I've noticed that after discussing a topic with friends, family, or coworkers, I sometimes see related videos and recommendations on my own apps. Usually my phone is nearby, so tracking, search history, personalization, coincidence, or the frequency illusion could explain it. However, one example confused me: phones weren't allowed inside my workplace, so my phone remained in my car while some coworkers had theirs with them. At work, we discussed Ronaldo Nazário's injury and surgery, and later, after returning home, my phone showed recommendations about that exact subject. Could another person's phone capture or recognize what I said and somehow connect the conversation to my own advertising or recommendation profile? Is that technically plausible, or are location signals, shared interests, prior searches, and recommendation algorithms a more likely explanation?
3 Answers
It’s technically possible for a phone or app to record sound, but that doesn’t mean companies are routinely identifying nearby speakers and attaching their conversations to those people’s advertising IDs. Doing that reliably would be difficult and risky, and widespread audio collection would likely be detected. More ordinary explanations include someone nearby having searched for the topic, devices being associated through location or social connections, and recommendation systems making broad guesses based on activity around you.
Frequency illusion is another major factor: once a subject has been discussed, you’re more likely to notice it when it appears in your feed. Topics can also be trending, previously searched by someone in your household or workplace, or selected because they resemble your existing viewing history. Those explanations are generally more plausible than another person’s phone secretly identifying your voice and linking it to your personal profile.
Advertising platforms can often infer that groups of devices were in the same place or near one another. If one person searches for a subject, watches related videos, or otherwise interacts with it, the system may test that interest against nearby or socially connected users. It doesn’t need to know exactly who had the conversation, and it may show the same recommendation to many people in the area. These systems are probabilistic, so an apparent match can feel much more precise than it really is.
That still seems difficult in my example because my phone was in the car and there were many employees nearby. I’m not sure how a system could determine which person I had spoken with unless it somehow recognized my voice or identity.