This strategy (showing more of what was searched) is certainly not optimal; but it's quite simple, and I'd say it's almost certainly better than showing random products. Ideally the ad provider should collect enough data to infer the most likely buy given the view history -- but this requires a vast dataset to provide a robust estimate, s.t. a decent approximation may be this strategy of just showing more of the same.

It's an interesting design point for me what the ad providers actually want to maximize. On one hand, they want to maximize the number of 'clicks', which is usually the direct revenue criterion -- but if those clicks don't generate sales their long term value is lost. But they also might not want to maximize sales exclusively: after all, they have multiple customers, and if they provide a large increase in sales for just a few of them, their overall service might see less demand on the long run. And finally, there are the consumers for which they also might want to maximize value -- otherwise they will be alienated on the long term (i.e. even the most naive user wont fall for one weird trick ads more than once). This should all come into play in the design of good ad decision system.

I believe all successful ad providers in the long run will have to give an overwhelming priority to customer value.