A zero-result search is a query that returns nothing, so the shopper sees “no results found.” Fixing them starts with sorting each one into one of four causes: a typo, a synonym your catalog doesn’t use, an item you don’t carry, or a filter that narrowed the results to nothing. Each cause has a different fix, and the order you fix them in matters.
Most teams treat zero results as one problem and try one fix, usually adding synonyms. That recovers some of them. The rest stay broken because they were never synonym problems. This playbook shows how to tell the difference, what to fix first and how to measure whether it worked. It is the method I use in my e-commerce search relevance work.
The short version: pull the data, label your top 50 queries by cause, fix typos first, then synonyms, then query relaxation, and track four numbers before and after.
What a zero-result query costs
Nobody files a ticket for an empty results page. The shopper leaves, and the only trace is a line in a query log.
You can estimate the cost with your own numbers. Take the share of searches that return zero results, multiply by how many searches you get a month, then by your search-to-purchase conversion and average order value. Not every one of those searches would have bought something, so treat the result as a ceiling, not a forecast. The article on how failed searches leak revenue works through this sum with an example.
Step 1: Pull the data
You need two things: a list of queries that returned zero results, with counts, and the total search volume for the same period. Your search platform, analytics tool or server logs will have both.
Sort by count. A handful of queries usually account for a large share of the zero-result volume, so start from the top. Fixing the top 50 is a far better use of a week than reviewing the long tail. Two checks before you trust the list:
Is the count real?
Some sites count a results page that has products but all are out of stock as “zero results,” others don’t. Know which yours is.
Are bots in there?
Scrapers inflate zero-result counts. Filter them out.
Step 2: Sort each query into one of four causes
Take your top 50 or 100 and label each one by hand. It takes an hour or two and tells you where to spend the next month.
Cause 1: Typos and spelling variants
“Samsng charger,” “adiddas,” “colour” versus “color.” The product exists; the spelling doesn’t match.
Fix: typo tolerance, also called fuzzy matching, which allows a small number of character differences. Set it to be stricter on short words (a one-letter difference between two short words is often a different word) and consider a “did you mean” suggestion on the results page. This is the cheapest fix and usually recovers the most queries.
Cause 2: Words your catalog doesn’t use
“Couch” versus “sofa,” “sneakers” versus “trainers,” a brand nickname, “boho summer dress under $50.” The shopper describes the product; your titles use different words.
Fix: a synonym list for the obvious pairs, and meaning-based matching for the long tail (see hybrid search vs keyword vs vector search). Build e-commerce search synonyms from the zero-result log, not from a generic list, so you cover your own shoppers’ words. Together, synonyms, typo handling and meaning-based matching are what people call query understanding.
Cause 3: Things you don’t carry
“Wireless charger for iPhone 15” when you don’t stock one. No search fix creates inventory.
Fix: don’t leave a dead end. Show close alternatives, popular products in the closest category and a way to be notified. And count these queries: they are a demand list (see Step 5).
Cause 4: Over-filtering and over-strict matching
The shopper typed a long query and every word was required, or they selected filters that together match nothing. Search finds products for “red running shoes waterproof size 14 under $40” only if one exists with all of those attributes.
Fix: relax the query when it returns nothing. This kind of query rewriting drops the least important term, or requires most words rather than all, and shows what you relaxed (“Showing results for red waterproof running shoes”). Disable filter combinations that lead to zero, or show counts beside each filter so the dead end never appears.
Step 3: Fix in order of effort
Start at the top of this list. Effort depends on your platform: on OpenSearch or Elasticsearch most of the first three are configuration; on a hosted service they may be settings or paid features.
1. Typo tolerance, tuned by word length
Hours to days. Recovers Cause 1.
2. Synonyms from your own zero-result log
Days. Recovers the common pairs in Cause 2.
3. Query relaxation when results are empty
Days to a week. Recovers Cause 4.
4. A better “no results” page with alternatives
Days. Softens Cause 3.
5. Filter counts and dead-end prevention
A week or more. Recovers Cause 4.
6. Meaning-based (vector) matching alongside keywords
Weeks. Recovers the long tail of Cause 2. See how hybrid search combines both approaches.
These effort ranges are my rough guide; yours will differ with your platform and team.
Step 4: Fix the “no results” page itself
Some queries will always return nothing, and the page should still work. A useful empty page does four things:
Says what was searched
Plainly, without blaming the shopper.
Shows close matches
Relaxed results or close alternatives, if you have them.
Offers a way back
Popular products or categories.
Lets people ask for the item
An email capture, if you stock a wide range.
Skip the apologetic paragraph. The goal is a next click.
Step 5: Turn the log into a demand list
The queries you can’t fix are the most interesting. Picture “wireless charger for iPhone 15” searched 400 times in a week and bought by nobody (illustrative numbers, not a client result). That tells merchandising something no dashboard will.
Sort the unfixable ones by count and hand them to the people who buy stock and write content. Treat the list as a standing report, not a one-off clean-up.
Step 6: Measure whether it worked
Track four numbers, before and after each change:
Zero-result rate
Searches with no results divided by all searches.
Recovery
Of the queries you fixed, how many now return results and get a click.
Search exit rate
How often a searcher leaves right after the results page.
Search-to-purchase conversion
Compared with the rest of your traffic.
Change one thing at a time where you can, so you know what moved the number. And a lower zero-result rate isn’t automatically better: if relaxing the query returns irrelevant products, you’ve swapped an empty page for a bad one. Check clicks on the results you now return, not just the count. For a quick read on where your own search stands, the free e-commerce search audit scores it against a checklist, and the ten-point search health check shows what to look at first.
Common questions about zero-result searches
What is a good zero-result rate?
It varies by catalog, so I don’t give a single target. Compare your own rate over time and by query type, and aim to push down the fixable causes. Treat unfixable ones, such as items you don’t carry, as a separate list.
Are synonyms enough to fix zero results?
They fix one of the four causes: words your catalog doesn’t use. They do nothing for typos, over-filtering or missing items.
Should I show products on a zero-result page?
Yes, if they’re relevant. Show relaxed results or close alternatives with a clear label. Showing unrelated bestsellers can look broken.
Will AI or vector search fix zero results?
It helps with the long tail of wording problems. It doesn’t fix missing inventory or over-filtering, and it can return plausible but wrong products, so measure click-through on what it returns.
How often should I review the zero-result log?
Weekly at first, then monthly. New products and seasonal wording bring new misses.
Can you do this for my store?
Yes. I work with e-commerce and marketplace teams on search relevance. See the e-commerce and marketplace page for how I work.
