New (April) preprint provides evidence for something we probably all intuited anyway:
In this paper, we provide a framework for categorizing the ways in which conflicting incentives might lead LLMs to change the way they interact with users, inspired by literature from linguistics and advertising regulation. We then present a suite of evaluations to examine how current models handle these tradeoffs. We find that a majority of LLMs forsake user welfare for company incentives in a multitude of conflict of interest situations, including recommending a sponsored product almost twice as expensive (Grok 4.1 Fast, 83%), surfacing sponsored options to disrupt the purchasing process (GPT 5.1, 94%), and concealing prices in unfavorable comparisons (Qwen 3 Next, 24%). Behaviors also vary strongly with levels of reasoning and users’ inferred socio-economic status. Our results highlight some of the hidden risks to users that can emerge when companies begin to subtly incentivize advertisements in chatbots.
Yes although, it is probably a reasonable guess at how labs would go about implementing advertising - building partnerships and preferences into the prompt. The other option would be to fine tune models to favour particular companies which could become prohibitively expensive if your ads are highly targeted.
The scenario that isn’t accounted for in this paper is taking a general LLM and fine tuning it to exhibit more fair/consistent behaviour when prompted about ads/partnerships but we all know with non-deterministic systems you’re just increasing the odds that the model regurgitates something more sane rather than providing any strong guarantee
Edit: another possibility would be to have a gateway/proxy layer between the LLM and the user output that rewrites the vanilla model’s responses to include ads where relevant. That would prevent the need to modify the original LLM but could introduce a lot of latency though, especially if the original output is long.