Writing card sort items people can actually sort

A card sort is only as good as the pile you hand people. Most studies that come back muddy were muddy before the first participant arrived.

Five rules, and the reason behind each.

1. One altitude

The commonest mistake, and the most damaging. A deck that mixes "Delivery", "Next-day delivery to a locker" and "Shipping and returns policy" is asking people to group a category, a feature and a document.

They will do it, because people are helpful. What comes back is a clustering of your list's shape rather than of its meaning, and the biggest split in your dendrogram will be between the broad cards and the specific ones.

Before you run anything, read your list and ask whether every card is the same kind of thing. If not, split it into two studies or cut one altitude out.

2. Their words, not yours

"Omnichannel fulfilment options" is a real phrase somebody wrote for a real deck. Nobody outside the company groups that card, they park it.

Internal language does not just confuse, it distorts. Cards nobody understands end up in a miscellaneous pile together, and that pile then appears in your analysis as a coherent theme. It is not a theme, it is a graveyard.

Source your wording from support tickets, reviews, search logs and sales calls. If a phrase has never been typed by a customer, do not put it on a card.

3. No near-duplicates

"Free returns" and "Returns at no cost" will be grouped together by everybody, and they will produce a beautifully strong cluster that tells you nothing except that you wrote the same card twice.

Worse, duplicates inflate a theme. Three near-identical cards make that idea look three times as prominent as it is.

Read the list and merge anything two people would call the same idea.

4. Short enough to hold

A card has to be readable at a glance and holdable in memory while somebody considers where it goes. Long cards get scanned rather than read, and a scanned card gets sorted on its first three words.

Aim for under about eight words. If a card needs a sentence to make sense, the idea is probably at the wrong altitude, or it is two ideas.

5. Enough, but not too many

Twenty is thin. You will get clusters but the structure will be coarse. Sixty is a sensible ceiling: readable, and it keeps the analysis legible.

Past sixty, two things go wrong at once. Participants tire, and their sorting gets less considered toward the end of the deck. And many tools stop rendering the analysis usefully.

If your genuine list is longer, cut it rather than splitting it across studies, because two sorts of different decks cannot be compared.

Common traps

Brand names on cards. People group by brand rather than by the thing, and you learn about brand awareness instead of structure.

Negatives. "Not having to phone anyone" is harder to place than "Sort it out online". Write the positive.

Cards that are answers to different questions. A deck mixing needs ("shoes that last"), features ("leather upper") and channels ("buy in store") will split along those three lines, and you will have learned that you wrote three lists.

Fifty cards and no descriptions. If a card could mean two things, some participants sort one meaning and some the other, and the disagreement in your results is your own ambiguity coming back.

A quick check before you launch

Read your deck to somebody who does not work on the project and ask them to say what each card means. Any card they hesitate on is a card that will produce noise. That five-minute test saves more studies than any amount of analysis technique.

After the sort

If you ran an open sort, standardise the group names before reading anything. Fourteen people will produce "Comfort", "comfort" and "Fit and comfort" for the same idea, and until those are mapped together your analysis is counting one theme as three.