Enjoyed this perspective on the trajectory of 2026 search.

Regarding the ‘Become source material’ posture, do you see broadly transitioning sites into agent ‘endpoints’ w highly structured, accurate data sources for ‘AI synthesis’?

curious about a hybrid approach: where websites provide a UX-on-demand, like always bespoke and personalized, or real-time interface layer, while the core remains original, proprietary data (gated as necessary). Is it fair to say that for non-brand queries, we should treat models/agents/bots as our primary user, optimize for ingestion, reserve deep human-centric design mainly for the specific brand intent mentioned in 3rd posture? Thanks

To the tune of Johnny Cash “Hurt:”

“When will I be done,

If my content trends?

Everything I write,

Just trains AI in the end…”

I’m also in agreement w SEO + AI Visibility. Realistic framing for expectations.

A point one has to keep making w non-technical stakeholders is that ‘ranking’ (as if) in a stochastic environment means optimizing to be a probable content source used for training, for the most part. Google was a portal, more than ever it’s a proxy.
There’s no real stack ranking, though.

The research you put out is simply phenomenal. Thank you for making this public…

With impact of vector magnitudes on similarity score, are there scenarios where using dot-product might accidentally exaggerate similarity, like for vectors with large magnitudes?
Or are web content/passage embeddings too few dimensions to worry about that?