Ecommerce site search best practices, a practitioner's checklist

Ecommerce site search best practices come down to a short list: make the search field easy to find and use, handle the ways shoppers type, fix zero-results searches every month, and measure search as its own conversion funnel. The checklist below covers each one, starting with a 30-minute audit you can run on your own store today. Grab the free audit scorecard, a Google Sheet you can copy, to log your results as you go.

In Constructor’s analysis of 609 million searches across 113 retail sites in Q4 2024, shoppers who searched made up 24% of visitors but brought in 44% of revenue, and they converted at 2.5 times the rate of shoppers who didn’t search. That’s Constructor’s data on its own customers, so use it to size the opportunity and measure your own store for the real numbers. Even as a rough guide, it says a quarter of your traffic may be carrying close to half your sales through one input field.

Why site search deserves its own CRO attention

Most conversion work focuses on product pages, the cart, and checkout. Search sits upstream of all three, and it fails quietly. A broken checkout throws an error that somebody reports. A search that returns the wrong products just sends a shopper away without a word.

On-site search CRO treats the search experience as a funnel with its own metrics, owners, and tests. If you already run a conversion rate optimization program, search should definitely be one of its workstreams.

How much revenue poor search costs you

No credible public number exists for revenue lost to poor site search. Figures that float around vendor blogs usually don’t cite a source, or they come from studies more than a decade old. You can estimate your own exposure with numbers you already have:

Monthly searches × zero-result rate × search conversion rate × average order value = monthly revenue at risk

Say your store gets 50,000 searches a month, 8% of them return nothing, searchers convert at 3%, and your average order is $85. That’s 4,000 dead-end searches. If those shoppers had converted like your other searchers, they’d have placed about 120 orders, or roughly $10,200 a month. The formula only counts searches that return nothing. Searches that return the wrong products lose sales too, so your real number is higher. That makes this a conservative estimate, which is the kind you want in a business case for search work.

Most stores still miss common query types

Baymard Institute’s 2026 search benchmark of more than 170 ecommerce sites and apps rated 56% of sites mediocre or worse on search UX overall. The weak spots cluster around particular kinds of queries: exact product names mostly work, while searches for a use case, a compatible part, or a store policy often fail. The audit below tests each type.

Run a 30-minute search audit

You don’t need a vendor or a data team for this. Open your store on a laptop and a phone, and run the searches below the way a shopper would. Make a copy of our free site search audit scorecard first. It’s a Google Sheet with tabs for the audit, the monthly reviews, and your search metrics, and it calculates your pass rates and revenue at risk for you.

First, rule out a technical failure

Before you judge relevance, check that search works at all. Run these three tests first:

• Search for a product you know is in stock. If nothing comes back, the problem is how your catalog connects to your search tool. Fix that disconnect first, because ranking and synonym changes can’t help until search returns results.

• Time the results. If a results page takes more than a couple of seconds to load on a phone over a cellular connection, fix speed before anything else.

• Search for something you added to the catalog this week. If it’s missing, your index isn’t syncing with your product feed often enough.

If any of these fail, send it to whoever set up your search tool or platform app, and come back to the rest of the audit once it’s fixed.

Test the eight query types

Baymard groups shopper searches into eight types. Pick two or three real queries for each type from your own catalog, run them, and mark whether the first page of results would satisfy a shopper.

Score each query as a pass or a fail. Exact and product type searches are the ones most stores handle well, so expect most of your failures among the other six, especially the last four. Those failures become your fix list for the sections below. Failures on product type and feature queries usually trace back to filters and product data, which the filters section and the product data section near the end of this guide cover.

Search bar placement and UX fundamentals

Make the field visible, wide, and labeled

Shoppers who want to search look for a field, so give them an open text field at the top of every page. A magnifying glass icon that expands on click adds a step on desktop, and shoppers who don’t see a field may assume you don’t have search. Make the field wide enough to show a typical query in full. Put placeholder text inside it that hints at what works, like “Search jackets, boots, or order status,” and pair it with a visible search button.

Keep the query in the field on the results page

When a shopper lands on results, their query should still be in the field so they can fix a typo or add a word. Baymard’s February 2021 research found that on 33% of desktop sites and 42% of mobile sites, search terms are cleared once the search is submitted. Clearing the field makes a shopper retype everything to refine a search, and on a phone, a lot of them won’t bother. That figure is from 2021, and the trend has been improving, but it takes a minute to check your own store.

