Someone using your search bar has already told you what they want. Most stores treat search as plumbing and never look at what gets typed into it, or at what returns nothing.
Every other signal your store collects is inferred. Analytics infers intent from a click path. Heatmaps infer attention from a cursor. Attribution infers a decision from a sequence of touchpoints and a model with opinions.
The search bar is different. Somebody typed words into a box because they wanted something specific and couldn't find it by browsing.
That's not inference. That's a customer telling you, in their own words, what they came for.
Most stores never read it.
The state of it is worse than you'd guess
Baymard runs a rolling benchmark on ecommerce search experience. The headline from the 2026 round: 56% of sites have mediocre or worse search UX.
The component failures are more specific and more fixable than the headline suggests. Around 69% of sites don't return useful autocomplete suggestions when a term is slightly misspelled. Only 19% get autocomplete genuinely right, even though 80% have some version of it. 37% fail to keep the query in the box after the results load, so refining a search means retyping it. On mobile, 21% don't put an obvious submit button next to the field.
None of that is exotic. It's the search equivalent of a door that sticks.
Zero-result searches are a free list
Here's the part worth acting on this week.
Somewhere in your admin or your search app is a report of queries that returned nothing. Most stores have never opened it. It's the highest-signal document in the business and it usually splits into four piles.
Things you sell, described differently. A customer searched "joggers" and you call them "sweatpants." A customer searched "vegan" and your product copy says "plant-based." You have the product. The search doesn't know your synonym.
Things you don't sell but could. Repeated searches for a product you've never stocked is demand data most brands would pay a research firm for. It arrives free and nobody reads it.
Things that aren't products at all. "Returns", "shipping", "size guide", "track my order." People use the search bar as navigation, and if it only searches your product catalog, every one of those queries fails. Baymard has a whole study on this. The fix is to index your policy pages.
Typos and noise. Which is exactly what a decent misspelling tolerance is for.
Three of those four piles are money, and none of them require a new tool to find.
What good search actually does
Strip it back and there are six behaviors that separate search that works from search that technically functions.
It tolerates misspellings. "Adiddas" finds Adidas. This is table stakes and two thirds of sites fail it.
It knows your synonyms. Both the industry ones and the ones specific to your catalog. This list should be maintained by whoever writes your product copy, because they're the reason the gap exists.
It keeps the query visible after submission. People refine searches. Making them retype is a small cruelty that costs conversions.
It searches more than products. Policies, FAQs, guides. The search bar is navigation whether you designed it that way or not.
It handles no-results like a human would. Not a dead end with "0 products found." Show the closest matches, show popular categories, keep the person on the site. Roughly half of sites don't do this.
It works with a thumb. Visible field, obvious submit, results that don't require pinching. Most of your traffic is mobile and most search testing happens on a laptop.
Merchandising the results
Search that returns the right products can still lose the sale, because relevance isn't the same as order.
A search for "running shoes" that returns 200 relevant results sorted by whatever the platform defaults to is a wall. The shopper who typed two words and got 200 answers is now doing work you could have done for them.
Which products appear first is a decision. Most stores never make it, and the default is usually creation date or a relevance score that knows nothing about margin, stock or what actually converts. Sorting by what sells, promoting what you have plenty of, and demoting what you're about to run out of are all decisions somebody should be making on purpose.
Where the tools fit
If you need a proper search engine, Algolia is the serious option and Searchanise and Klevu are solid Shopify-native choices. We don't build search engines and don't intend to, so if your catalog is large enough that relevance ranking is a real engineering problem, those are the right places to run that.
What a search tool won't do is tell you which of the resulting changes actually made money. It'll give you a better search experience and a dashboard, and then the question of whether the new result ordering beat the old one sits with you.
That's the half we took. The Qosmic audit finds the opportunity, then we build the variant and adjudicate it against real revenue, both halves, so a change to how search results are ordered comes back as a revenue number rather than a feeling that it looks better.
You can run this yourself. Export the zero-result queries monthly, keep a synonym list, and have somebody own it. That's a spreadsheet and a recurring calendar invite, and plenty of sharp operators do exactly that.
Do this first
- Open the zero-result report. Sort by frequency.
- Fix the synonyms in the top twenty. That's usually an afternoon.
- Index your policy pages so "returns" stops failing.
- Search your own store on your phone for something you sell, with a typo.
- Decide what order results come back in, on purpose, and then test it.
Step four takes ninety seconds and it's where most people find the problem.
