How often should the zero-result list be reviewed?

In most shops a weekly rhythm is the right compromise. It is tight enough to catch seasonal terms early and wide enough for the list to carry meaningful frequencies. Very large shops with several thousand searches a day benefit from a daily glance at the ten most frequent queries without results, because there a new trending term becomes measurable within hours. After an assortment change or a relaunch, a tighter cadence pays off for two or three weeks.

Should synonyms be created one-way or two-way?

One-way is the default case: the customer term points to the catalogue term, not the other way round. Two-way equivalences make sense only where the meaning genuinely matches, such as mobile and cell phone. With any fuzziness a two-way rule pulls results into the list that nobody asked for, and result quality degrades gradually without a visible fault. That is why every new rule deserves a short counter-check with five typical queries.

How high may the share of searches without results be?

There is no universally valid target, because assortment breadth and customer mix influence the figure strongly. What carries meaning is the movement: a share that declines over weeks shows that upkeep is working. If it jumps, that is almost without exception a technical signal - an aborted index run, a field lost after an update, or a search component that works with different defaults after an upgrade.

What belongs on a good zero-result page?

The entered term visibly, so that a typo is spotted; a suggestion for an amended query where the search component knows one; two to four entry points into matching categories or popular articles; and, for recognisable content questions, a pointer to shipping, returns or contact. The search box stays filled and focused so the second query needs no detour. The page should return a clean status code and stay out of search engine indexes.

Why does search miss articles that exist in the shop?

Four causes cover most of it. The attribute searched for lives only in the description text and in no searchable field. The customer term is a synonym that has not been stored. The index is stale or incomplete because its build aborted. Or search requires all entered words to match, and a single unsuitable word wipes out the entire result list. The order of investigation should follow exactly that list.

Can search upkeep be part of a maintenance contract?

Yes, and it suits one particularly well because it is plannable and measurable. Typically agreed are the review cadence, the number of queries handled per pass, monitoring of the index age with an alert threshold, and a monthly report showing the trend of the zero-result rate. That turns a task without urgency into a service with evidence which can be verified against the movement of the metrics.

Shop search upkeep: zero results and synonyms

Someone types three words into the search box, presses enter and reads: no results found. For the shop this is an unremarkable event with no error message, no log entry, no spike in monitoring. For the visitor it is the end of the shopping trip. Search is the shortest path from a need to an article, and it is the one place where customers tell the shop in their own words what they want. That is exactly why it is a maintenance object: a catalogue changes, vocabulary changes, and an index nobody has touched since the last relaunch slowly loses contact with the language of its customers.

Key takeaways

Zero results are not random, they are a list. Reading it weekly produces a work list for the catalogue, synonyms and content, sorted by frequency.
The most common causes are linguistic, not technical: spelling variants, missing synonyms, attributes that live only in body text, and questions that mean no product at all.
Synonyms should mostly be maintained one-way. A two-way rule quickly pulls results into the list that nobody asked for.
The search index has an age. Without watching that age, search sells yesterday's stock and prices while the product page already shows today's.
Search upkeep becomes reliable once it has a cadence, an owner and three or four metrics in the maintenance report.

Search is the shortcut through the catalogue

Whoever uses the navigation works through the structure the shop provides. Whoever searches skips that structure and phrases the request personally. That makes search queries the most honest data source in the entire shop: they contain the words of the customer, not the words of the catalogue. The whole upkeep effort - and the whole return - lies in that difference. A shop whose search understands the language of its visitors saves each of them several clicks; a shop whose search knows only its own catalogue vocabulary loses exactly the visitors who already knew what they wanted.

How widespread the gap is shows in the ongoing benchmark of search usability in online retail: 56 percent (Baymard Institute) of the audited sites support their users' search needs inadequately. The study draws on 170 (Baymard Institute) benchmarked sites and apps. Broken down by device, 46 percent of desktop sites, 58 percent of mobile sites and 64 percent of apps (Baymard Institute) rate as mediocre or worse overall. That is not a fringe group but the majority - which means a well-maintained search box is a comparatively cheap point of difference.

