This may or may not apply to the anecdote you shared about your wife, but since these apps know your relative proximal location to your weak/strong social connections, they also can know what your friends may search for and is often-used flavor of targeting.
e.g. You and a bunch of friends go to dinner and have a conversation about <topic x>, at least one of those friends googles something about that topic. You later see an ad related to <topic x> because you were targeted based on the search your friend did while they were near you.
If your wife potentially did anything digitally, related to the diabetes topic, its likely that you were targeted based on that.
Again, no idea what happened resultant to the story you shared, but whenever this sort of thing happens to me I try to appeal to Occam's Razor based on how much I know about how this tech works under the hood.
I thought of this. Just to be clear, neither she nor I did any searching on diabetes or anything like that. I can understand this being driven by search, chats, emails, etc. (basically any type of keyboard input). But here, the mode of communication was voice-only.
Yea I've definitely had situations where I've been challenged in this way (in which I definitely was having a voice-only conversation that seemed to be targeted after the fact). It can be vexing.
It was mentioned elsewhere in this thread, but this is where Baader Meinhof may apply. I don't how many times I've seen a targeted ad that was a "miss" in terms of recency bias. But I absolutely remember every time there was a "hit".
Both situations were targeted based on my digital behavior, but they're playing the volume shooter game. Taking as many shots as possible hoping eventually they score. This could be true in your case. The fact that you and your wife were having a recurring conversation about stem cell research, diabetes, etc. suggests that its likely that this is in your digital fingerprint at least once in the past (recent or otherwise).
Something I try to do now, when I'm being mindful about it, is note how often I see ads that are definitely in my interest bucket but are completely uncorrelated with any recent conversations I've had. That helps at least establish an anecdotal ratio of hits-to-misses that makes the Orwellian/dystopian much less reasonable on balance.
Occam's razor, in this situation, is that the corporate entity who profits from collected data while deceiving it's product into perceiving itself as a "customer" might, in fact be... collecting data from its product to sell to it's customers.
The core tenet of the principle is that the more entities you have to posit in any given postulate, the more assumptions you have to make, the less likely it is to be true.
I was making the point that I have to assume much less to make the statement that companies are not passively recording, tokenizing, and analyzing every conversation users have without their consent.
e.g. You and a bunch of friends go to dinner and have a conversation about <topic x>, at least one of those friends googles something about that topic. You later see an ad related to <topic x> because you were targeted based on the search your friend did while they were near you.
If your wife potentially did anything digitally, related to the diabetes topic, its likely that you were targeted based on that.
Again, no idea what happened resultant to the story you shared, but whenever this sort of thing happens to me I try to appeal to Occam's Razor based on how much I know about how this tech works under the hood.