My biggest apprehension about using llms for shopping is that there is no reason to think the models won't sell priority/positive attribution as an alternative to traditional marketing. It's the only monetization scheme that I think makes the absurd valuations make any sense. Condition everyone to offload the tiresome tasks of comparison shopping by just telling the llms what they want. Once people form the impression that the top results are what they themselves would pick or at least close to it, then the llm's will take over the actual purchasing task. Then the stage is fully set to sell consumers as a product to various companies. Imagine if HP could move away from scammy ink subscriptions and instead pay OpenAI to get millions of people to buy their printers and ink anytime they ask chatgpt to find them a printer.
I recognize that this is borderline conspiratorial, but llm's smell like the next evolution of enshittification and removing that pesky rational actor of pricing models.