What counts as search upkeep in this article

Everything between the entry in the search box and the result list that comes back: the search index and its refresh cadence, the word lists for synonyms, spelling variants and abbreviations, the mapping of product attributes to searchable fields, the handling of queries without results, and the evaluation of what visitors actually type. Not included are recommendation rails, filter logic and the sorting rules of category pages - those have their own rules and their own upkeep.

Zero results are a list, not an incident

The first step of search upkeep costs nothing but access: an evaluation of the queries that returned zero results, sorted by frequency. Nearly every shop system and every search component can produce this list; in many installations nobody reads it. Yet it is the only place in the business where customers write down, unprompted and in their own words, what they miss in the shop. Six causes typically explain most of that list, and each one leads to a different upkeep step.

Spelling and typing errors

Split instead of joined, hyphen instead of space, one transposed letter. The query is correct in substance but matches no indexed word. The remedy is a variant list, not fuzzy matching across the whole catalogue.

The synonym is missing

The catalogue says hooded sweatshirt, the customer says hoodie. Both mean the same shelf. Without a stored term pair the list stays empty although the article is in stock.

Attribute only in body text

Wall thickness, connector type or cut are described in the copy but exist in no searchable field. Searching for the attribute returns nothing, even though the text describes it.

Content question, not a product

Return period, delivery time, invoice payment: queries that mean no product. A pure product search finds nothing for them and comes across as clumsy.

Article discontinued

The product existed, the demand still does. The successor carries a different name that nobody knows. Without a pointer the search ends in emptiness.

Not in the assortment

The most honest category and the most valuable one for purchasing: documented demand, sorted by frequency, for articles the shop does not carry yet.

Synonyms, spellings and abbreviations

The biggest lever sits in the word list, and the biggest mistake is to create it once. Retail language moves: brands swap designations, technical terms drift into everyday speech, and every season adds new short forms. In the search usability benchmark these linguistic query types rank among the weakest - for abbreviations and symbols 54 percent (Baymard Institute) of audited sites show weaknesses, for use-case queries 43 percent (Baymard Institute) and for feature queries 39 percent (Baymard Institute). Six kinds of rule cover daily practice.

One-way mapping

The customer term points to the catalogue term, not the other way round. Searching for hoodie finds hooded sweatshirts - but searching for hooded sweatshirt does not suddenly return everything ever labelled hoodie. This is the default case.

Two-way equivalence

Two terms are treated as interchangeable. Sensible only where the meaning truly matches, such as mobile and cell phone. With any fuzziness the rule pulls in results that nobody asked for.

Spelling variants

Joined, split, hyphenated, with and without accents in transliteration. This list grows out of the zero-result list and needs observation rather than theory.

Units and symbols

Inches, centimetres, volts, the inch mark, an x or a cross between two measurements. Without normalisation you lose exactly the queries that sit closest to a purchase.

Stop words that carry meaning

Without, for, against: short words that search engines like to discard although they carry the sense. A query for cream without fragrance otherwise turns into a query for cream with fragrance.

Accessory relations

Spare part, battery, filter, bag: queries that mean an accessory to a main article. Without a maintained relation between the two the list stays empty although the part is in the assortment.

These rules deserve documentation, not just entry. Every line in the synonym list needs a date, a trigger and a note on which query produced it. Without that provenance the list turns into an inventory after two years that nobody dares to touch, because it is unclear which rule causes which side effect. The same care pays off twice in multilingual shops: a synonym list maintained in one language only creates exactly the gaps in the second language that the article on translation gaps in the shop describes.

One-way is the default case

Two-way equivalences feel convenient because one line covers both directions. They are also the most common reason for result lists that degrade gradually: equate bag and backpack, and every bag query returns backpacks too - the result list loses its precision without anybody spotting a fault. Check a handful of typical queries after each new rule before you move it into the live index.

The index and its age

In most shops search does not work on the database but on a dedicated index - a copy of the catalogue optimised for fast text matching. That copy trails the original by some margin. By how much is an operational figure that few people know and few people watch. As long as the refresh runs every minute, nothing shows. If it breaks, search displays week-old prices and stock levels while the product page already states the current ones - and the contradiction is noticed by customers first, not by the operator.

