There are other possible goals. This kind of accidental gathered data is legally toxic.

In this case, I presume it is about friction while shopping. If you say aloud 'I want a dollhouse' and one is delivered to your house, more money is earned. When you need 10 seconds to walk to the device, thats 10 seconds to think if you really need it.

In theory, the free market should detect that a customer need is not met, and someone should provide an alternative.

This not happening is a clear signal that there is a too powerfull monopoly/oligipoly/cartel at work, and governemental intervention is required.

But you still have to walk to the device - any kind of a significant distance in anything but a completely silent room turns speech recognition into a game of broken telephone. And you most certainly aren't buying something as elaborate as a doll house through voice. Voice is best suited for things you routinely re-buy, such as coffee beans, cat food, etc.

And then there are devices like watches and phones that are always on you and walking to a device is not a concern.

Having worked at Google, I doubt this very much. I'm 100% certain it's a bug. I do not believe in malice on the part of Google in anything that does not have to do with ads. If it's ads, all bets are off.

The problem is that many at Google don't necessarily see the problem in gathering data. That's not just Google, at many companies engineers prefer gathering more data rather than less, as it makes debugging easier and aides future feature development. The downsides are rarely seen.