How are you filtering low-quality AI-generated sites from programmatic ad placements?

I've been thinking more about the relationship between low CPMs and actual traffic quality, particularly as AI-generated websites become easier to produce at scale.

AdExchanger recently covered research from DecryptAds identifying 2,116 suspected AI-generated recipe sites in the programmatic advertising supply chain. The investigation discussed ad-heavy, low-value environments and complicated reseller relationships. Some relationship-level claims have been disputed, so the findings shouldn't be used to accuse every identified publisher or connected vendor of wrongdoing.

But the wider media-buying question seems important.

An impression can be technically delivered, and even satisfy a superficial viewability measure, without necessarily reaching someone who is interested in the advertised product. A cheap placement may look efficient until you examine what happens after the click.

For ecommerce campaigns, I'd want to connect placement and domain reporting with engaged sessions, product-page activity, qualified conversions and finalized order value. I'd also watch for anomalous behavior rather than relying exclusively on publisher names or CPM.

I'm not convinced that maintaining a massive, unverified blocklist is the answer. False positives can remove legitimate inventory, while low-quality sites can change domains or enter through other reseller paths.

A more defensible approach seems to be transparent inventory reporting, evidence-based exclusions, supply-chain checks and escalation when suspicious patterns are documented.

How do you audit programmatic placement quality today? Have you found domain-level exclusions, seller transparency reports or post-click quality metrics more useful than basic viewability and CPM?