Journal

How to design site search that helps people find things

A practical guide to designing search: the field, suggestions, results, filters and the no-results page, built around the words people actually type.

How to design site search that helps people find things

Search is the feature people reach for when your navigation has already failed them. They arrive with a specific need, a few half-remembered words, and very little patience. If the results page shows nothing useful in the first few seconds, most of them leave rather than try again.

The good news is that search is mostly a design problem, not just an engineering one. The input, the suggestions, the results layout, the filters and the "no results" state all shape whether people find what they came for. Here is how we approach site and in-app search so it actually helps.

Start with what people search for

Before you draw a single search bar, find out what people type. If the product already has search, export the query logs and sort them by frequency. If it doesn't, look at support tickets, sales calls and the words people use in usability sessions.

You will usually see three kinds of queries:

  • Known-item searches: people who know exactly what they want, like a product name, an invoice number or a team member.
  • Exploratory searches: broad terms like "pricing" or "templates" where people want to browse a set.
  • Problem searches: questions or symptoms, like "can't export PDF" or "change billing email".

Each kind needs a slightly different results page. Known-item searches want one strong match at the top. Exploratory searches want filters and a scannable list. Problem searches want help content, not product records. Design for the mix you actually have.

Make the search field easy to find and easy to use

A magnifying-glass icon on its own is easy to miss, especially on content-heavy sites. If search is a primary way into your product, show a visible input field with a short placeholder that hints at what can be searched, such as "Search orders, customers or invoices".

A few details make a real difference:

  • Make the field wide enough for a typical query. A narrow box makes people shorten their searches.
  • Support a keyboard shortcut (/ or Cmd+K) in tools people use every day, and show the shortcut inside the field.
  • Keep the query in the field after searching, so people can edit it instead of retyping.
  • Add a clear button once there is text in the field.

Use suggestions to shorten the path

Autocomplete is where good search starts saving time. As people type, show a short list of suggestions, grouped by type when you have more than one: pages, products, people, help articles. Six to eight suggestions is plenty; a longer list turns into a second results page.

Highlight the part of each suggestion that matches what was typed, so people can see why it appeared. Let the arrow keys move through the list and Enter open the selected item. For products with returning users, show recent searches when the field is empty and let people remove them.

Good suggestions answer the question before the person finishes asking it. If someone can jump straight to the right page from the dropdown, they never need the results page at all.

Design results people can scan

A results page should help people decide quickly which item to open. Each result needs just enough information to make that choice, and no more.

  1. A clear title with the matching words highlighted.
  2. A type label when results mix content types, such as "Article", "Product" or "Customer".
  3. One line of context: a short excerpt around the matching text, a price, a status or a date.
  4. A thumbnail only when the visual helps people tell items apart, like products or templates.

Show the number of results and repeat the query at the top ("24 results for 'invoice template'"). This confirms the search worked and helps people judge whether to refine it. Keep the order predictable: relevance by default, with sorting options only where people need them, like price or date.

Add filters that match how people think

Filters help with exploratory searches, but only if they match the way people describe what they want. Use the categories that appear in your query logs, not the ones that mirror your database tables.

On desktop, a left-hand filter column works well for more than three filters. On mobile, put filters behind a single button that opens a sheet, and show the active filters as removable chips above the results. Always show how many results each filter option will return, and hide or disable options that would return none.

Apply filters instantly when the result set is small and fast to load. For heavier queries, an explicit "Show results" button prevents the page from jumping around while people are still choosing.

Treat "no results" as a design state

The empty results page is where most search experiences let people down. A blank page with "No results found" gives people nothing to do except leave.

Instead, help them recover:

  • Check for typos and offer a corrected query: "Did you mean onboarding?"
  • Suggest removing filters if they are the reason nothing matched.
  • Show popular items, categories or help articles as a starting point.
  • Offer a way to contact support or request the missing item.

Also log every zero-result query. That list is one of the most useful pieces of product research you can get for free: it tells you what people expect to find and can't.

Handle synonyms, typos and partial matches

People rarely use your internal vocabulary. One person searches "bill", another "invoice", another "receipt". Work with your developers to add a synonym list based on real queries, and make sure the search tolerates small typos and partial words.

As a designer, your job here is to collect the language. Keep a shared document of terms people use, mapped to the terms your product uses. Review it every month alongside the zero-result log. Small changes to this list often improve search more than any visual redesign.

Test search with real tasks

Search is hard to judge from static mockups because the experience depends on real content. Prototype with real data where you can, even if it is a spreadsheet of a few hundred items.

In usability tests, give people tasks rather than queries: "Find the invoice you sent to a client last March" instead of "Search for invoice". Watch what they type first, whether they use suggestions, how they read the results and what they do when the first attempt fails. After launch, track a few simple numbers: the share of searches that end in a click, the share that return no results, and how often people search again straight away.

Conclusion

Good search is quiet. People type a few words, see the right thing near the top and move on. Getting there means designing every state, from the empty field and the suggestions to the results, filters and the page with nothing on it, around the words people actually use. Start with your query logs, fix the zero-result searches, and keep testing with real tasks. Each round makes your product easier to find your way around.

Oybek Ruziev

Designer & Framer developer · AidanCreates

Keep reading

All posts