· · Matthew Ford · 7 min read
Search UX best practices: a practical guide

A search results page should answer three questions quickly: what did the system find, which result is worth opening, and how can I narrow the set?
The answers depend on the content and the task. A product catalogue, a document library and a staff directory need different metadata and filters. Start with those differences rather than copying another site's layout.
This guide covers the interface decisions. It does not treat relevance ranking, indexing or query understanding as visual-design problems.
Show the query and the result set
Keep the submitted query in the search field and repeat it near the results. A person who has followed a link, returned with the Back button or opened several tabs should not have to remember what they entered.
Show the number of results where it helps set expectations. If the number changes as someone filters, update it at the same time as the list. When results span different kinds of content, label the content type or group the results only where those distinctions help people choose.
Do not silently replace a query with a corrected or broader one. If the system searches for something different, show both the original query and the query used. Give people a way to search for their original wording.
Make each result easy to assess
Give each result a specific title that links to the destination. Add a short description or snippet that helps someone judge relevance without opening every item. Highlighting matching terms can help, but it should not turn the snippet into a row of disconnected fragments.
Choose metadata according to the decision someone needs to make:
- a publication or review date for time-sensitive material;
- a file type and size for a downloadable document;
- a price, availability or delivery constraint for a product;
- a location, role or team for a person or service;
- a content type when the same result set includes pages, documents and media.
Use an image when it helps identify the result, not as decoration that pushes useful metadata down the page. Keep titles, metadata and actions in a consistent order so people can scan the list.
Save, compare and similar actions can be useful on result cards. Make their selected state clear and do not reveal essential actions only on hover. An expandable result can work for a small preview, but it should not become a second, harder-to-use version of the destination page.
Make sorting and filtering predictable
Use a sort order that matches the task. Relevance is a reasonable default for a text search, while a catalogue may also need price or availability and a document library may need date. Name the active option in plain language, such as "Newest first", rather than relying on an icon.
Filters narrow the current result set. Sorting changes its order. Keep those controls separate, and preserve the query when either changes.
- Put frequently used filters where they are easy to find. Collapse secondary options instead of presenting every possible control.
- Show available values and counts when they help someone choose. Remove or explain options that cannot return results in the current set.
- Show applied filters close to the result count. Let people remove one filter or clear the whole set.
- Keep the query, sort and filter state in the URL where practical. Reloading, sharing a link and using Back should not unexpectedly reset the results.
- Decide whether a change updates results immediately or after an Apply action. For a long filter panel, an explicit Apply action avoids repeated updates while someone is still making choices.
A clear-empty action is not a substitute for clear recovery. If a combination has no matches, keep the selected filters visible so someone can change one choice rather than start again.
Use faceted search without hiding the results
Faceted search offers filters on separate attributes of the same result set. A document search might offer content type, department and year. The useful facets depend on the information people know before they search and the metadata your content actually contains.
Start with a small set that supports common decisions. Allow multiple values within a facet when that matches the task, and make the combination rule clear. For example, selecting two content types usually broadens that facet, while adding a year narrows the overall set. Do not make people infer this behaviour from changing numbers alone.
Update facet counts from the current result set and test how each selection affects the other facets. Avoid a long column of empty checkboxes. If a value is unavailable, either remove it or disable it with an explanation that works for keyboard and screen reader users.
On every screen size, keep the results and applied selections understandable. A polished facet panel is of little use if people cannot tell why an expected item has disappeared.
Design for small screens
A desktop sidebar rarely transfers directly to a phone. Keep the query, result count, sort control and applied filters visible, then offer a clearly labelled way to open the remaining filters.
A filter drawer or full-screen panel can provide enough room, but it needs an obvious Close control and an Apply or Show results action. Show how many filters are active. When someone closes the panel, return focus to the control that opened it and preserve any selections they made.
Do not make the results wait behind a filter panel on every visit. Let people scan the first results before deciding whether to refine them. Keep filter summaries and result actions within the viewport without horizontal scrolling, and test long facet names as well as short sample content.
Check the whole journey at narrow widths with touch and keyboard input. Opening a filter, changing it, applying it and returning to the same place in the results should feel like one continuous task.
Make search accessible
WCAG 2.2 requires functionality to be operable through a keyboard interface, apart from a limited exception for path-dependent input.[1] It also requires labels or instructions when content requires user input.[2]
Use native form controls where possible, with persistent labels and clear grouping for related filters. Placeholder text can show an example, but it should not be the only instruction. Make focus visible and test every action without a pointer, including removing a filter and closing a mobile panel.
When results update without a page load, a brief message such as "18 results returned" can be a status message. WCAG 2.2 calls for these messages to be programmatically determinable without a change of focus.[3] Expose the count through an appropriate status role or live-region properties. Do not announce the entire result list or move focus on every filter change.
Also check that selected states are not communicated by colour alone, that text and controls remain usable when zoomed, and that errors identify both the problem and how to fix it. Test with assistive technology; automated checks will not establish whether the journey makes sense.
Give empty states a useful next step
"No results found" should not be the end of the journey. Repeat the query, keep the search field available and show any filters that contributed to the empty set.
Offer recovery in this order:
- check for a likely spelling correction or close match;
- suggest removing or changing a filter;
- offer a broader category or a revised query;
- provide a small number of relevant alternative destinations.
If a correction is available, either offer it as a link or show its results with an explicit explanation. Never present corrected or popular results as though they matched the original query. If there is genuinely nothing relevant, say so and provide another route, such as browsing a category or contacting the team responsible for the content.
Review unsuccessful searches with the people who maintain the content and the search engine. Some will need interface changes. Others will reveal missing content, synonyms, metadata problems or indexing errors.
Test the behaviour, not just the layout
Use realistic tasks rather than asking whether someone likes the page. Include a known-item search, a broad query, a query with a typo, a filter combination with no matches and a search using language that differs from the content's preferred terminology.
For implementation checks, cover the transitions that visual reviews often miss:
- the submitted query remains visible and editable;
- sort and filter choices work together and survive reload and Back;
- removing one filter leaves the other choices intact;
- result and facet counts agree with the displayed items;
- loading, no-results and error states are distinguishable;
- keyboard focus remains sensible after every update;
- the journey works at narrow widths without clipped labels or hidden actions.
When a result is poor, record whether the problem was ranking, metadata, query handling or interface behaviour. That distinction gives design and engineering teams something they can act on.
For the engineering side, read our article about improving search with OpenSearch and Elasticsearch. If you need to review an existing search journey, our UX design service covers interaction design, responsive interfaces and usability testing.
Sources
- [1] W3C WAI. Understanding Success Criterion 2.1.1: Keyboard.
- [2] W3C WAI. Understanding Success Criterion 3.3.2: Labels or Instructions.
- [3] W3C WAI. Understanding Success Criterion 4.1.3: Status Messages.
Discuss your software, technical leadership or AI plans
Tell us what you need to build, improve or decide.
We can help with software development and ongoing support, fractional CTO leadership, and practical AI adoption. Speak to our senior team about your priorities, our approach and availability.
Choose 15 or 30 minutes. Booking requests require confirmation by email.


