(smart) advertising customers will want to see metrics and reporting on how their ad campaigns are doing, and making ads that are too subtle runs the risk that the customer is being charged for a weakly or sneakily worded message that they perhaps don't like. Also curious how they're gonna generate reliable, deterministic reporting for crafty embedded ads when LLMs are famously non-deterministic.
> advertising customers will want to see metrics and reporting
You're assuming the customers who want to advertise on ChatGPT will want it to look like the typical low quality internet ads, not like billboards or "company branding marketing", something I think is much more likely.
Instead of selling impressions/clicks, they'll sell "injections" that hopefully (from their PoV) at least has some impact, like a billboard or TV ad today.
LLM providers know a lot about their customers. I expect that, with their post-privacy approach, they will not have qualms about scanning a customers’ future chats for mentions of the advertised product (oh, the analysis potential!).