Let shoppers search within a category

A shopper browsing women’s running shoes who types “waterproof” probably wants waterproof running shoes. Give them a way to keep that scope. Pick at least one of these:

• A scope option in the search field. On the running shoes page, the search box shows a small label or dropdown that says “Search in: running shoes.” The shopper can leave it on or switch it to “All products.”

• Category suggestions in autocomplete. When the shopper types “waterproof,” one suggestion reads “waterproof in running shoes,” so a single tap gives them the narrower search.

• A prompt on the results page. The shopper runs a store-wide search, and the results page offers a link like “Only show results in running shoes.”

When a query exactly matches a category name, like “rain jackets,” consider sending the shopper straight to that category page, as long as the category page has filters and sorting.

Autocomplete, typeahead, and query suggestions

Autocomplete, which many search tools call typeahead, is the dropdown that appears under the search field as a shopper types. It can suggest completed searches, called query suggestions, like “rain jacket women’s” after a shopper types “rain.” It can also suggest categories to jump to and individual products. Done well, it saves shoppers typing and steers them toward searches that return results.

Baymard’s 2022 autocomplete research found 80% of ecommerce sites offer autocomplete, and only 19% get all the implementation details right. The misses tend to be small, like a list too long to scan or rows too cramped to tap accurately. In Baymard’s testing, details like these were enough to make shoppers skip the list and type out their whole query, or tap the wrong suggestion and land on results they didn’t want. The sections below cover what to get right.

What to show in suggestions

There are three types of suggestions we recommend adding to your search to improve user experience:

• Query suggestions

• Category suggestions

• Product suggestions

Start with query suggestions, meaning completed search phrases drawn from what shoppers search and buy. You don’t have to write query suggestions by hand. Most search tools build them from your store’s own search history, ranking the phrases shoppers type most often. Algolia’s Query Suggestions, for example, pulls the most frequent searches from your search analytics and can also combine catalog attributes, like brand and product type, into new phrases. If your tool lets you shape the list, favor popular searches that lead to clicks or purchases, and use catalog-based phrases to cover new products that don’t have search history yet.

Add category suggestions so a shopper can jump to “boots in women’s” without a results page in between. They help most on broad searches, where one word could mean several departments. A shopper typing “boots” might want women’s boots or kids’ rain boots, and a category suggestion lets them choose before they land on a mixed results page. Show a couple of categories under the top query suggestion, ideally the departments where shoppers who search that phrase most often click or buy. Style them differently from query suggestions, with “in” before the category name or a small label, so shoppers can tell a search apart from a jump to a department. Baymard’s autocomplete research recommends this distinction for that reason. Category suggestions also do the scoping work described in the section on letting shoppers search within a category.

Product suggestions show specific items, with a thumbnail and price, right in the dropdown. They help most when the shopper already knows what they want, like someone typing “Hoka Clifton,” because one tap takes them straight to the product page and skips the results page entirely. Keep them to a handful, three or four on desktop and fewer on mobile, and place them beside or below the query suggestions so they don’t crowd out the searches a shopper might still want to run. Choose which products appear with the same signals you use for ranking, like sales and inventory, and leave out anything that’s out of stock unless you label it clearly.

A few rules apply to the whole dropdown. Keep the full list short, around 10 items on desktop and fewer on mobile, so it doesn’t push the page out of view. Style the part of each suggestion the shopper hasn’t typed yet differently from what they’ve already typed, so they can scan the completions. And leave out suggestions that return zero results. Check your autocomplete source against your no-results log every month, covered later in this guide.

Handle misspellings in autocomplete too

Shoppers lean on autocomplete most when they aren’t sure how to spell something, which is the moment many implementations give up. In Baymard’s 2026 mobile benchmark of more than 150 sites, 28% of mobile sites don’t offer relevant autocomplete suggestions for closely misspelled queries. If your search engine tolerates typos on the results page, make sure the autocomplete does too. The section on typo tolerance below covers how to set it up without creating false matches.

