Pull up your on-site product search report and find the list of zero-result search queries. At a mid-market retailer, it usually runs a few hundred rows, each one a query a shopper typed that returned an empty page. Now try to answer a single question about it: which of those rows are queries you could have answered, and which are queries you had nothing to answer with?

The report will not tell you, and that distinction decides everything that happens next. One kind of row is a configuration problem owned by whoever runs on-site product search, and it can be closed out by Friday. The other kind is an assortment problem owned by whoever decides what the company stocks, and no amount of search tuning will touch it. Both appear in the same list, sorted by the same volume column.

Separating them takes an afternoon. Teams that do it stop treating zero-result searches as housekeeping and start reading the report as the only unfiltered record they have of demand they did not serve.

Every zero-result search is one of two problems

A zero-result search query is a search a shopper runs on your site that comes back empty, and every one of them is either a query you could have answered or one you could not. That sounds too obvious to be useful until you try to act on it and find that the two kinds carry different owners, different clocks, and different fixes.

Pile one holds queries where something in your catalog would have satisfied the shopper and your on-site product search did not return it. Someone searched using a word your product data does not contain, misspelled a brand name, or filtered themselves into an empty page. You already own the inventory, so recovering the sale means changing search configuration, product attributes, or the zero-results page itself, and any of that can happen inside a week.

Pile two holds queries where nothing in your catalog would have satisfied the shopper. Synonym work cannot reach those, because the product isn’t there to be found. Acting on pile two means changing what the company buys, which runs on a seasonal calendar and belongs to a team that does not attend your search reviews.

Almost every retailer works pile one and leaves pile two alone, for an understandable reason: pile one has a button. Oliver Bonas monitored its zero-result searches and added unrecognized terms by hand, accumulating more than 2,000 synonyms over time. That is careful work, and it addresses half the report.

Pile one: tell the four causes apart before you fix anything

Pile one splits into four causes that look identical in a report and respond to completely different work. Diagnosing which one you are looking at, before touching configuration, keeps you from adding synonyms to a problem that synonyms cannot solve.

1. Vocabulary mismatch: the shopper used a word your product data does not contain, and a near-equivalent product sits in the catalog waiting. Hartville Hardware finds these by reading its zero results report against its own naming conventions, then either adding synonyms or rewriting product copy to use the customer’s term. You can spot this cause because you can read the query and name the product it should have returned.
2. Misspellings and brand terms: volume concentrated on a handful of near-identical strings, with visible typos. Incu’s traffic runs heavily on category and brand queries, and slight spelling errors were leaving shoppers at dead ends on a catalog that carried the brand they were asking for.
3. Thin product attributes: the product exists, and nothing in its product data matches what the shopper asked about. Baymard Institute’s 2026 search benchmark sorts queries by type and finds that sites fail use-case searches 43 percent of the time, and compatibility searches 44 percent, against 12 percent for exact searches.¹ Those rates track how complete the product data is rather than how good the search engine is, which is why they resist search-side fixes.
4. Filter and stock dead ends: the query returned results, and then a filter combination or an out-of-stock rule emptied the page. This one gives itself away when the query works for you and failed for the shopper.

Whatever the cause, the shopper saw an empty page, and most of those pages offer nothing to do next. Baymard’s benchmark of 344 ecommerce sites found that 68 percent have a no-results page that is a dead end, “offering no more than a generic set of search tips.”² Rebuilding that page does not require finishing the diagnosis first. ITS, a UK power tool and workwear retailer with a 30,000-SKU product catalog, placed personalized product recommendations on its zero-results page, and those recommendations influenced more than $57,000 in revenue across 2025.

Pile two: nobody types a query by accident

Pile two goes unread because of a reasonable objection. When a shopper searches for a product you do not sell, that query reads as a non-event, since you cannot lose a sale you were never in a position to make. That same objection makes pile two the most valuable part of the report: someone came to your site, told you in their own words what they wanted to buy, and you paid nothing to collect it.

Milton & King, which sells designer wallpaper to direct-to-consumer and trade buyers, reads zero-result search queries as a design brief.

A wallpaper company changed what it manufactures based on a report most retailers use to schedule synonym maintenance. SuperATV, which sells aftermarket off-road vehicle accessories, reads its report the same way. Josh South, its Director of Marketing and Technology, credits the zero results report with driving “new product development within the organization because it showed us that people were on our site looking for certain products that we didn’t have.”

No conversion report could have produced either decision, because conversion reports only describe products you already stock.

Send pile two to the people who choose the assortment

Pile two is worth nothing while it sits in the ecommerce team’s reporting console, because nobody on that team can approve a purchase order. Moving it is more of a scheduling problem than an analysis problem, and two Athos Commerce customers have settled on cadences that work.

Hartville Hardware runs a monthly send. “Every month, I pull our search terms and send them to the buyers so they can see if customers are searching for products that we don’t carry, or products that we’ve been on the fence about introducing,” says John Mueller, the retailer’s Ecommerce Manager. FastGrowingTrees.com works quarterly, which suits a business where roughly 60 percent of revenue arrives between March and June. “If people are regularly searching for something that returns zero results, we should probably pick that product up,” says Director of Marketing Justin French.

Send the query and its volume, then add the 90-day trend and whether a near substitute already sits in your catalog. That last column keeps the conversation honest, since a buyer needs to know whether the answer is a purchase order or a synonym.

Some rows can never be filled, and saying so is part of the handoff. BHFO receives significant search volume for a shoe brand it holds no online sales rights to, so rather than absorb the loss, it built a merchandising campaign presenting comparable products to shoppers searching that term.

Athos Commerce runs on-site product search, merchandising, and personalization on the same product data it optimizes for external channels, so a zero results report can be read against what the catalog covers without exporting anything.

Where to start this month

1. Pull 90 days of zero-result search queries and split them. The zero results report already sits in your reporting console, and the split is a judgment call per row rather than an analysis project. Athos Commerce provides Customer Success and Education at no charge, so a first pass with someone who has sorted these reports before costs nothing beyond the meeting.
2. Fix pile one’s top 20 by volume, and put product recommendations on the zero-results page. Do both in the same sitting. Recommendations recover sessions while the diagnosis is still in progress, and search configuration, product attributes, and merchandising rules all read the same product data on a single platform, so none of the three waits on the others.
3. Book a monthly pile-two send to whoever owns assortment. Include the substitute column. A recurring 20-minute slot beats a report nobody has agreed to read.
4. Report two rates instead of one. A blended zero-results rate can be driven down with synonym work while unserved demand grows unnoticed, because resolving rows makes the remainder a smaller share of a smaller total. Track queries you could have served separately from demand you cannot serve yet.

The report you already paid for

Every retailer running on-site product search owns a demand study that updates daily and costs nothing to run, reported in the shopper’s own vocabulary. Most read it as a maintenance queue. Sorting it into the queries you could have answered and the queries you could not turns one list into two decisions with two owners, and one of those decisions reaches the assortment.

That reading matters more as AI shopping agents begin querying product catalogs directly, since an agent gets no second attempt and no “did you mean” prompt when it cannot match a product to a request.

Sources

1. Scott, Edward. “Ecommerce Search UX Best Practices 2026.” Baymard Institute, updated April 29, 2026. https://baymard.com/blog/ecommerce-search-query-types.
2. Baymard Institute. “‘No Results Page’ Design Examples.” Baymard Institute, accessed August 24, 2026. https://baymard.com/ecommerce-design-examples/35-no-search-results-page.