Card sort —
three modes,
one engine.
Card sorting is how you find the categories real users think in — for a menu, a taxonomy, a product line, a help centre. Fieldbase runs all three modes in-app with the same drag-and-drop respondent UI: pick a mode, upload the cards, field it, read the results. Each mode exports a co-occurrence matrix, a best-fit cluster map, and a raw dataset you can take to R or Python if you want to cluster the data yourself.
Same cards, three experiences.
What the respondent sees in each mode — on the same card set.
The same twelve content cards (Apples, Milk, Bread, Frozen pizza, and so on) run through all three modes below. Only the shape of the sorting task changes — fixed buckets, free buckets, or a mix — and that shape drives whether you're discovering an IA, validating one, or doing both at once.
Pick by the question you're trying to answer.
The mode isn't a preference — it's determined by where you are in the design cycle.
Open sorts give you the vocabulary users actually use. Closed sorts tell you whether the labels you've chosen land. Hybrid lets you check both at once, with the trade-off that the fixed labels will bias where free-form additions appear.
- Open — "what categories exist?" New product area, blank-slate navigation, research into mental models. Respondents name the groups; you cluster the labels afterwards.
- Closed — "does my IA work?" You already have a candidate menu or taxonomy. Respondents can only place cards into your buckets — a stress test of how obvious the labels are.
- Hybrid — "validate, and catch what I missed." Fixed labels for the known structure, an "Add a group" affordance for what's genuinely new. Guards against the main failure of closed sorts: making users squeeze cards into categories that don't fit.
Three analysis views, one clean export.
Everything the respondent did, in the shape a cluster analysis expects.
Fieldbase computes the three views card-sort analysts actually use directly from the raw data, and ships the raw dataset too so you can rerun the clustering with your own parameters in R or Python if you want to.
card_sort_long one row per respondent-card-assignment rid · card · group_label · group_origin (fixed / open / added) · group_std_labelCard sort, in plain terms.
What is card sorting?
Card sorting is a research method where respondents group a set of cards — usually topic names, content items, or product features — into categories that make sense to them. It's used to design information architectures, taxonomies, menus and product-line structures that match how real users think.
What's the difference between open, closed and hybrid card sorting?
Open lets the respondent create and name their own groups — best when you don't yet know the right categories.
Closed gives respondents a fixed set of groups to sort into — best when validating an existing structure.
Hybrid gives the fixed groups but also lets respondents add their own — best for validating a structure while catching what you missed.
When should I use each card-sort mode?
Use open for discovery — IA from scratch, new taxonomy work, qualitative exploration of mental models. Use closed for validation — checking a proposed menu, tree-test alternative, A/B between two IAs. Use hybrid when you want to validate a candidate structure but still learn where users would add categories.
How many participants do I need for a card sort?
Spencer and Tullis' widely-cited guidance suggests 20–30 respondents reaches around 0.90 category correlation stability for most studies. Smaller studies (10–15) are fine for exploratory open sorts; validation studies often run 50+ if you want segment-level cuts.
What data do I get back from a card sort?
Fieldbase returns one row per respondent-card-assignment, plus a category co-occurrence matrix, a best-fit cluster map, and (for open sorts) a standardised list of category labels. The raw card_sort_long dataset exports as CSV, Excel or SPSS .sav alongside the main survey data for downstream clustering in R or Python.
Is card sorting the same as tree testing?
No. Card sorting asks respondents to build or populate a structure; tree testing asks them to find a target in an existing structure. Card sort designs the IA; tree test validates it under task pressure. Fieldbase runs card sort in-app today — tree testing is on the backlog.
Card sort in the same project as the rest of your research
Most card-sort tools are standalone — upload cards here,
export data there, then paste the clusters into a slide
deck. On Fieldbase the sort is a question type like any
other. It sits in the same survey, under the same quota, in
the same project as your screeners and follow-ups, and the
analysis views live alongside your topline and crosstabs.
The raw card_sort_long is one click away for
anyone who wants to cluster the data themselves.