At least in the US the ads must be labeled as such by law, so at a bare minimum I expect the ad blocker devs will be able to remove them with some work.
AI firms already managed to successfully weasel their way out of copyright laws. They will try to do the same with advertising laws.
I wonder if the current laws are written in a way that accounts for these models. Sure, if a specific tool call results in a paid product card for pepsi, that ought to be labeled. But what if the number on some pepsi-related weights is massaged just a bit, way early on in the process? What if the training data is tweaked to include some additional pro-pepsi inputs?
I look grimly forward to the future of adblock, which I predict will literally involve a media interception and re-rendering agent that sits between us and everything we see, hear, read, etc. AR goggles that put beach pictures over bus stop posters and red squigglies under sentences with a high enough adtech confidence score. This shit's gonna get real weird in our lifetimes.
There's a whole design niche dedicated to making that label as subtle and hard to see as possible.
And I'm skeptical ads will remain outside of the ChatGPT output for very long. You can hide a div tag, but you can't hide an advertisement streamlined into the "conversation" with ChatGPT. Is ChatGPT recommending product X because they're an advertiser, or because that's what it "learned" on the internet? Did it learn from another advertisement?
I fully expect them to exploit the plausible deniability.