Search query distributions follow a power law: the top few hundred queries (head) account for ~30–40% of traffic, the next thousand or so (torso) for another 30%, and everything else — queries seen 1-3 times — for 30–40%. That last bucket is the long tail, and it’s where the worst search experiences live.
Head queries get hand-tuned: synonym lists, merchandising rules, category boosts. Long-tail queries are too varied for that — they get whatever the ranking algorithm gives them. They’re also where natural-language and semantic search pay off most, because traditional keyword matching on a never-seen-before query has nothing to anchor to.
Measure long-tail health separately from aggregate metrics. A 5% improvement in head-query NDCG looks identical to a 5% improvement in long-tail NDCG, but the second is much harder and means more for new and exploratory shoppers.