I think these conspiracy theories happen because people don’t understand how easy it is to leak data and how easy it is for data collectors to gather metadata and make a conclusion. Metadata is incredibly powerful and a lot of non-data scientists don’t realize the level of sophistication that companies have in their possession.
The classic example is Target predicting your pregnancy based on specific purchase behaviors. All they have to know is a consistent identifier and your purchase history and they can predict whether you’re pregnant. There’s no need to listen in on conversations or obtain other more detailed user data.
Also, a lot of “private” services and apps really don’t promise jack shit in their privacy policy. They are probably all gathering and selling the data nearly in real time. Their privacy policies are often far more broad, vague, and permissive than their PR will tell you.
You’re with your wife, your devices are often on the same networks, so it’s likely that advertisers know you know each other when you browse. Despite what your wife says, you really don’t know if she interacted with a Pacifica ad or piece of sponsored content. Even if she didn’t search for a Pacifica, it doesn’t have to be specifically something related to minivans, because that information that you are potentially more interested in minivans can come from other metadata.
TikTok manages to figure out your perception of a particular video based on how your fingers are moving on the screen, how long you’re spending on a video, what’s happening when you’re lingering or swiping, etc. You never really have to tell TikTok directly what things you like.
The game of 20 questions works from a similar concept. You can start knowing absolutely zero and ask a very small amount of binary questions to find the specific item the person has on their mind, only metadata.