Blog

Does good search actually sell more? I checked our data

Six Icelandic stores, one week: searchers order 3× as often, searches that find nothing get almost no clicks, and most famous search stats have no source.

By Dagur Leo, founder of Findlane · 21 October 2026

Every article about store search starts with the same numbers. "Shoppers who search are 2–3× more likely to buy." "68% of shoppers won't come back after a bad search." "Up to 30% of your visitors use the search box."

I build search for online stores, so believe me, I wanted those to be true. Then I went looking for where they come from, and most of them don't come from anywhere.

So I looked at our own data instead. Findlane runs search for a handful of online stores here in Iceland: a party shop, a homeware store, a couple of supplement stores and a few more. In six days in October, real shoppers on six of them made almost 30,000 visits and placed 707 orders. This is what that week shows, plus the outside research that actually holds up.

The short version

• People who search order about 3× as often as people who don't. Part of that is just that searchers came to buy. But even against visitors who were just as engaged, it's still about 2×.
• A search that finds nothing is a dead end. Those searches got 0.1 clicks per 100. Searches that found something got 41.
• Typo tolerance saves one search in eight. Without typo tolerance, synonyms and compound words, 44% of searches on these stores would have found nothing. With them, 23%.
• The outside research agrees on the direction, not the crazy multipliers. Better search is worth a few percent of revenue. That's a lot more than it sounds like, and I'll get to why.

The numbers everybody quotes

I traced the most repeated site search stats back to where they started. Here's what I found:

Stats pages from search companies are how a lot of these spread. Algolia's "40+ stats on e-commerce search" is a typical one. It says "Amazon's conversion rate shoots up 6x when visitors do a search (2% → 12%)" and that "Autocomplete can boost sales and conversions by as much as 24%". Both link to posts on Nacho Analytics' blog, and I couldn't find any data behind either. Walmart's "1.1% conversion rate goes to 2.9% with a search" on the same page has no source at all. I'm not picking on Algolia: almost every search company has a page like this, and it's easy to see how it happens. But if you're deciding where to spend money, you want numbers that come from somewhere.

What six Icelandic stores show

The data is from six stores running Findlane: searches from 9 to 19 October 2026, and visits, clicks and orders from 13 to 19 October. Bots and crawlers are left out, and so are our test and demo stores. How I counted has the details.

People who search are the ones who buy

Visits with a search placed 3.1× as many orders as the same store's visits without one, and brought in 2.8× the revenue per visit.

Now, people who search usually already know what they want, so of course they buy more. To see how much of the gap is just that, I compared only engaged visits, where the shopper looked at two or more products or clicked something in a product list. Search visits still ordered 1.9× as often.

One thing I didn't expect: searchers' baskets weren't bigger. On the stores with enough orders to compare, orders from search visits were the same size or a bit smaller. So search doesn't make people spend more per order. It gets more of them to an order at all.

Across all six stores, about a quarter of the orders had a product in them that the shopper clicked in search results earlier in the same visit. How much a store sells through search depends a lot on how visible the search box is: between 2% and 40% of visits used search, depending on the store.

A search with no results is a dead end

Searches that never found anything got 5 clicks in 5,877 searches. That's 0.1 clicks per 100 searches. Searches that always found something got 41.

Sounds obvious, right? But think about what happened there. Someone typed exactly what they wanted into your store. They literally told you. And the store answered "we don't have that", which a lot of the time isn't even true.

How often that happens

About one finished search in four found nothing: 25% over the whole period, 24% in the last week. That's a lot higher than the roughly 6% that search companies report for their customers (Prefixbox, Luigi's Box), and I think the reasons are worth knowing:

• Small catalogs. These stores have 400 to 6,500 products. There's just less to find.
• English in an Icelandic store. People search suit, cowgirl, elf and alien in stores whose products are named in Icelandic.
• Things the store doesn't sell. Brands they don't carry, products from somewhere else. That isn't search failing, that's a shopping list.

On the bigger stores, popular searches almost always work. The misses are in the long tail, the searches only a few people make, and together those are a lot of people.

How you count also matters a lot. With search as you type, 59–69% of search requests are sent while the shopper is still typing. Count every keystroke as a search and you'll think you have two and a half times more searches than you do. I only counted the search a shopper stopped at.

Typo tolerance saves one search in eight

To see what each feature actually does, I replayed real searches against the stores' catalogs with typo tolerance, synonyms and compound words switched off, one at a time and all together.

Almost all of the difference is typo tolerance:

• 12.4% of all searches found something only because a typo was forgiven. hárkollua found hárkolla (wig), sjóræningji found sjóræningi (pirate), and andlits maling found 238 andlitsmálning products (face paint). It's October in Iceland, so yes, it's mostly costumes.
• Another 4% found something because the word matched the start of a longer one, like jack finding jacket.
• Synonyms rescued 0.5% of searches on their own. Where they really help is adding results: they added more to about one search in ten. They're also how you catch the English searches in an Icelandic store.
• Compound words didn't rescue any search on their own in the sample. Good to have, but not the big one.

