Yeah, I don't see the issue there. Some people buy a mattress, don't like it, and then buy another different mattress. Some people like the mattress they bought for their kid so much they buy a similar one for themselves. Some people like the mattress they bought for their rental house so they buy one for their primary residence. And so on. Sure, it might not be super likely that someone buys multiple mattresses in a short period of time, but if someone has mattress buying on the brain, it is possible.

I don't think people assume there is literally zero signal. I think their intuition is that the repurchase rate is low. Low enough that that if the advertiser knew the purchase was made, then the advertiser would serve something else with a higher predicted chance of getting a click.

It's easy to assign a straw-man a belief there is zero signal and talk down to it. The more nuanced idea that is less obvious goes like this: advertising is such a boondoggle that the aforementioned weak signal actually does represent a higher click-through than everything else in their ad inventory.

Advertisers working in predictive targeting like to tell tall tails about their analytical genius. That story about the Target ad campaign predicting someone was pregnant is their ur myth. The basis for their big swinging statistical analysis prowess. Imagine how embarrassing it would be to their carefully crafted image if they had to admit the peak of their profession is hoping people need a 2nd of something that's usually a less-than-once-a-year purchase.