Instead of speculating whether something like this could or could not be true, there should be a way to test it scientifically.
* Have pairs of mobile devices set up from factory configuration with WhatsApp and Instagram installed.
* Simulate conversations between each pair from select topics.
* Collect all ads from Instagram after the WhatsApp conversations from each device.
* Categorize ads to broad topics.
* Search for significant bias.
There are probably a lot of factors I'm missing here, and it's probably easy to introduce bias when there is none there. For example it's probably a good idea that a different person categorizes the ads into topics than the person handling the specific phone, otherwise the person might bias the categorization of the ads based on the conversation they had on WhatsApp beforehand. The person categorizing the ads should have no knowledge of the WhatsApp conversation
that happened on the phone. The devices should probably be on different networks. There is probably a lot that I am missing here.
* Have pairs of mobile devices set up from factory configuration with WhatsApp and Instagram installed.
* Simulate conversations between each pair from select topics.
* Collect all ads from Instagram after the WhatsApp conversations from each device.
* Categorize ads to broad topics.
* Search for significant bias.
There are probably a lot of factors I'm missing here, and it's probably easy to introduce bias when there is none there. For example it's probably a good idea that a different person categorizes the ads into topics than the person handling the specific phone, otherwise the person might bias the categorization of the ads based on the conversation they had on WhatsApp beforehand. The person categorizing the ads should have no knowledge of the WhatsApp conversation that happened on the phone. The devices should probably be on different networks. There is probably a lot that I am missing here.