Joe English / October 27, 2026
You visited a page. That isn't a biography.
I asked an AI system to look at my activity history and describe my interests. The result was interesting. It was also too sure of itself. Page visits became completed actions. A checkout became something closer to a purchase. A category of clothing became a label for the person browsing it. That is a lot of biography to squeeze out of a URL.
So I asked for the source records. Not just an impressive table: the actual filename, the array position, the timestamp, the activity title, and the recorded link. I wanted to be able to go back and find the thing the system claimed it had seen. A footnote that points to the wrong record isn't evidence just because it looks technical.
The distinctions matter. Searching, saving, liking, subscribing, dismissing, and visiting are different actions. None automatically explains the motive. A saved performance might be a favorite, a future listen, or a reference for someone else. A dismissed clip might mean dislike, repetition, or simply enough of that for today.
Dates matter too. A dense recent export with a few old entries is not a continuous history of a life. And a browser can accumulate background traffic that says more about an extension or a reward system than about the person sitting in front of it. Silence about those limitations makes a profile sound more complete than it is.
I am interested in sharing, monetization, privacy controls, and being recognizable. Exploring several of those at once doesn't hand a model a motive. Ask me. Keep the observation separate from the interpretation. If the example cannot be located, withdraw it rather than decorating it with another confidence label.
I still want to see what these systems make of me. I want a profile that can be challenged and corrected. The interesting part is not how easily a machine can turn me into a category. It is what remains when the category has to answer to the record.
Written with AI assistance from my requests and descriptions of the profiling experiment. Edited commentary, not a verbatim transcript or a publication of the underlying activity logs.