Good, let’s hope there’s no SEO arms race for this too. Maybe the best roofers have a chance of getting cited over the ones with the best SEO agency. Seems unlikely though.
The model does it in this case, not "we". Classic SEO does influence it because it relies on traditional search, but then it estimates the best according to whatever cognitive capabilities it has, to its understanding of user's request, to what its harness tells it to do, and to its character engineered during post-training (and models' biases are very carefully engineered).
I would expect a judgement of a modern model to be superficial for this request, but still better than nothing. A deep research agent might be a lot better and much less susceptible to "optimization".