Card Sorting and Generative Methods

Card sorting is a foundational research method for understanding how people mentally organize information. You create a set of cards, usually 30 to 100, each representing a content object, then ask participants to work with them to reveal their mental model of how that content should be structured and labeled. Sessions typically involve 6 to 12 participants, individually or in small groups of 3 to 5, and participants should think aloud so you understand the reasoning behind their choices, not just the outcome.

Open card sorting is the most common generative method: participants group the cards however feels natural to them, without predefined categories, then label the groups they've created themselves. This reveals both how people naturally group content and what terminology they'd use for the categories. Modified-Delphi card sorting, devised by Celeste Lyn Paul, takes a different approach: rather than each participant sorting independently, a seed participant creates an initial sort, and each subsequent participant refines that same evolving sort in turn, until the group reaches consensus. Because it produces a single structural model instead of many individual ones to reconcile: it's often more reliable and simpler to analyze, but it takes longer per session, since participants work sequentially rather than in parallel.

Open card sorting shows you many individual mental models you then have to reconcile. Modified-Delphi trades that reconciliation work for a slower, more deliberate path to one shared model.

Exercise

The scenario: A UX team is redesigning a large corporate intranet. Two team members are arguing about which card-sorting method to use.

The project has a tight six-week timeline. The team already has a rough sense of the intranet's content domain from earlier stakeholder interviews, but no idea how employees would naturally group hundreds of policies, forms, and resources. One team member wants to run open card sorts with 10 individual employees, in parallel, this week. The other wants to run a single modified-Delphi session with a small group of employees representing different departments.

Given the six-week timeline and the need to understand how a broad, unfamiliar employee base thinks about this content, which method should the team choose?
Try It With Your Data: Card Sort Analysis

Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.

What this prompt is made of (RICE breakdown)
R

Role

Casts the AI as: You're a UX researcher specializing in card sorting analysis and information architecture.

I

Instructions

Names 5 specific outputs to produce, so the response comes back structured rather than a general summary.

C

Context

Domain, Current IA and Research question tell the AI what your specific situation is, not a generic one.

E

Expected format

Structured analysis with actionable recommendations

Choose the method you ran.
Whole number, e.g. 25
Whole number, e.g. 20
Paste raw notes or a summary: e.g. "20/25 grouped Card A + Card B together."
Short phrase, e.g. "Payment API documentation"
One line, or "none"
One sentence, what decision this should inform
Assembled prompt

        
        
      
Try It With Your Data: Category Label Synthesis

Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.

What this prompt is made of (RICE breakdown)
R

Role

Casts the AI as: You're a UX researcher synthesizing category labels from user research data.

I

Instructions

Names 5 specific outputs to produce, so the response comes back structured rather than a general summary.

C

Context

Domain, Audience, Existing terminology and Brand voice tell the AI what your specific situation is, not a generic one.

E

Expected format

Label recommendations with supporting data

One cluster per line, e.g. "Cluster 1: Settings, Preferences, Config (8 participants)"
Short phrase, e.g. "Developer documentation"
e.g. "New developers"
Current labels in use, or "none"
Pick the tone that fits your site.
Assembled prompt