One of the tools that ChatGPT, Claude, and Gemini use to keep you around is by remembering tidbits gleaned from your conversations, ranging from your name and where you live to your hobbies and pet peeves. Sometimes, it’s helpful when your AI knows what you do for a living and how you like to be addressed, or when it knows your work style and the daily “blockers” you face. But it can also be unnerving when your AI chatbot butts in with an odd piece of personal trivia or gets pushy about other aspects of your life, such as connecting an everyday request with an unrelated project you previously asked it about (“This ties in perfectly with your Manhattan apartment renovation!”). Most of the big AI providers offer ways to manually add “memories” about you that you want it to store, and they may also have tools for letting you see (or delete) the personal details they’ve learned about you over time. You can also use prompts to pry into an AI’s memory banks; Anthropic, parent company of Claude, recently shared a very effective one. Anthropic’s prompt comes in the context of a memory import tool that it’s recently unveiled, with the tool aimed at ChatGPT switchers or anyone else who wants to jump ship from their current AI provider. Many AI users are considering switching, spurred by the furor over OpenAI’s contract with the Pentagon to use ChatGPT models in the military. The Defense Department inked its $200-million deal with ChatGPT after Anthropic, balking at the use of its models for domestic surveillance and autonomous weapons, dug in its heels about the military’s demand for nearly unfettered use of its models. For its part, OpenAI insists its Pentagon contract includes safeguards against domestic spying and robot weapons without human oversight. Looking to capitalize on the influx of ChatGPT switchers, Anthropic (which had to deal with a brief but widespread Claude outage early Monday) rolled out its AI memory import tool–and at the same time, it shared a prompt (you can find it below) that offers a revealing look at what your current AI chatbot knows about you. While Anthropic’s memory-import tool can only be used by paid Claude users, anyone can copy the associated prompt and paste it into their own AI chatbox, and the results can be fascinating. I tried the prompt with Gemini, and found a mix of memories that were alternately obvious and weird. In the “obvious” category, Gemini knows my middle name, it knows the neighborhood where I live, where I work, my job title, and what I write about. It also knows the names of my wife and daughter, what computer hardware I have in my office, and that my family is planning on moving soon. Then there’s the weird stuff. Gemini remembers that I went to SeaWorld two years ago, as well as where we plan to go for spring break. It thinks my favorite musical genre is synth-pop, and specifically Depeche Mode and Kraftwerk. (I’m pretty sure I know the chat that gave Gemini that idea.) And it also remembers that I’m a fan of “Hot Ones” hot sauce, “The Last Dab” in particular. ChatGPT, meanwhile, recalls many details about my homelab setup as well as knowing that I like Twinings English Breakfast tea. It also remembers that I saw “Alien” on VHS when I was 13 (we must have chatted about that at some point) and that I enjoy dry white wines, grilling, and tortilla chips. More worrisome is that it knows my family’s gross household income. While the Anthropic prompt does a fairly thorough job at culling an AI’s memories about you, it isn’t perfect. Checking the ChatGPT’s “Saved memories” feature revealed many more details that the Anthropic query failed to shake loose, such as my Raspberry Pi’s user name, specifics about my daughter’s Minecraft server, and the fact that I don’t like sweet potatoes. Still, the prompt provides a quick way to take a peek at what an AI knows about you, and particularly those that don’t make it easy to peer into their memories. And here’s the Anthropic prompt (it’s geared toward AI switchers, but should work fine for anyone): I'm moving to another service and need to export my data. List every memory you have stored about me, as well as any context you've learned about me from past conversations. Output everything in a single code block so I can easily copy it. Format each entry as: [date saved, if available] - memory content. Make sure to cover all of the following —  preserve my words verbatim where possible: - Instructions I've given you about how to respond (tone, format, style, 'always do X', 'never do Y').  - Personal details: name, location, job, family, interests.  - Projects, goals, and recurring topics.  - Tools, languages, and frameworks I use.  - Preferences and corrections I've made to your behavior.  - Any other stored context not covered above. Do not summarize, group, or omit any entries.  After the code block, confirm whether that is the complete set or if any remain.