I have to say, I'm very sceptical about all of this, but one day a friend showed me a product (some speakers that use the table surface for resonance) and the ad showed a couple of minutes after on Facebook.
I started thinking on how devices can emit audio signatures for other devices around it to be aware of its presence. A couple of years after, Chromecast uses audio for authentication of nearby devices.
I honestly dismissed my paranoia all of these years with the argument that it would show in battery and data usage. But with offline language models, I'm not so sure anymore.