The connection to other maintenance topics is close. Building the index is a background process like any other and shares its failure patterns, as the article on silent job failures in cron jobs and imports describes. While the refresh runs it can put noticeable load on the database - a topic of its own in database maintenance for shops. And a search component installed as an extension belongs in the same view as every other extension, for which the plugin audit sets the frame.

Reading index age and the zero-result list

# Age of the search index in seconds against the last catalogue write
LAST_CATALOGUE=$(mysql -N -B shop -e \
  "SELECT UNIX_TIMESTAMP(MAX(updated_at)) FROM product;")
LAST_INDEX=$(cat /var/lib/shop/index/last_run.stamp)
echo "Index age: $(( LAST_CATALOGUE - LAST_INDEX )) seconds"

# The twenty most frequent queries without results over the last seven days
mysql -t shop -e "
  SELECT term, COUNT(*) AS queries
  FROM search_log
  WHERE hits = 0 AND created_at > NOW() - INTERVAL 7 DAY
  GROUP BY term
  ORDER BY queries DESC
  LIMIT 20;"

Two numbers, one signal

A search index needs two measurements, not one: the age of its last complete run and the number of documents it contains. The age shows whether the refresh still runs. The document count shows whether it ran completely. An index that finishes on time while holding only half the articles is more dangerous than one that visibly stalls.

Eight query types, eight checks

Search upkeep becomes plannable once queries are separated by type rather than by wording. The benchmark mentioned distinguishes eight types and reports for each how many audited sites show weaknesses. That breakdown works directly as a test plan: for every type you phrase three to five example queries from your own assortment and look at what search returns. The pass takes a good hour and yields a solid baseline.

Query type	Example from the assortment	Sites with issues
Exact designation	Model number, full article name	12 percent
Product type	Women's jeans, coffee machine, wall shelf	20 percent
Feature	Waterproof jacket in blue	39 percent
Use case	Gift for coffee drinkers	43 percent
Abbreviation and symbol	TV instead of television, inch mark instead of inch	54 percent
Compatibility	Battery matching a device model	44 percent
Symptom	Cause described instead of product	37 percent
Non-product	Return period, invoice payment, contact	66 percent

The last type deserves particular attention because it is underestimated most often. Non-product queries score worst at 66 percent (Baymard Institute), and in usability testing 34 percent (Baymard Institute) of participants used the search field for content that is not a product at all - the returns policy, say, or unsubscribing from a mailing list. Pointing the search box exclusively at the product catalogue lets a third of its usage run into the void. The remedy is unspectacular: take content and service pages into the index and mark them out as a separate section in the result list.

The zero-result page: dead end or signpost

Even with the best upkeep a residue of queries without results remains. The question is therefore not how to abolish zero results entirely, but what the shop offers at the moment of a zero result. Here the benchmark paints a similarly unfavourable picture: almost 50 percent (Baymard Institute) of the audited sites offer no effective way to recover from a search that yields nothing. An empty page stating that nothing was found is technically correct and commercially a dead end.

A usable zero-result page answers three questions at once: what did I search for, why did nothing come back, and how do I move on. It repeats the search term visibly so that a typo becomes obvious. It offers an amended query if the search component knows a similar one. And it leads into a selection close to the request - a category, the bestsellers of the matching area, or customer service where the query was clearly a question. Technically the page belongs in the same view as other dead ends in the shop, as described in maintaining 404 pages and redirects.

The entered term appears visibly on the page so that a typo is spotted at once.
A corrected or broadened query is suggested where the search component knows one.
Two to four entry points lead onwards: matching category, popular articles, new arrivals.
Where the shop recognises a content question, the page points to shipping, returns or contact.
The search box stays filled and focused so the second query needs no detour.
The page returns a clean status code and is kept out of search engine indexes.
Every call is logged with term, timestamp and device type and feeds the weekly list.

A working rhythm for search upkeep

Search upkeep rarely fails on skill and often on cadence. It is an activity without urgency: nobody calls because a synonym is missing. That is why it needs a fixed slot in the maintenance calendar, the way updates and backups have one. Five steps have proven themselves as a weekly routine and cost around an hour in a mid-sized shop.

