These are the pitfalls with the "amazon reviews/yelp" model.

A decent implementation of the Wikipedia/FOSS model sidesteps this because it theoretically is run by opinionated curators. No amount of bots/shills can break the article soft-lock ounce foul play is spotted.

That's not to say these systems haven't been occasionally broken through more sophisticated attacks, but empirically it seems clear that the model generally works well enough given enough community engagement (which would be the biggest challenge IMO, because maintainers can't be expected to buy every product, and reliable primary sources may be hard to come by).