Keyboard and tap behavior

Shoppers use autocomplete as a starting point and then edit it. On desktop, when a shopper arrows down to a suggestion, copy that suggestion into the search field so they can add a word before pressing Enter. On mobile, add a small arrow or plus button next to each suggestion that fills the field without submitting the search. Baymard’s August 2024 research found 58% of sites fail to support users attempting to use autocomplete as a starting point for query generation.

Typos, synonyms, and what shoppers actually type

Search tools that understand full sentences rely on natural language processing, which goes deeper than this guide does. This section sticks to what you can control: typo tolerance, synonyms, and how you store units and sizes.

Typo tolerance without false matches

Typo tolerance lets “jeens” find jeans. Set it too loose, though, and “boot” starts matching “boat.” Most search tools scale tolerance with word length, and the defaults vary by vendor. Algolia’s defaults allow one typo in words of four or more characters and two typos in words of eight or more. Elasticsearch’s AUTO fuzziness setting allows one edit in terms of three to five characters and two edits in longer terms. Neither is an industry standard, so check what your platform does before you change anything.

Then turn typo tolerance off where precision matters. SKUs, model numbers, and part numbers should match exactly, because a one-character difference is usually a different product. Short brand names and numeric values deserve the same treatment. Algolia, for example, lets you switch off typo tolerance for specific attributes, specific words, and numbers. Your search tool may also offer fuzzy or phonetic matching as settings. Test either one against your own catalog before you leave it on.

To check whether your settings work, review your zero-results log. Misspellings that still return nothing mean tolerance is too strict. Complaints about irrelevant results for short words usually mean it’s too loose.

Build synonym lists from your query log

Your shoppers’ vocabulary is in your search log. Export the last 90 days of queries and look for words that should match products but don’t: “sneakers” when you call them trainers, “couch” when you call it a sofa, “hoodie” when the product title says pullover. Add those as synonyms.

Decide which way each synonym should work. Two-way synonyms like sofa and couch match each other. One-way synonyms point a broad term at a narrower one, so “outerwear” finds jackets, and “jacket” doesn’t return every vest and raincoat. Review the list quarterly, because seasons, trends, and new product lines change how people search.

Abbreviations, symbols, and units

Baymard’s benchmark found abbreviation and symbol queries are among the most mishandled, with 54% of sites having issues. Shoppers type “13in,” “13 inch,” and “13"” interchangeably, along with “XL” and “extra large,” or “3/4 sleeve” and “three quarter sleeve.” Normalize units and sizes in your product data, then map common abbreviations to them. If you sell anything with dimensions, test every format a shopper might use.

Use-case, compatibility, and symptom queries

These are the queries where a shopper describes a need: “dress for a beach wedding,” “charger for a Dell XPS,” “shampoo for dry scalp.” Keyword matching struggles with them because the words in the query often aren’t in the product title. Baymard’s numbers show the gap: 43% of sites have issues with use-case queries, 44% with compatibility queries, and 37% with symptom queries.

Product data does most of the work here. Add use-case tags like “wedding guest” or “travel,” compatibility fields that list the models a part fits, and benefit attributes like “for dry scalp” to your catalog, then make those fields searchable. For your highest-volume needs, a curated landing page that the search redirects to can work better than a results page. Some of these queries are questions in disguise, like whether a charger works with a specific laptop. Tags help search match them, and for the ones a product page can’t settle, an AI assistant like Gladly can answer the shopper directly. More on that at the end of this guide.

Fix zero-results searches before they cost a sale

A zero-results page is the clearest signal search gives you, because the shopper told you exactly what they wanted and you had nothing to show them.

Triage the no-results log every month

Pull every query that returned zero results in the last month and sort by volume. Your search tool’s analytics or your platform’s search reports usually include a list of searches with no results. Shopify’s Search & Discovery app has one, for example. If you rely on GA4, you’ll only have this list once you’ve set up the results count described in the measurement section below. Then put each query in a bucket:

• Synonym or spelling gap. You carry the product, but the shopper used a different word or misspelled it. Add a synonym or adjust typo tolerance.

• Missing product. You don’t carry it. Send the list to merchandising, because repeated searches for something you don’t sell are free demand data.

• Content question. The shopper searched for information, like “size chart” or “warranty.” Make sure that content is indexed or redirected.

• Redirect. The query is so specific that a single page answers it, like “gift cards.” Set up a redirect.

