Ecosia routes AI overviews by intent
Product manager at Ecosia on building AI-powered summaries with search
The gating logic shows that AI search is being treated as a scarce, expensive layer for hard queries, not as a default replacement for the search box. Ecosia is using live query shape plus past click behavior to separate low value intents, like go to one site or buy one product, from research intents where people normally open several tabs and compare sources. That matters because AI overviews cost money on every call, while regular search keeps ad inventory intact on most traffic.
-
In practice, the cutoff is simple. One word queries and obvious destination searches, like LinkedIn or Facebook, stay on standard web search. Open ended questions and queries that historically lead to multi site research are routed to Exa for a synthesized answer with source links.
-
This is as much an economics system as a relevance system. Around 30 to 40% of queries trigger the AI overview, about 500,000 times a day, on roughly $300,000 of monthly Exa spend. Ecosia says about 70% of queries remain regular search because AI search raises cost per user and does not directly monetize on its own.
-
The broader pattern matches AI search product design elsewhere. Traditional search wins when intent is known and speed matters. AI search gets strongest when the job is research, comparison, or synthesis across many pages. That is why Perplexity has concentrated on longer, more complex queries rather than trying to answer every search.
This routing layer is likely to become a core control point in consumer search. As AI answer costs fall, the frontier will move from deciding whether to show an overview at all, to deciding which intents deserve synthesis, commerce actions, or plain links, with the best products using behavior data to tune that split market by market.