It Can Be Surprising What AI Knows About You

I was going over expenses and budgets today and thought I’d ask Grok what I might want to improve on my Excel budget sheet. I didn’t upload my budget information. I also didn’t use a project or a skill. This was the sum total of my prompt:

When constructing my own budget tracker so that I can mark bills, for a family of two adults and three children, listing utilities, work expenses, groceries, gas, insurance, and luxuries, so that I can enter monthly expectations and update them with actual expenses, what Excel fields and equations do I need to use and what templates, if any, are recommended?

Its response took a surprisingly long time. Two minutes, forty-nine seconds. Much longer than anything else I’ve asked since 4.0 dropped. Normally even complicated questions only take about sixteen seconds or so. (I haven’t been keeping a log, but “thought for 16 seconds” pops up frequently enough to notice.)

The response was very detailed. I normally don’t load in much private data; old habits due to my father’s government positions and his desire to keep away from attention, both from the media spotlight and to avoid too much scrutiny from groups like the Mafia, who once had a contract out on him. I’ve made it a habit to obscure my family’s digital trail since long before companies like Incogni and Delete Me were around. So I know how much hard information Grok has on me.

What I didn’t know was how much it could figure out about me. I asked it a very general question about templates, and it provided very specific advice on what additional categories to use.

And after a very, very long and detailed list of suggestions, it then provided a ready-made Excel sheet specifically for my family, using everything it had on us and everything that was statistically likely about us. I once asked it a maintenance question about one of our vehicles, and it put the exact make and model down as part of vehicle expenses. It had a very good guess about our household income; it wasn’t right for my wife, but it knew our professions and it knows my rates from my website, and both sets of income, plus royalties, were extremely good guesses based on the statistical likelihood of our jobs.

It had a bunch of other details in there. An incorrect but likely guess at our specific power provider, a close guess on our typical utility and insurance bills, and even the ages of our children (each digit wrong, but still close).

If I didn’t know how AI analyses worked, I’d be spooked. As it is, I was still very surprised and impressed.

Because I know how much of my personal information is put online, and particularly on X where Grok has direct access to all my public posts, I know most of the identifying information is based on things I’ve given it directly to solve certain problems. So I know most of the information isn’t floating around, and that it’s also not easily obtained by a hacker (not the same as impossible). But a more ‘normal’ person, without my digital habits and without as much knowledge of how AI works, would probably be terrified.

Even so, I highly suggest that if you don’t want information out there, be vague, introduce a few deliberate inaccuracies, and be careful how much you feed an AI. I’m not worried; but it’s still a good practice to be very, very cautious.