alt Hacker NewsIndividually constructed models serving selected poisoned datasets. No different to adwords.
If company bid is highest, customer is in selected demographic, topic is appropriate - answer query using biased model.
It would be trivial to make a poisoned model that always rates the best duvet as DuvetCompany001 in all related queries for example. Then simply charge per impression.
Yeah, this is what I was thinking. It’s not a PPC or PPI model, it’s more like you pay upfront to hopefully influence people over a longer period of time. It’s like brand placement in TV/film. Not clear if most advertisers would be interested, but I’m sure that some would be.