On a marketplace, buyers arrive with a more or less precise idea and leave as soon as they cannot find it. Marketplace search, with its filters, carries much of that first impression. Yet I often see it designed last, from the database fields, once the rest of the site is finished.
I designed and then built a marketplace with advanced search for business buyers. That kind of project forces you to separate what the team knows about the catalogue from what the buyer understands of it. Here is the order I work in today, alongside what usability research and Google's documentation say.
01Start from buyers' words, not the database
The temptation is strong: the database has a "material" column, a "supplier reference" column, a "status" column, so they become filters. The result looks like a back office. Buyers do not think in columns.
Kate Moran at Nielsen Norman Group sums up the questions to ask before drawing anything: which characteristics weigh most in the choice, which words users use to describe them, and whether our labels make sense to them or read like jargon. To answer them, she suggests interviewing users, talking to sales and support staff, and reading the site's search logs.
It is the same rule as in the brief of my content machine: the FAQ comes from questions people have actually asked, noted with their source, because an invented FAQ answers questions nobody asks. For a search feature, I do the same with the queries typed into the box, the questions support receives and the words people use in meetings. I sort them before opening Figma.
One last point before filters: check that people can see the search box. In its video How to Design a Good Search UI, NNgroup points out that low search usage in your analytics does not prove nobody wants to search. The box may simply be invisible. On desktop, the video recommends an open field at the top of every page rather than a lone magnifying glass, which forces a click before typing.
02Filters or facets: choosing the depth you need
The two words are often mixed up. Kathryn Whitenton, also at Nielsen Norman Group, draws a clear line. A filter excludes items that do not meet a criterion. Faceted navigation combines several filters that describe every aspect of the content, and also gives people a sense of what the catalogue holds.
Facets are more flexible, but they cost more: every listing, current and future, needs metadata for each facet, and the interface presents more options to understand. The same article advises checking that users really need them before investing.
On a marketplace, this comes down to a simple question: who fills in the listings? If sellers do, a facet is only reliable when the listing form enforces it with a closed list of values. A free-text "colour" field filled in by two hundred sellers gives you two hundred ways of writing "navy blue". I would rather have three well-populated facets than ten half-empty ones.
03Filter values people can trust
Once the categories are chosen, the work moves to the values. According to Nielsen Norman Group, they should suit the product type, be predictable, avoid jargon and be ordered by importance. The article mentions an animal shelter whose "Behavior" filter offered "Purple", "Orange" and "Green": an internal code that meant nothing to visitors.
Units and formats
On an integration between two tools, I saw a mix of centimetres and millimetres distort an entire rendering. Since then, I ask for the reference data to be explicit: one unit stored, converted for display. In search, the error is sneakier, because nothing breaks. A "width from 50 to 100" filter returns results, just not the right ones. The same source also recommends providing suitable currencies and units as soon as you serve an international audience.
Labels and order
For labels, I reread each value and ask whether a new buyer would understand it without help. If a technical term is unavoidable, a short explanation next to the filter beats a link to a help page.
For order, numerical values go from smallest to largest, brands alphabetically, and everything else by what matters to the buyer. General filters go at the top, specialist ones at the bottom, unless one specific criterion drives almost every purchase. In that case, it moves up.
04Applying filters straight away or with a button
This question comes up often on usability forums, and there is no single answer. Katie Sherwin at Nielsen Norman Group links it to user intent. People who are exploring learn the catalogue by filtering: they tick, look, tick something else. Interactive filtering suits them. People who arrive with several criteria in mind prefer to tick everything and then see the result, with an Apply button.
Speed matters too. If results arrive in under a second, interactive filtering is not much of a nuisance. If they are sometimes slow, especially on mobile, the same article recommends batch filtering. It also suggests a middle ground: wait one or two seconds without a new selection before refreshing, while showing straight away that the click has registered.
Two details I always check. The page must not jump back to the top while the buyer is still ticking boxes, or they will click the wrong one. And when the list updates, the results area dims as a block with a loading indicator, rather than every thumbnail flickering.
When the team cannot decide, I build two clickable prototypes at the same level of polish, one with a button and one without, and have a few buyers test them. Showing several directions at the same level unblocks a decision far faster than a debate. I have described the method in the article on building a clickable prototype before development.
05Zero results, counts and faceted URLs
Zero results is the moment when buyers decide whether to stay. The first line of defence is showing the number of results next to each value: according to the Nielsen Norman Group article on applying filters, this helps people avoid empty searches. I grey out values that would lead to zero rather than hide them, so the buyer can see what exists elsewhere.
If zero results happen anyway, the page must offer a way out: remove the last filter, see the whole category, or contact support.
There is one more subject product teams often forget: every filter combination creates a URL. Google's documentation on managing faceted navigation URLs explains that these parameter URLs form an almost endless space, which crawlers explore before realising the pages are useless, at the expense of new pages. Google offers two routes:
- If filtered pages do not need to be indexed, block their parameters in the robots.txt file, or put filters in a URL fragment (after the #), and let crawlers reach the listings and an unfiltered list page.
- If some combinations should be indexed, use the standard & separator, keep the filter order always the same, with no duplicates, and return a proper 404 status code when a combination returns nothing.
I settle this with whoever handles SEO before development, not after go-live.
06Where to start on your marketplace
If you are taking over an existing search feature, I would start with half a day of reading: the queries typed into the box, searches with no results, support questions. You will already know which words are missing and which filters nobody uses. If you are starting from scratch, a few interviews with buyers do the same job, and useful personas help keep their words in front of you.
Then, the checklist I follow before handing over:
- filters use buyers' words, not column names;
- every facet has a closed list of values on the seller side;
- a single unit is stored, and conversion happens at display;
- the choice between interactive filtering and an Apply button has been tested;
- every value shows its result count, and zero results offer a way out;
- the indexing rule for filtered URLs is written down.
I also check search as a visitor would, in a private window, with no account. A site can look fine to the logged-in team and still be broken for visitors because of caching. And looking at how two or three competitors arrange their filters, with a UX benchmark, quickly gives you ideas and counter-examples.
This method has a limit: it takes time at the start, when everyone wants to see screens. I think that time is recovered during development, but I cannot promise it for your catalogue. If you would like to talk it over, the UX and product engagements I run often start with this diagnosis, and the portfolio shows what I have built.