Applied filters

The controls that set a filter and the summary of what is set are two different things, and the second one is what this term names. Faceted search gives the reader a rail of checkboxes; applied filters gives them a sentence made of tokens sitting directly above the results, saying “this is why you are seeing 3 of 240”. The row earns its place because the rail is so often not on screen: collapsed behind a Filters button on a narrow viewport, scrolled past on a long one, or in a drawer that closed when the reader applied. Without the row, the only evidence that a filter is on is a result count that looks wrong.

Every token removes exactly one constraint, and the removal has to land back on the control that set it, so the checkbox unticks in the same beat. The remove affordance inside a chip is small, so it needs a target larger than the glyph drawn in it, and the accessible name has to say what pressing it does (“Remove filter: Remote”) rather than just “Close”. Clear all is a second, separate control, because clearing six filters one chip at a time is six round trips through a re-rendering result list. Keep it out of the reading order of the tokens themselves, usually at the end or set apart as a link, so it cannot be hit while aiming at the last chip.

Two things make the row feel unstable if they are not designed for. The first is order: tokens should sit in a fixed order, by facet group, so a chip does not move under the pointer when a neighbour is removed, and the row should reserve its height so the results do not slide up when it empties. The second is scope: a chip should only exist for a constraint the reader chose. Putting the default sort, the current tab, or an implicit “in stock only” in the row invites people to remove things the interface will immediately put back, and a chip that reappears after being dismissed reads as broken.

The row is also where the result count belongs, and where a change of results should be announced. Removing a filter changes a list somewhere else on the page, which nothing tells a screen reader about unless the count lives in a polite live region. Note that the word chip is doing double duty here: a filter chip reports and removes a constraint, while the chips inside a tag input are values being composed into a field. They look alike and behave nothing alike, so the tokens above a result list are worth calling applied filters rather than just chips.