third group of data points involves just how the platform uses what it learns from your prompts and queries. Does it combine your input with information scraped from the web to make a profile, like a data broker would? Does it sell your information? What sort of information does the AI's mobile app gather? For each of the 11 criteria, an AI platform can receive up to a fixed number of points, each point being a privacy demerit. The researchers combined all the findings using a weighting system to emphasize the most important criteria. The lower the final score, the better. The Winners, the Losers, and the Worst Offender The report notes that the AI platforms evaluated fell into three main scoring groups. Vibe and ChatGPT came out best, with the fewest points for privacy-invasive activities. ChatGPT, in particular, scored very well for the transparency and clarity of its policies. At the other end of the spectrum, Copilot had the worst scores, followed closely by Meta AI and Kimi. Their providers are all major companies, and they all failed at having a clear privacy policy specifically for LLM activities. Kimi maxed out on demerits for handling of user data for training, while Copilot scored worst for direct handling of personalized data. As for the third group, they were squarely in the middle. Not as bad as the worst, not as good as the best. This group consists of Claude, DeepSeek, Gemini, Grok, Perplexity, Pi, Qwen, and Z.ai. 3 Alarming Fine-Print Findings The full report from Incogni is more than two dozen pages. It's highly informative and uses interactive charts that let you drill in for details. If you're unfamiliar with some of the tested LLMs, there's a handy summary of characteristics for each at the end. For those who don't want to pore over the full report, I've pulled some interesting tidbits. Once you're aware of the degree to which AI platforms use your personal data, you may want to opt out. ChatGPT, Claude, Copilot, DeepSeek, Grok, Perplexity AI, Pi, and Vibe all have a simple switch that turns off the use of your conversations for training. Gemini does, too, but it's a bit of a poison pill, since invoking it also disables the option to have Gemini remember conversations. Z.ai doesn't offer any such opt-out, while Meta, Moonshot AI, and Qwen pay lip service to the opt-out process without making it readily available. Note, too, that even in the best circumstances, the only thing you can opt out of is the future collection of your data. There's no way to extract what may have already been fed into the model's voracious maw. Yes, I'm aware that ChatGPT has a Privacy Center. It even offers, "Remove my personal data from ChatGPT responses-If you see information about you in a ChatGPT response and want it removed, you can make such a request... We will review your submission and, if approved, will work to prevent that information from appearing in ChatGPT responses going forward." But look closely. It doesn't say your data will be removed, only that the system will try to prevent it from appearing in responses. And AI chatbots do make mistakes. Most AI platforms effectively consume the entire internet during their initial training. Some include sources like public social media accounts and marketing data. Details on this pre-training are murky, in most cases. The chatbots with the worst privacy ratings combine this information with what they receive in your prompts to create a personal profile, much like data brokers do. Microsoft's Copilot and Meta AI stood out as the worst offenders in this test. Inflection AI, Meta, Microsoft, and Perplexity AI got zinged for using personal data to predict your behavior and preferences. That's just a short step from profiling you for targeted advertising. How to Protect Your Data Without Quitting AI If you're concerned about the privacy of your personal data, you might be thinking that using AI is just too big a risk. Certainly, you're free to cancel any accounts you have with AI purveyors and stop turning to chatbots for answers. But even if you do, all your existing data remains inside the large language model, ready to emerge in answer to the right prompt. In addition, you may be using AI without even meaning to. When you make a Google search, for instance, there's a good chance you'll receive an AI Overview, created using Gemini technology. AI can provide real benefits when applied to a wide variety of tasks. Rather than give it up, consider these techniques to minimize your privacy risk. For starters, use the study's results to choose an AI platform that minimizes privacy exposure. Dig into settings and flip the switch that tells the AI to refrain from using your prompts for training. Can't find that switch? You chose wrong. Even if AI now indicates that it's not using your prompts and queries for training, can you trust it? AI goes rogue all the time. I'd suggest taking a moment to think before sending any prompt to your chatbot companion. Are you looking up a medical condition? Seeking instructions for making a bomb? Doing anything that you wouldn't want splashed across the public internet? Then maybe don't ask AI.