Step 1

Pull the list

Export the queries without results from the past seven days, sorted by frequency, from the shop system or the search component. Twenty lines are enough; below that the return per line drops quickly.

After a few weeks the character of the list changes visibly. Linguistic causes decline because the word list grows. What remains is the hard core: articles the shop does not carry. That is no failure but the actual result - documented demand that carries more weight in a purchasing conversation than any assumption. Anchoring search upkeep in a maintenance contract with committed response times turns that routine into a dependable service rather than a good intention.

Metrics for the maintenance report

For the work to become visible it needs numbers. Four to six figures suffice, and all of them can be derived from the search log without building extra measuring points into the shop. What matters is the view as a trend: a single zero-result rate says little, its movement across twelve weeks says a lot.

Share of searches without results among all searches, weekly and split by mobile and desktop.
The twenty most frequent queries without results, with their change against the previous week.
Number of newly created synonyms, spelling variants and pointers per week.
Age of the search index in minutes plus the document count per run.
Share of searches that lead to a product page view - the actual success figure.
Number of non-product queries and their share of all searches.

The last figure is the most interesting because it points beyond search. Where queries about delivery time, returns or invoice payment accumulate, that indicates the information is hard to find in the shop. Search upkeep thus exposes gaps that really belong to navigation and service copy. The same direction shows in the observation that questions about withdrawal are regularly asked through the search box - how the corresponding element is built in and tested is described in the article on the withdrawal button in the shop.

The commercial frame makes the effort comprehensible. Gross goods revenue in German e-commerce reached 83.1 billion euros (bevh) in 2025, and revenue in mail order and internet retail rose by 10.1 percent (Destatis) in real terms over the previous year, with the federal statistical office pointing to a special effect from the restructuring of a larger company. At the same time 43 percent (Bitkom) of online shoppers call the size of the assortment and 42 percent (Bitkom) call product presentation and usability very important when choosing a shop. A large assortment that cannot be found through search feels small to a visitor - so both criteria meet precisely in the search box.

Historically the direction is encouraging. A recurring study of search in online retail measured search success of 92 percent (Nielsen Norman Group) in its 2017 round, against 74 percent in the 2011 round and 64 percent in the year 2000. The distance between that success rate and today's findings on usability is explained by the fact that a technically successful search is not yet a usable result list. The upkeep work sits exactly in between.

Legal edges of search upkeep

Search has a legal side that is rarely considered in practice. Where the result list shows an article as available that has long been gone, that concerns the essential characteristics of goods. Availability is expressly among them under Section 5 subsection 2 (UWG) of the German act against unfair competition - which makes a search index that has not completed a full run for days more than an operational matter. The same care that the article on maintaining legal texts describes for withdrawal and price statements applies by analogy to what search claims about availability.

Accessibility adds to that. Services in electronic commerce fall under Section 1 subsection 3 (BFSG) into the scope of the German accessibility act, and the search box is one of the most used controls of all. Labelling, focus order, announcing the result count to assistive technology and a zero-result page that does not carry its message through colour alone therefore belong in the same review as the rest of the shop - the article on accessibility in the shop puts that in context. How stable search stays under load is in turn a question of load testing before peak traffic, because during the Christmas season search is one of the most expensive queries in the entire system.

A search without results is not a customer mistake, it is an open task in the catalogue. Reading it hands you a pre-sorted work list every week.

Principle of search upkeep

It starts with an inventory: which search component runs in your shop, at what cadence is its index built, which fields are searchable, and how many queries currently run into the void? Out of that inventory comes the upkeep plan with a weekly rhythm, owners and metrics. We set up the evaluation, take over the ongoing upkeep of synonyms and attributes and watch the age of the index together with the rest of your shop monitoring - the evaluation of the logs is described in the article on log monitoring and alerting, and how all of this affects load times is shown by performance maintenance. A short conversation is enough for a first assessment.

Sources and Studies

This article is based on data from Baymard Institute, Bitkom, bevh, Destatis and Nielsen Norman Group. The figures quoted refer to the state of the respective publication.