What 1,037 card sorts reveal: how users actually organize information
We analyzed every published card sort study on ValidateThat — 1,037 studies, 4,044 participant responses, 7,476 user-created category labels. Here are the practical patterns that emerged: study sizes, sort type mix, category counts, and the canonical vocabulary users reach for when they organize content from scratch.
Last updated: June 4, 2026 · 1037 studies · 4,044 responses
▶Methodology — what's in the dataset
ValidateThat is a self-serve card sorting tool used by UX researchers, product managers, and student researchers. This report aggregates 1037 published studies — open, closed, and hybrid sorts — and 4,044 participant responses across them.
All numbers are aggregate-only. No individual study titles, user identities, or response-level content is retained in the published dataset — only distributions and frequency counts. Studies are excluded if they're still in draft status. Auto-generated category names ("Category 1," "Group 2," etc.) and obvious test labels are filtered before computing the top user-created labels.
The extraction script (open source) is read-only and can be re-run by anyone with read access to the platform data; this report will be refreshed periodically as the dataset grows.
Open, closed, and hybrid sorts are roughly evenly used
Classic UX literature frames card sorting as a binary choice between open (users create categories) and closed (users sort into yours). Platform data shows a three-way split: open at 41%, closed at 29%, hybrid at 30%. Hybrid sorts — "use my categories but also create new ones if mine don't fit" — are now roughly tied with closed sorts in real-world usage.
Users create their own categories
Users sort into pre-defined categories
Users sort into yours + can add their own
Why this matters: if you're picking a method for your next study, the "open vs closed" binary underestimates how often hybrid is the right call — when you have a candidate IA but want users to surface what's missing. See the differences.
Most studies use 11–30 cards. The textbook number is wrong.
The classic UX literature recommends 30–60 cards per study. Real-world usage clusters lower — the median study runs in the 11–30 range. Studies above 50 cards are rare and their response rates suffer.
Practical recommendation: if you're planning a card sort, start in the 15–25 range. The data suggests you'll get more usable response data than at 40+ cards, and your participants will finish.
The same 10 navigation labels appear in 5%+ of card sorts
Across 1,037 studies and 7,476 user-created category labels, a tight canonical vocabulary emerged. Home, About, Contact, Help, Support, Account, Settings, Services, Resources, Events — each appears in 2–5% of all card sorts across wildly different content domains (restaurant menus, real estate apps, edtech platforms, hospital websites). Below the top 10, labels split into domain-specific vocabulary that's idiosyncratic to what was being sorted.
The data is also multilingual: Korean "마이페이지" (My Page) is the single most-used label in the dataset (6.1% of all studies), and Spanish/Portuguese labels like "perfil," "especialidades," and "viajes" appear repeatedly across non-English studies. Users reach for their language's equivalent of the same concepts, not a translated version of English IA conventions.
Why this matters: if you're a UX team trying to predict how users will group your content before testing, the safest assumption is that they won't reach for the labels in your competitor research. Run an actual open card sort with your domain's vocabulary.
The 5±2 rule holds: most categories cluster at 4–8 in real studies
In closed and hybrid sorts where researchers pre-define categories, the count of categories per study clusters in the 4–8 range — close to Miller's classic 5±2 working-memory bound. Studies with fewer than 3 categories or more than 12 are rare.
Five takeaways for your next card sort
Start with hybrid, not open
1 in 4 platform studies are hybrid sorts. They give you the discovery benefit of open sorts (users can add categories) plus the comparability benefit of closed sorts (your categories appear in every response). If you're undecided, start hybrid.
Aim for 15–25 cards, not 40
The platform median is 11–30 cards. Studies above 50 see worse response rates. Pick the 15–25 sweet spot unless you specifically need wider coverage.
Cap your categories around 8
Real category counts cluster in the 4–8 range. Higher than that and users struggle to keep them in working memory, which biases the sort.
Don't import competitor labels
Researcher-created category names are wildly idiosyncratic across the dataset. The 'top labels' that emerge in your study will be specific to your content domain, not borrowed from competitors.
Run a debrief alongside the sort
Quantitative card sort data shows what users grouped together; it doesn't say why. Pair every card sort with a 10-minute debrief on a subset of participants. See our 25 qualitative research questions guide for prompts.
Citation: Byrne, R. What 1037+ Card Sorts Reveal: A Patterns Report. ValidateThat, 2026. validatethat.io/research/card-sort-patterns
Want to cite or republish? These numbers are free to use — please link back. Methodology and extraction script are open source.
Run your own card sort, free
ValidateThat's free plan covers 3 card sorts (open, closed, or hybrid) with unlimited responses. Your study contributes anonymized aggregate data to future versions of this report.
Start a card sort — free