Meaning-Based Search Reshapes Publishing
Will Bryk, CEO of Exa, on building search for AI agents
Meaning based search changes the publishing game by rewarding pages that actually solve a question, not pages that merely repeat the right words. In Exa's model, retrieval quality depends on whether a document semantically matches a complex request and clears a quality filter, so the winning article is the one that best explains the topic, stays current, and contains the concrete detail an agent or summary system can use downstream.
-
Exa is not trying to index the whole web first. It deliberately crawls a high quality subset, then ranks by embeddings and quality signals. That means publishers are pushed toward substance and trustworthiness, because weak pages are more likely to be filtered out before ranking even begins.
-
In production, customers use Exa on queries where keyword search is weak. Ecosia routes only more complex searches to Exa for AI overviews, and an Exa data pipeline runs 5,000 daily prompts to find fresh technical articles, with roughly 80% of results judged relevant after filtering.
-
The competitive contrast matters. Perplexity is optimized around consumer answers and subscriptions, while Parallel is stronger on multi step research tasks. Exa's wedge is raw retrieval, large result sets, and full text extraction, which favors content that can be directly consumed by agents, not just clicked by humans.
Over time this should split the web into content made for human attention and content made for machine retrieval. The highest value pages will increasingly look like clean, specific, evidence rich reference objects that agents can reliably pull from, and search infrastructure providers that reward that format will shape how professional and educational content gets written.