>This incentive is inevitably going to lead to ad supported LLM's not answering questions properly the first time so more ads served.

The extra inference cost is going to be greater than whatever fraction of a penny they would get from doing this.

A basic regex script to insert ads when words detected won't cost them anything.

If I can tell that the LLM didn't answer my question properly, why would I engage with the ad the answer contains, which would likely be equally irrelevant?

The google ad revenue and Meta ad revenue is hundreds of billions a year.

They have behavioural experts who know how mainstream minds work and how to manipulate.

Remembers ads are predominately not for the educated and intelligent, that type of person will think critically. People who are more naive are the golden gooses