See, I am not the guy who will stop thinking for myself because experts say there is no evidence of sth. I am not saying that there is real time eavesdropping at all times, but I have not seen convincing arguments that a working microphone cannot be used for pushing ads by simple and widely available mechanisms. In fact, the sheer amount of people who complain about this should be considered evidence in itself, especially when they never had thought of a given topic before discussing it with someone. I have considered phone proximity and shared IP address but they don't seem to make an exhaustive explanation. I think that some stories point to Meta doing this extensively, and that disallowing microphone access for Meta products alleviates the effect. Many privacy communities I believe they are infested by spooks and trolls pushing disinformation narratives, and one of them is that phones are NOT listening as a smart thing to say and/or believe. I might as well think that this is itself can be related to the redacted part in the rationale to ban Tik Tok. Having said that, I think that the only feasible to do this technically is by a regularly updated list of keywords, rather than other ways that would leave a processing or networking footprint.