I agree. The current feedback loop between site SEO and search engine algorithms is going to get orders of magnitude more convoluted if SEO turns into the art of "getting noticed" by language models.

I'd expect these models to start being run against each other, adversarially, to try to generate optimally "noticeable" information. And not just in commerce - but also in politics and organised disinformation efforts.

It's really hard to see anything good coming of this, and my hope is that the reputational risk consideration (mentioned in the article) ultimately stops it.