Icelandic is hard mode

Between 2% and 10% of searches, depending on the store, had a word typed without its Icelandic letters: buningar for búningar, risaedla for risaeðla. Phones, English keyboards, being in a hurry, all valid reasons. A search that thinks á, ð and þ are completely different letters from a, d and th misses every one of those.

As far as I could find, nobody has published data on store search in Icelandic, or in any small language like it. So here's some.

Where people click

When a search has 10 or fewer results, people pick from the top: 72% of clicks land in the top three. When it has more than 50, like búningar in October, they browse: 73% of clicks land below the tenth result.

So for broad searches, getting the first result right isn't the whole job. The order of the first 50 decides what people actually see.

What the outside research says (the parts that hold up)

Most stores' search isn't good. Baymard Institute is independent and doesn't sell search. In its 2026 benchmark of more than 170 sites and apps, 56% had mediocre or worse search, and phones did worst:

Shoppers say a failed search costs the sale. Google Cloud paid The Harris Poll to survey 1,008 US shoppers in late 2022. After a search that didn't work, "3 in 4 consumers reported that it resulted in a lost sale for the retail website. 51% went elsewhere to purchase the item, and 25% decided not to." It's a survey, so it's what people remember, not what they did. But it fits with 0.1 clicks per 100.

Searchers bring in way more than their share. Constructor, a search company, so add a pinch of salt, went through 609 million searches on 113 retail sites. Searchers were 24% of visitors and 44% of revenue, and converted 2.5× as often as people who didn't search. Our 3.1× is right in that neighbourhood.

When stores actually test better search, it wins, by a few percent. Not On The High Street, a UK marketplace, tested new search against its old one and got 2.2% more revenue per visitor across the whole site. Datart, a Czech electronics store, saw searchers convert 14% and 19% better on its two country sites. Etsy's better search raised conversion 2.63% across the site, and JD.com's raised it 1.29% overall and 10% on rare searches. Rare searches are exactly the long tail that small stores live in.

Speed counts too. In controlled tests, every 100 ms was worth 0.6% of revenue at Bing and 0.7% of revenue per session at Zalando, and on eBay's search page every 100 ms faster meant 0.5% more add-to-carts.

A few percent doesn't sound exciting. But in an analysis of 6,700 online store A/B tests, 90% moved revenue by less than 1.2%. Most things you can change on a store do close to nothing. Search is one of the few that reliably does something.

What I'd do if I ran a store

1. Read your searches with no results every week. It's the most honest feedback your store gets. Some are typos your search should catch, some are your shoppers' words for things you sell (add them as synonyms), and some are products you should maybe start selling.
2. Test your search on your phone. A typo, a missing accent, an English word, a plural. The typo post has a ten-search test.
3. Make the search box easy to find. On our stores, between 2% and 40% of visits used search, and the stores at the high end sell much more through it.
4. Check the broad searches, not just the exact ones. That's where people scroll, and where the order of results decides what they see.
5. Don't trust anyone's "search triples your sales", including mine. Compare orders from visits with search and without on your own store. Then compare only engaged visits, so you see what search itself adds.

Where Findlane fits

I'm obviously biased here, I built it. But this is exactly what Findlane is made for:

• Typo tolerance from the first search. One typo in words of four to six letters, two in longer ones, swapped letters count as one, and numbers always match exactly.
• Accents and special letters. buningar finds búningar, kobenhavn finds København and strasse finds Straße.
• Synonyms and compound words, so pirate can find sjóræningi and sjóræningjabúningur finds "sjóræningja búningur".
• Analytics that show what failed. Every search that found nothing, what people click and where the clicks land, plus a weekly email so you actually look at it.
• Easy to switch. If you're on Algolia, Findlane imports your settings, and here's how the prices compare.

The playground runs on a demo store, so you can try typos on it right now. It's free to start, no card needed.

How I counted

• The stores. Six Icelandic online stores running Findlane, with 400 to 6,500 products each. Four of them started sending real traffic on 13 October.
• Searches, 9–19 October 2026. A finished search is the last thing a shopper typed before they stopped, not every keystroke. The share that found nothing leaves out searches with filters.
• Visits, clicks and orders, 13–19 October 2026: 29,662 visits from 20,024 visitors, and 707 orders. A search visit has at least one click on a search result or on a search suggestion. Each store's search visits are compared with its own visits without search, then the stores are added up.
• The replay ran each store's 200 most popular searches and a random 200 of the rest against today's catalogs, with each feature on and off, and scaled the results up to all finished searches. The sixth store had too few searches to include.
• What it can't show. It's one week, six stores, one country, two weeks before Halloween. It shows what goes together, not what causes what. And search requests don't carry a visit ID, so a visit where every search found nothing counts as a visit without search. That's the next thing I want to measure: what people do right after a search that found nothing.