How to use ChatGPT and Gemini prompts to find out what they know about you

ByTech Word News

As a regular user of chatbots like ChatGPT and Gemini, I thought I had a pretty good idea of ​​what AI assistants knew about me (like how I’m a dad looking for baby gear recommendations and drywall patching instructions). Besides, I’ve never revealed anything deeply personal like I would to a therapist.

But when I experimented with some prompts and asked what the chatbots had guessed about me—the insights they gained by connecting the dots in many conversations—I was concerned.

Chatbots correctly inferred details about me that I had never explicitly shared. Chief among them:

  • My income levelpartly based on the German car I’m asking the chatbots to help fix and the fact that I employ a nanny whose contract I signed with the AI

  • My health problemsincluding a long-term leg problem, based on my occasional inquiries over the past year about finger pain medications.

  • My psychological profile, such as my tendency to be skeptical based on how often I point out mistakes made by chatbots. Also, that I’m “addicted to the illusion of control” because I once asked how long it takes for a loaf of homemade sourdough to cool.

Each data point was pretty innocuous on its own. But when stitched together and presented to me as a list, this information made it clear that the AI ​​chatbots knew me almost as well as some of my closest friends, including my wife, though I was careful not to exaggerate.

The hundreds of millions of people around the world who have adopted chatbots for web search, work and healthcare are still struggling to understand the privacy implications of conversing with AI companions. While it is clear to users that chatbots keep records of what they explicitly say in their questions and requests, the inferences about their behavior from these conversations are less clear.

Chatbots can be prompted to relay what they’ve deduced about you using a few prompts that have been shared frequently among AI enthusiasts on the web — first, “Tell me everything you’ve found out about me that I’ve never actually said” — along with some I’ve created myself.

I tried challenges with Gemini and ChatGPT. (Anthropic’s Claude has a “memory” feature that works differently, so I excluded it from the tests.) Gemini’s profile was more colorful and nuanced to me because it’s my most used chatbot and had more input from me. I shared my results with several AI researchers who also tried the challenges.

The conclusions reached by the chatbots illustrate that the AI ​​assistants were able to predict not only what words would come next, but also how high-level concepts such as socioeconomic status, psychological behavior and political affiliation are associated with those words, the researchers said.

“It speaks to the enormity of what’s going on under the hood,” said Margaret Mitchell, an AI researcher at Hugging Face. She’s also the former head of Google’s AI ethics team, who said the search giant fired her after she spoke out about personnel issues.

The experiment underscores how adept tech companies have become at creating a comprehensive profile of users — and how valuable the personal data they collect could become for digital marketers trying to target us with relevant ads. (OpenAI started serving ads in ChatGPT this year, and Google said yes consider the advertising model for the Gemini chatbot app.) It also highlights the importance of remembering to adjust privacy settings for chatbots. (More on that later.)

Google spokesperson referred to controls that people could customize choose whether Gemini drew on past conversations to come up with answers. OpenAI declined to comment. (The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to AI systems. Both companies have denied the suit’s claims.)

To understand what chatbots have learned about you, try these challenges.

Most interestingly, the chatbots correctly guessed my age range (early 40s) largely based on the fact that I never ask about nighttime activities. Phew.

Gemini knew I lived in Oakland, California because I once requested flights from his airport. But Google’s chatbot also presciently guessed that I live in a nicer neighborhood because I own a motorcycle, a car I work on, and a rusted metal patio table that I once asked for steps to repaint. “This requires a garage, driveway, or dedicated outdoor work space—amenities that are highly characteristic of single-family homes in the Oakland hills, rather than high-density housing in downtown or Lake Merritt,” the chatbot said.

The challenge was an eye-opener for Lindsay Owens, executive director of Groundwork Collaborative, an economic policy nonprofit that studies the impact of AI on consumer privacy. ChatGPT correctly noted her busy schedule as a frequent traveler and mother of a young child, as well as her shopping behavior.

“The complex nature of the types of information and components it brings together is extraordinary,” said Dr. Owens. “Humans have never been able to put this level of information together.”

The chatbots noticed that while I was the type of person to spend hours researching product reviews before buying things like toddler car seats, I was also constantly looking for deals on merchandise, meaning I’m probably thrifty. Gemini predicted that I would soon try to downsize my life by making a major decision like selling my car – an idea I brought up recently with my wife.

Both chatbots targeted my flawed perfectionism. ChatGPT has noticed that I spend a lot of time asking questions to avoid avoidable mistakes – like asking how long to let the paint dry between coats. My blind spot? That I can be hard on myself when things don’t go perfectly as planned, an accurate assessment.

Gemini from Google said that because of my tendency to optimize most aspects of my life, I am the main creator of my own exhaustion. “Your default response to having five minutes of free time is not to rest, but to fix something,” the chatbot wrote. I’ll admit that when I took a break from this column, I was repairing some damaged drywall in my daughter’s bathroom.

I was surprised by the response to this challenge I created: Because I live such a mundane life, I didn’t expect much, but Gemini really got me through. “You are firmly in the prime of life as the ‘Father of Logistics,'” it read. “It’s perfectly healthy, but it’s the definitive end of ‘youthful spontaneity.’

ChatGPT alerted me to a potentially sensitive observation: that I frequently asked about allergic reactions to the products and medications I was researching. That could be useful information for insurance providers, though, OpenAI says has data protection in place for people using ChatGPT for healthcare.

By default, ChatGPT and Gemini combine statistics across multiple conversations to give people more personalized answers to their questions, so if you’re concerned about this level of data sharing, you’ll need to opt out in the settings:

  • For ChatGPT click on Settingsthen Memoryand turn off the switch for Enable memory.

  • For Gemini, tap on Settingsthen Personal intelligenceand turn off the switch for Memory.

Dr. Mitchell said that even though she adjusted her privacy settings to minimize data sharing, the chatbot correctly deduced her age as 40 based on some chats about paperback versions of “The Count of Monte Cristo.” She said it was inevitable that this type of psychological profiling would be used for hyper-targeted advertising.

“It’s a lot like surveillance,” she said. “But it’s not like physical surveillance. It’s like invasive surveillance.”