Ecommerce site search works best when the search bar is visible as an open field (not hidden behind an icon), suggests popular searches and products the moment a shopper taps it, tolerates typos and synonyms, and never ends on an empty results page.
In our database of A/B tests, making mobile search visible is one of the most consistent wins in ecommerce, while busier "guided" search panels have lost.
This guide covers what on site search is worth, the practices that held up in real tests, and how to run site search optimization as a testing program.
Every claim about lift links to a test you can check.
Blend Commerce
SplitBase
Visually
Clean CommitOn site search: why it matters more than its usage share suggests
On site search is the search box inside your store, as opposed to Google.
Only a small share of visitors use it, but those visitors usually know what they want, so they tend to convert at a higher rate than browsers.
The SNOCKS case from DRIP Agency shows the pattern.
Fewer than 0.1% of visitors used the barely visible search, even though searchers converted far better than non-searchers.
Making search prominent moved conversion from 17.29% to 17.52% in their A/B test.
The lift is small, which is a useful reminder that search improves a segment, not the whole store.
The quality problem is widespread. Baymard Institute's search query type research (updated April 2026) rates 56% of benchmarked sites "mediocre or worse" on search.
Failure rates vary by the kind of query shoppers type:
| Query type (Baymard) | Example | Share of sites with issues |
|---|---|---|
| Exact search | A product name or model number | 12% |
| Product type | "sofa" | 20% |
| Symptom | "stained shirt" | 37% |
| Feature | "waterproof jacket" | 39% |
| Use case | "camping gear" | 43% |
| Compatibility | "iPhone 16 case" | 44% |
| Abbreviation and symbol | "16in" vs "16 inch" | 54% |
| Non-product | "return policy" | 66% |
Baymard also found that 34% of test participants used the search field to look for non-product content such as shipping or returns.
If your search only indexes products, a third of searchers can hit a dead end before they ever see one.
Ecommerce search best practices backed by A/B tests
These practices come from tests in our filters and search element hub.
Each one names who ran the test and what it measured.
Many are vendor or agency case studies without published sample sizes, so treat single results as directional and the patterns across stores as the stronger evidence.
1. Show an open search field, especially on mobile
This is the best-supported search change in the database.
Replacing a mobile magnifying-glass icon with a visible search bar has won for many different stores:
- Aeromotive (automotive parts): +88% conversion rate, +109% revenue per user, in a Blend Commerce test.
- Titan Casket: +22% conversion rate and +76% average purchase revenue per user, mobile only, in a Blend Commerce test.
- Roots of Fight: +19.4% mobile conversion rate and +24.8% revenue per user over about three weeks, run by SplitBase.
- Wilson's Pet Food: +9.97% conversion rate from a visible mobile search bar instead of an icon (Blend Commerce).
- Panduro (craft supplies): +14% search usage and +2.67% purchases in a one-week VWO case study.
This cuts against older guidance. Nielsen Norman Group's 2014 magnifying-glass icon article advised against icon-only search on desktop but said it "makes sense on mobile devices" because space is tight.
For stores where shoppers arrive knowing the product, the tests above say the open field earns its space.
Not every result is as clean as its headline.
The Best Choice Products test reports 30% more conversions in its headline but only +0.1% revenue in the text, after one week with no significance given.
2. Put search where shoppers already look: the header and the menu
Oliver Kenyon has tested search inside navigation several times.
Adding a search bar inside the mega menu was worth +$81,050/month, and a mobile menu with search at the top (combined with a Clearance-first layout) was worth +$92,000/month.
The second test changed several things at once, so the search placement can't take all the credit.
On desktop, Blend Commerce saw +13.4% conversion rate from a full search bar in the desktop header.
3. Don't add search where it competes with buying
Location matters. Clean Commit placed a search bar above product images on product pages and conversion fell 4.2% across 40,182 visitors (not significant).
ConvertPolo went the other way for Attrangi, removing the product page search bar while making zoom more prominent, and saw +12% conversion.
That was a bundled change, but both point the same way: on a product page, the job is buying, not searching.
4. Fill the empty search panel before anyone types
When a shopper taps the search box, showing popular searches, top categories or products tends to help:
- Safety Gear Pro: +20.4% purchases when tapping the bar immediately showed top searches and categories (VWO, 9 days, 50/50 split).
- Dorado Fashion: 30% of mobile visitors clicked the search icon but only 1% searched. A focused search screen with popular searches lifted conversion 26.55% over 33 days.
