Search Ranking for Long-Tail Queries: When the Best Match Has No Sales History

A shopper typed seven words into the search box and your best product for it has never sold. The ranking engine has no sales history to lean on, and the product that should win is sitting on page three behind products that happen to have history.

Long-tail queries are specific multi-word searches, four words or more, with low volume and high intent. The ranking problem is that the best match often has no sales history, so the engine cannot use its usual signals. The fix is borrowed signals: category, attributes, price band, and imagery, plus a controlled exploration budget that lets new products earn their own history. Amazon's A10 algorithm weighs conversion by keyword and sales velocity, which is exactly why a new product with no sales cannot rank on its own merit.

The long tail is where intent lives

Long-tail queries are four or more words, specific, and low volume individually. A shopper who types waterproof hiking boots size 10 wide is not browsing. They know what they want, and they are close to buying. The volume is small per query, but the tail is long: 200 long-tail terms at 500 searches a month is 100,000 monthly searches, and the intent behind them is higher than any head term on the site.

The ranking problem is that the best match for a long-tail query is often the product with the least history. A new product, or a product that has never sold for that specific query, has no conversion data to rank on. The engine's default signals, sales velocity and click-through rate, both favor the products that have already sold. The new product is not losing a fair contest. It is not in the contest.

Related reading on this site: RelevanceRail and Query Understanding: When Search Has to Guess What the Shopper Means cover the neighbouring problems in more depth.

Amazon's A10 shows the trap clearly

Amazon's A10 algorithm weighs two signals above the rest: conversion by keyword, and sales velocity. A product that converts well for a keyword ranks higher for that keyword, and a product that sells fast ranks higher overall. Both signals need history. A new product has neither, so it ranks low, gets few impressions, sells nothing, and stays low. The loop is self-reinforcing, and it is the reason new products on Amazon need external traffic or advertising to break in.

The same trap exists on any store's own search. A ranking engine that leans on sales history will always favor the incumbents, and the incumbents are usually the products that match the long-tail query worst. The fix is not to remove the sales signal. It is to add signals that do not depend on it.

Borrowed signals and a controlled exploration budget

Borrowed signals come from the catalog, not from the order history. Category tells the engine what the product is. Attributes tell it what the product does: waterproof, wide fit, leather, insulated. Price band tells it whether the product fits the shopper's budget. Imagery tells it whether the product looks like what the query describes. A new product with strong attribute matches to a long-tail query should rank on those attributes, even with zero sales.

Controlled exploration is the second half. Give new products a small, measured share of impressions for the queries they match, and watch whether they convert. If they convert, they earn their own history and the borrowed signals can step back. If they do not, the exploration budget shrinks and the product drops. The point is to let the product prove itself, not to promote it forever on the strength of its attributes alone.

What to measure after the change

Judge the change on the long-tail queries themselves, not on sitewide search revenue. Track the zero-result rate for long-tail queries, the click-through rate on the first result, and the conversion rate for queries that previously returned only incumbents. A ranking change that lifts the first-result click-through rate on long-tail queries is working, even if the head terms do not move. The head terms were never the problem.

Frequently asked questions

What is a long-tail search query?

A specific search of four or more words, with low volume per query and high buying intent. A shopper who types waterproof hiking boots size 10 wide knows what they want, and 200 long-tail terms at 500 searches a month is 100,000 monthly searches.

Why do new products rank badly for long-tail queries?

Because ranking engines lean on sales history, and a new product has none. Amazon's A10 weighs conversion by keyword and sales velocity, so a product with no sales ranks low, gets few impressions, sells nothing, and stays low.

What are borrowed ranking signals?

Signals from the catalog rather than from order history: category, attributes such as waterproof or wide fit, price band, and imagery. A new product with strong attribute matches to a long-tail query should rank on those attributes, even with zero sales.

What is a controlled exploration budget?

A small, measured share of impressions given to new products for the queries they match, so they can earn their own sales history. If they convert, the borrowed signals step back. If they do not, the exploration budget shrinks and the product drops.