Rerun the top queries after each fix. Then fix the page itself: a zero-results page should keep the query in the field, suggest alternative spellings or broader searches, show popular categories, and offer a way to ask a person or an assistant for help. Don’t leave it blank.

Route policy and order questions to an answer

Shoppers use the search bar for more than products. They type “return policy,” “where is my order,” and “do you ship to Canada.” Baymard’s benchmark found that 66% of sites have issues with these non-product queries, the worst rate of the eight types.

At minimum, index your help content so policy searches return the right page, and use knowledge base best practices so those pages answer the question. Some questions can’t be answered by a page, though. “Where is my order” needs that shopper’s order, and “can I return opened items” often needs a judgment about their specific item. Those queries need a route to something that can look up the order or apply the policy, whether that’s a person or AI. More on that at the end of this guide.

Filters and search result merchandising

Filters on the results page

Filters let shoppers narrow a broad search like “dress” to the 40 products that fit. Show filters on the results page that match the query, so a search for boots offers heel height and shaft height, and a search for tents offers capacity. Keep applied filters visible and easy to remove, and make sure filters are reachable on mobile without scrolling to the bottom of the page. Our guide to faceted search best practices covers filter choice, ordering, mobile layout, and zero results caused by filter combinations in detail.

Give every merchandising rule an owner and an end date

Merchandising rules boost, bury, or pin products in search results. They’re useful for launches, promotions, and clearance. They’re also how a search experience slowly gets worse, because rules pile up and nobody remembers why a winter coat is pinned to the top of “jacket” results in July.

Treat each rule like a small campaign. Record who created it, why, which queries it affects, and when it ends. Set an expiration date on every boost and bury rule, and review any rule without one every month. Keep the total number of active rules small enough that one person can review them all in an hour. And freeze rule changes on any query that’s part of a running A/B test, because a new boost halfway through a test makes its results meaningless. The audit scorecard has a tab for logging rules and flags any that have expired or have no end date.

Mobile site search vs. desktop

Mobile search is harder to get right, and most stores haven’t. In Baymard’s 2026 benchmark, search UX performance was mediocre or worse on 46% of desktop sites, 58% of mobile sites, and 64% of apps. Everything earlier in this guide applies to mobile too, but a phone makes a few problems worse: typing takes more effort, the screen fits fewer results, and some shoppers would rather not type at all. The sections below cover each one.

Typing costs more on mobile

Every character takes more effort on a phone keyboard, so shoppers type shorter queries and make more typos. That makes autocomplete and typo tolerance more important on mobile than on desktop.

A few small changes to the search field make typing on a phone easier. Each one is a quick fix for whoever manages your site’s code:

• Show a “Search” key on the keyboard. Mark the field as a search box with input type="search" and add enterkeyhint="search", so the return key on the phone keyboard reads “Search.”

• Stop automatic capital letters. Set autocapitalize="off" so the keyboard doesn’t capitalize the first letter of every search.

• Consider turning off autocorrect and spellcheck in the field. Phone keyboards like to “correct” brand names and model numbers into regular words, which sends the search in the wrong direction. Your search tool’s typo tolerance is a better place to handle misspellings.

• Put a visible submit button next to the field. Some shoppers look to the page for a search button and never notice the one on the keyboard. Baymard’s 2026 mobile benchmark found 27% of mobile sites don’t place a submit button next to the search field.

Voice search is also worth a look, since speaking is the easiest way to skip the keyboard. Most phone keyboards already have a microphone button that dictates into any text field, so some shoppers are speaking searches into your site today without you building anything. Those queries tend to be longer and more conversational, like “waterproof hiking boots for wide feet,” and dictation can mangle brand names. The same fixes covered earlier handle both: typo tolerance, synonyms, and product data that covers use cases. A dedicated microphone button in your search field is an option once those work, but browser support for building one yourself is uneven, so don’t build around voice before your search basics are solid.

Less screen, fewer results

A phone shows two to four products per screen, so the first few results carry more weight than they do on desktop. Lead with the most relevant products and keep product titles short enough to read in a two-column grid. Broad queries like “shirt” are where mobile shoppers most need filters, so keep the filter button pinned where a thumb can reach it as the shopper scrolls. The faceted search guide covers mobile filter layout.

