How ChatGPT and Gemini can reveal what they infer about you

Generative AI chatbots can feel conversational and helpful, but they can also be surprisingly revealing. In a New York Times Technology report, the focus is on how prompts can be used to test what ChatGPT and Gemini appear to know or infer about a user, and why that can feel unsettling when private details emerge more easily than people expect. The central takeaway is simple: these tools are not just responding to a single question in the moment. Depending on what has been shared with them over time, they may surface patterns, assumptions, or personal details that users did not realize were being retained or inferred. That can include everyday information, workplace context, interests, or other fragments that create a fuller picture than the user intended. For readers, the privacy lesson is less about any one chatbot and more about the way people use AI assistants in daily life. Many users treat them like a note-taking app, search engine, or sounding board, without pausing to consider how much they are disclosing in ordinary conversation. The more specific the prompt history, the more likely the system is to reflect back details that feel personal. The Times piece points to a growing tension in consumer AI: the tools are designed to be useful, but usefulness often depends on memory, personalization, and context. That convenience can blur the line between assistance and exposure. Even when a chatbot is not storing sensitive information in the way a human would, it may still produce answers that make a user feel known in uncomfortable ways. This is also part of a broader public conversation about how much data people should share with AI systems and how clearly those systems should explain what they retain, infer, or use to personalize responses. Consumers may assume they are having a private exchange, when in practice the experience can involve logging, model training policies, account history, or connected services, depending on the product and settings. The exact behavior varies, which makes the privacy landscape hard to summarize in a single rule. What remains unclear from the feed information is the exact set of prompts the Times recommends, the specific settings involved, and how each platform handles memory or retention in this case. Those details matter, because privacy controls differ across products and can change over time. Readers should treat any AI assistant as a system that may remember more than expected unless they have reviewed the relevant controls and data policies themselves. The article arrives at a moment when AI companies are trying to make their tools more personal while users are increasingly asking what that personalization costs. For anyone experimenting with ChatGPT, Gemini, or similar assistants, the bigger question may not be whether the systems can say something surprising about you. It is whether you are comfortable with how they learned it in the first place.

Source: New York Times Technology - https://www.nytimes.com/2026/07/23/technology/personaltech/chatgpt-gemini-prompts-privacy.html