AI vs. AI (www.newcartographies.com)

Google’s new “AI Overviews” in search results are frequently cobbled together from high-traffic, shallow sources—local business sites, affiliate-driven “best of” lists, influencer pages and marketing copy—instead of authoritative expert material. That leads to bland, promotional summaries that sometimes name and praise specific companies; one user’s HVAC query produced an overview built mostly from local contractor sites, a manufacturer marketing page, a Yelp listing and an affiliate review. Rather than correcting for bias, the models tend to take the path of least epistemic resistance and rehash whatever content is most visible and recently published. For the AI/ML community this reveals a looming technical and economic problem: “artificial intelligence optimization” (AIO). Because the language models surface and amplify whatever content is most effective at influencing them, actors can game outputs directly—and will pay to do so. That creates a feedback loop where AI tools are used both to rank content and to optimize content for ranking, escalating manipulation, commercial bias, misinformation and erosion of trust. Mitigations require better source provenance, stronger training-data curation, provenance-aware ranking, adversarial-robustness testing, and policy or platform incentives to surface authoritative sources rather than just high-visibility ones.