Let shoppers stop typing

The best mobile search is sometimes a conversation. A shopper who types “gift for my dad who hikes” is asking a question that a results page answers poorly, and a person answers well. Guided search, quizzes, and conversational assistants let shoppers describe what they need in their own words and get a recommendation back. On a small screen, a few taps or a short message is a lot less work than five rounds of query, filter, and back button.

Measuring site search performance

The search funnel and six metrics

Map search as a funnel with five steps: search, click a result, view a product page, add to cart, and purchase. Then track these six metrics against it:

• Search usage rate: the share of sessions that include at least one search. This tells you how much traffic flows through search.

• Search conversion rate: the share of sessions with a search that end in a purchase. This is your main outcome metric.

• Revenue per search: search-driven revenue divided by the number of searches. It catches changes in order value that conversion rate misses.

• Zero-result rate: the share of searches that return nothing, which means the funnel failed at its first step.

• Search exit rate: the share of searches after which the shopper leaves the site from the results page. High exit with a low zero-result rate usually means results came back but weren’t relevant.

• Refinement rate: the share of searches followed by another search in the same session. Some refinement is normal. A spike means the first results aren’t landing.

Segment every metric by device, and segment search conversion rate by the eight query types from the audit. A store with a healthy overall rate can still be failing compatibility searches, and those shoppers often know exactly what they want to buy. For a deeper look at the overall number, see our guide to ecommerce conversion rate.

Setting it up in GA4

GA4 can track site search through enhanced measurement. When a URL contains a search query parameter such as q, s, or search, GA4 records a view_search_results event with a search_term parameter. First, check that your store’s search parameter is in the list in your data stream settings, and add it if it isn’t. Once search events are coming in, set up three more things:

1. Register search_term as a custom dimension. Make it event-scoped so you can report on what shoppers searched. It only collects data from the day you register it, so do this one right away.

2. Build a funnel exploration. Use the steps view_search_results, select_item or view_item, add_to_cart, and purchase to see where searchers drop off.

3. Send a results count with each search. GA4 doesn’t know how many results a search returned, so have your developer push that number into the data layer with each search. That lets you filter for searches that returned zero and calculate your zero-result rate.

For search usage rate, build a free-form exploration with Event name and Device category as dimensions and Sessions as the metric, then filter to view_search_results. That gives you the number of sessions with at least one search on each device. Divide it by total sessions for the same dates.

For revenue per search, apply a segment of sessions that include view_search_results, take purchase revenue for that segment, and divide it by the number of view_search_results events. This counts all revenue from sessions with a search, so treat it as an upper bound on what search drove.

For refinement rate, subtract the number of sessions with a search from the total number of view_search_results events, then divide by total view_search_results events. Every search except the last one in a session was followed by another search.

GA4 has no built-in equivalent of the search exits report from Universal Analytics. You can approximate it in a path exploration that starts from view_search_results and shows what shoppers did next, or estimate it as sessions where the results page was the last page viewed.

Tools worth using

Pick tools based on how mature your search program is.

• Built-in platform reports and GA4. Start here. Shopify’s Search & Discovery app, for example, includes reports on searches by query, searches with no results, and searches with no clicks, along with click and purchase rates. These show what shoppers searched for and whether they clicked. But they don’t show why a shopper gave up.

• Session replay on results pages. Watching recordings of shoppers on your results pages shows what the numbers can’t: scrolling past irrelevant products, opening and closing filters, retyping the same query. Filter recordings to sessions with a search and a quick exit.

• Search vendor query analytics. If you use a dedicated search platform, its analytics usually include query-level click-through, revenue attribution, and relevance testing. These are the deepest, but they only measure what happens inside that vendor’s search.

Whatever you use, track all six metrics and share three with your team or leadership monthly: search conversion rate, zero-result rate, and revenue per search. They cover outcome, failure, and value without burying anyone in dashboards.

Build a continuous testing and relevance process

Search relevance drifts over time. New products arrive with thin data, seasons change what shoppers type, and merchandising rules pile up. Keeping search sharp takes a routine, and the order matters. Start with your product data, because every ranking setting and test depends on it. Then review results monthly to catch what slips, and A/B test the changes big enough to prove.

Fix product data before you tune ranking

Search can only match what’s in your product data. If half your dresses don’t have a sleeve length attribute, no ranking setting will make “long sleeve dress” return all the right ones. Before you adjust relevance, fill gaps in the attributes shoppers search for most: size, color, material, fit, compatibility, and use case. The same clean data helps AI shopping agents read your catalog, which our guide to preparing your product feed for AI agents covers.

