Find the pages quietly fighting over one meaning

How we read meaning: embeddings and triplets

Search and AI engines stopped reading words long ago — they read meaning. So do we: every page becomes an embedding, a vector of hundreds of numbers where closeness of vectors means closeness of meaning. “How to buy cheap flights” and “tricks to save on airfare” use different words, yet sit next to each other in meaning-space — and so compete for the same intent.

On top of that we break content into semantic triplets — “entity → relation → entity” links. A flights page, for instance, yields triplets like “Aviasales → finds → flights Moscow–Istanbul”, “airfare → is cheapest → on Tuesdays”, “low-cost carrier → flies from → Vnukovo”. Triplets reveal exactly which entities, routes and facts a page is about, so we catch the moment two articles describe the same route in different wording — “tickets to Istanbul” vs “Moscow–Turkey flight” — and quietly choke each other’s rankings.

The result isn’t keyword guesswork but a meaning-map of the whole site: where the meaning-level duplicates are, where a page body steals another’s route, where an article has drifted out of the site’s core — with a clear merge / differentiate / refocus verdict per pair.