Ecommerce site search: what helps people find the product

A store’s search works when the box is easy to spot, understands the way people actually type and never ends on an empty page with no way out. That matters because, in Baymard Institute testing, about half of participants prefer searching to browsing the menu. Even so, 56% of the stores Baymard benchmarked don’t support search well enough.

Why the first try decides

People who search want to go straight to what they’re after, not follow the path the site laid out. Search is also the escape hatch for anyone who got lost in the menu. That’s why Jakob Nielsen at NN/g wants search available on every page.

The same 2001 article includes a figure from a study of people shopping on several stores. The first query worked 51% of the time. The second worked 32% of the time, and the third only 18%. Almost half of those whose first search failed gave up right away (NN/g).

Baymard saw the same effect from another angle. When a product doesn’t show up, people decide the store doesn’t sell it and leave, even when the item is in the catalog (Baymard).

A box, not a link

NN/g’s advice is simple. On the homepage, search is a field you type into, not a link to another page. It goes at the top, usually in the right corner, and the left corner works almost as well. The field should be wide enough to fit a typical query without the text scrolling.

In code, HTML already has an element for this. <search> marks the search area as a landmark for screen readers, so you don’t need role="search" (MDN). The results themselves don’t go inside it. They belong in the page’s main content.

<search>
  <form action="/search/">
    <label for="q">Search the store</label>
    <input type="search" id="q" name="q" />
    <button type="submit">Search</button>
  </form>
</search>

What people type

Baymard splits store queries into eight types. Next to each one is the share of benchmarked stores with problems on that type. The examples are mine.

Query type	Example	Stores with problems
Exact, by name or model	“air fryer model X200”	12%
Product type	“sandals”	20%
Feature	“leather sofa”	39%
Use or occasion	“wedding gift”	43%
Abbreviation and symbol	“50 in tv”	54%
Compatibility	“case for phone model Y”	44%
Symptom or problem	“rug stain”	37%
Non-product content	“return policy”	66%

Three fixes cover a good part of the table. When a query matches a category, search can take the person straight to that category page. When it includes a feature, like color or price, that feature becomes a visible, pre-applied filter the person can remove. And search also covers shipping, returns and help pages. In Baymard’s accounts and self-service testing, 34% of participants used the field to find that kind of content (Baymard).

Suggestions while typing

The dropdown under the field is now expected, but few people click it. In NN/g’s ecommerce study, a suggestion was picked 23% of the times it appeared: 31 out of 136, with a 7% margin of error. It still helps as a reference, because it shows what the store carries and how a term is spelled.

The same article lays out the rules:

never suggest a term that leads to poor or empty results;
visually separate what the person typed from what the system filled in;
show the suggestion’s category when the store splits search by section;
on mobile, cut the list down to plain text suggestions.

Scope: useful, but risky

Scoped search limits results to one section, like “electronics only.” The risk is that people miss the limit and decide the product doesn’t exist. NN/g recommends defaulting to “all” and showing the scope at the top of the results. It also calls for a one-click way to widen the search to the whole site and a first suggestion with no scope at all.

Keep the query in the field

Few people get it right the first time, so they tweak the query. They change “dress” to “red dress,” drop a word, fix another. In Baymard’s testing, that happened 2.2 times on average. When results stayed poor, people usually tried three to five versions before leaving the store.

Still, 33% of desktop sites and 42% of mobile sites clear the term from the field after the search (Baymard). Then people retype everything, which costs even more on a phone. The fix is to leave the term in the field and repeat the query in the results page heading.

When nothing comes up

The no-results page isn’t the problem. The problem is when it turns into a dead end. NN/g asks for three things: say clearly that nothing was found, offer a way forward and don’t make fun of the person who searched. That way forward can be the term kept in the field, a spelling fix or a similar query that does return results.

Baymard adds that about 50% of stores fail on this page. Search tips alone aren’t enough, because few people read or apply them. What works, according to the same research:

categories related to the searched term;
alternative queries that return products;
recommendations based on what the person already viewed;
a visible sales phone number and chat;
popular products and categories.

How it shows up in my work

At Newfold Digital, in 2022, I developed a domain search and filtering system that streamlined the user selection process. The role is listed on my full profile. The way I see it, domain search is almost always an exact query: people type the name they want. When that name doesn’t work out, search and filters have to work together so the choice moves on without starting over.

At Vexor, the assistant finds an old file by what the business owner remembers about it, not by its exact name. It’s the same idea behind symptom queries: take the term the way the person thinks of it. Neither case comes with a measured number. They’re here as examples, not as proof.

Where the sources disagree

NN/g’s 2001 article says people who fail on the first query rarely succeed on later ones. Baymard’s 2021 testing shows people redoing their query several times. Both readings fit together. The rewording that works is small, like adding or dropping a word. That’s why keeping the term in the field helps so much (NN/g, Baymard).

Suggestions are another point of tension. NN/g measured few clicks on them. Baymard recommends suggesting categories precisely for people who search by problem and don’t know the product’s name. One source measures use, the other points to where suggestions help most.

The common mistake: never opening the search log

The log of what people type is the cheapest research a store has. NN/g suggests listing the most common queries and picking the best page for each one. Baymard uses the same log to learn how customers describe a problem. Without that reading, the team tunes search for the words it uses, not the words customers type.

To compare options once search has narrowed the list, see the post on comparison tables. What interrupts the purchase after the product is found is covered in ecommerce checkout.