Once the data is in shape, most ranking comes down to four signals:

• Popularity

• Inventory

• Shopper behavior

• Margin

Popularity pushes best sellers up. Inventory keeps out-of-stock items from crowding the top. Shopper behavior, like clicks and add-to-carts on past searches, shows which results people choose for a given query. Margin boosts the products that earn you the most.

The first three all follow what shoppers want, while margin follows what the store wants, so it’s the one to watch. A margin boost can lift revenue per search, but if it pushes the products shoppers came for down the page, search conversion falls with it. Test margin boosts before you roll them out. Personalized ranking is a further step, covered in the AI search section below.

A monthly relevance review

Block one hour on your calendar each month. Pull your top 50 queries by volume and your top 20 zero-result queries. For each top query, run the search yourself and look at the first row of results. If you were a shopper who typed that query, would you click any of them? Fix what’s broken with synonyms, attribute fixes, or a merchandising rule with an end date. Then rerun the eight query type tests from the audit and compare against last month. The audit scorecard has a tab for this.

How to A/B test search

You can test search, but plan for it to take longer than a sitewide test.

Start with what’s testable. Ranking changes, autocomplete design, typo-tolerance settings, zero-results page layouts, results page layouts, and the default sort order can all be split-tested. Infrastructure changes, like switching search vendors, are better evaluated with a before-and-after comparison on a holdout.

Make sure to measure the right thing. Use search conversion rate and revenue per search as your primary metrics, measured only on sessions that include a search. Sitewide conversion rate dilutes the effect because most of your visitors never touch search, and a real improvement to search can disappear into the noise.

Then do the traffic math up front. A search test is measured on searchers only, and searchers may be about a quarter of your visitors, as in Constructor’s data. With a quarter of the traffic, a test needs roughly four times as long to collect the same number of sessions as a test across all visitors. Your baseline search conversion rate also changes how many sessions you need, so run a sample size calculator with your own search traffic and baseline rate before you launch. If the answer is three months, test a bigger change.

Four tests worth starting with:

• A redesigned zero-results page with suggestions and a help option, against your current page

• Autocomplete with product thumbnails, against text-only suggestions

• A ranking boost for in-stock items, against your current ranking

• Looser or tighter typo tolerance on your 100 most misspelled queries

While a test runs, freeze merchandising rules on the queries it covers. A boost added in week two changes the results for one group and wrecks the comparison.

Where AI search fits

Once the fundamentals are in place, AI search is the next layer. AI search tools use language models and behavioral data to interpret queries, so a search like “something warm for a ski trip” can return relevant products even if those words aren’t in any product title. They can also personalize ranking for each shopper. They work best on top of the fundamentals in this guide, because AI search still depends on clean product data, measurement, and someone reviewing results.

When search can’t answer, let shoppers ask

AI search can read a messy query better, but some of the most valuable searches on your site are questions. “Will this fit a 15-inch laptop?” “Is this safe for sensitive skin?” “Can I return opened items?” “Where’s my order?” A search bar can match keywords to products and pages. It can’t weigh a product against a shopper’s situation, look up their order, or apply your return policy to their purchase.

That’s where a conversation helps. Gladly is customer experience AI that answers product questions on your site at the moment a shopper is deciding, covering fit, value, and what pairs well, and it guides them to the right products. It recognizes returning shoppers. It can also take action, like looking up an order, processing a refund, or canceling an order, within the guardrails and policies you set, across chat, SMS, and voice. When a question needs a person, Gladly hands the conversation to your team with full customer context, so the shopper doesn’t have to repeat themselves.

Crate & Barrel sees this with its own customers. Melissa Fye, Innovation and Improvement Manager, says, “Our customers often have detailed questions about customization and design. Gladly takes care of the simple requests instantly and ensures that when team members step in, they already have the full picture—streamlining even the most complex purchases.”

You don’t have to choose between search and conversation. Fix search for the shoppers who know what they want, and give everyone else a way to ask. Our guides to AI shopping assistants and conversational AI for ecommerce go deeper, and if search is only one of the leaks you’re chasing, start with how to reduce cart abandonment.