- AnotherSole: +54.8% revenue from popular searches and highlighted collections, run by ConvertPolo, with no sample size given.
The effect is not guaranteed. Blend Commerce's Fragrance Hub test of the same idea finished flat: +1.92% revenue per visitor and -1.33% conversion rate, not significant.
5. Keep the suggestion panel simple
More guidance is not always better. Clean Commit made search more prominent and guided, with example queries, product previews, suggested collections and a "Customize Your Own" banner.
Conversion fell 12.0% over 22 days (not significant), even though average order value rose 11.5%.
A separate Clean Commit test of live search with category links showed +13.7% conversion across 53,622 visitors but was also not significant.
The same lesson shows up in results tiles.
Contorion stripped search tiles to image, title and price and clicks to product pages rose 2.3%, though no purchase data was reported.
Our best practices that failed page collects more cases where adding features backfired.
6. Tolerate typos, synonyms and natural language
Baymard's search research overview reports that 69% of sites lack relevant autocomplete suggestions for closely misspelled terms.
The tests suggest fixing the engine, not just the design, pays off:
- La Fée Maraboutée kept the search preview looking the same but swapped in an engine with typo handling and semantic matching: +11.7% revenue per user and +9.4% add-to-cart rate in the first two weeks. AB Tasty ran it on its own search product.
- The Bee's Knees moved from Boost AI search to a Google Vertex AI-powered engine built by Nimstrata and saw +20.6% conversion rate at 99% confidence over five weeks, with the agency 36 running the work.
- Andaaz Fashion replaced exact-name search with smart suggestions and predictive results: +28% search engagement and +18.39% purchases (ConvertPolo).
Vendor-run tests of their own product deserve extra caution.
The direction is consistent, but the sizes are probably optimistic.
7. Never show an empty results page
Zero-result searches are a fixable leak. Laura Geller Beauty showed relevant products when a search returned nothing and conversion rose 30.7% (Visually case study; scope and sample size not stated).
An Optimizely retail customer redirected zero-result searches to recommendations and saw +2.11% click-through and +2.74% revenue.
8. Write placeholder text that invites a search
Small copy changes can count. Blend Commerce rewrote Azuna's mobile search placeholder to invite product searches and saw +3% revenue per visitor and +1% conversion rate.
Conversion Fanatics gave Featsocks a clearer mobile search field with placeholder text and orders rose 18.73% at 98.5% significance over 14 days.
Blend Commerce
Optimizely
Conversion FanaticsSite search optimization: how to improve search with testing
Site search optimization means improving how easily shoppers find and use search, and how relevant the results are once they do.
Treat it as two separate problems, because they need different fixes and different metrics.
| Problem | Typical fixes | Primary metric | Evidence in our database |
|---|---|---|---|
| Shoppers don't find or use search | Open field on mobile, header placement, placeholder copy, prefilled suggestions | Search usage, then conversion or revenue per visitor | Strong and repeated across many stores |
| Search returns poor results | Typo tolerance, synonyms, semantic matching, zero-result fallbacks | Revenue per searcher, zero-result rate | Fewer tests, mostly vendor-run |
Start with your own search data
Before you test, pull three numbers: the share of sessions that use search, conversion rate for searchers vs non-searchers, and your most common zero-result queries.
GA4 tracks site search through its enhanced measurement setting; our GA4 for CRO guide covers the reports.
Dorado Fashion's 30% icon clicks vs 1% searches is exactly the kind of gap that tells you where to start.
Read your zero-result list closely. It shows the jargon shoppers use that your catalog doesn't, which is often the cheapest fix: add synonyms and redirects before buying a new engine.
Pick the right metric and segment
Measure revenue per visitor or orders as the primary metric, not search clicks.
GANNI's more visible search bar raised mobile search engagement 21.9% but conversion 7.5%, so engagement overstates the business result.
Jackson's new search bar showed +28% conversion for returning visitors and +5% for new ones, so segment by visitor type and device.
Size the test for the search segment
Because only a fraction of visitors search, a result-quality change dilutes when measured across all traffic.
Run results tests on searchers only, and expect to need several weeks of traffic.
Visibility tests can run on all traffic because they change who searches in the first place.
Test one change at a time when you can
Several of the biggest search wins bundled other changes, which makes them hard to repeat.
If you want to know whether the search bar did the work, isolate it.
For the broader process, see our mobile CRO guide and the full test database.
Want to see more evidence across the funnel?
Our CRO hub links every guide built on these tests.