Card Sort Plan

Use this when you need to understand how users organize, categorize, or label content -- typically to inform information architecture, navigation design, or taxonomy decisions. Produces a complete plan: sort type selection, card list, facilitation approach, analysis method, and IA recommendations framework. Card sorts answer "how do users think about this content?" -- for testing whether an existing IA works, use tree testing or usability testing instead.

Related skills: Feeds into /information-architecture for IA design decisions. Use /usability-test-plan to validate the resulting IA with real tasks. Pair with /research-readout for stakeholder presentation of findings. References method selection in practices/user-centered-design/research-ops-method-chooser.md.

Process

Step 1: Gather inputs

Ask the user to provide:

1. IA challenge -- what information architecture problem are you solving? (e.g., "Users can't find settings," "We're restructuring the help center," "New product area needs navigation.")
2. Content inventory -- what content, features, or items need to be organized? (Provide a list, or point to an existing sitemap/feature list.)
3. User segments -- who should participate? (Different roles may organize content differently -- that's valuable data.)
4. Current IA -- does a structure exist today? What's broken about it? (This determines open vs. closed sort.)
5. Decisions this informs -- what navigation, menu, or taxonomy will change based on results?
6. Constraints -- remote vs. in-person, moderated vs. unmoderated, budget, tools available (Optimal Workshop, UserZoom, Miro, physical cards).

Step 2: Choose sort type

## Card Sort Plan -- (Domain/product, date)

### Sort type selection

| Type | When to use | Tradeoffs |
|---|---|---|
| **Open sort** | No existing IA, or starting from scratch. Want to discover how users naturally group content. | Richer insights, harder to analyze, more time per participant. |
| **Closed sort** | Existing categories. Want to validate whether users can place content into your proposed structure. | Faster, easier analysis, but won't reveal alternative mental models. |
| **Hybrid sort** | Have proposed categories but want to see if users would add, merge, or rename them. | Best of both, moderate analysis effort. |

**Selected type:** (Open / Closed / Hybrid)
**Rationale:** (Why this type fits the research question)

Step 3: Design the card list

### Card list

**Total cards:** (Target 30-60. Under 20 is too few for meaningful patterns. Over 80 causes participant fatigue.)

| # | Card label | Description (shown to participant) | Source |
|---|---|---|---|
| 1 | (Item name -- use language users would use, not internal jargon) | (Optional clarifying text if the label is ambiguous) | (Feature, page, help article, etc.) |
| 2 | | | |
| 3 | | | |

Card list design rules:

• Use user language, not product language. "Change my password" not "Authentication settings." If unsure, test the labels.
• Keep labels concise. 2-5 words per card. Participants scan, they don't read essays.
• Add descriptions only when labels are ambiguous. Descriptions should clarify, not lead.
• Include edge cases. Add 3-5 items that don't fit neatly -- these reveal where your IA has gaps.
• Exclude obvious outliers. Don't include "Contact us" alongside 40 feature cards. Keep cards at the same conceptual level.
• Randomize card order. Never present cards in your proposed IA order -- it anchors participants.

For closed/hybrid sorts, also define the categories:

### Proposed categories (closed/hybrid only)

| # | Category name | Description |
|---|---|---|
| 1 | (Category label) | (What belongs here) |
| 2 | | |

**Rules:**
- (Can participants create new categories? -- hybrid only)
- (Can participants rename categories? -- hybrid only)
- (Can a card go into multiple categories? Typically no.)

Step 4: Plan facilitation

### Facilitation approach

| Parameter | Detail |
|---|---|
| Format | Moderated / Unmoderated |
| Setting | Remote / In-person |
| Tool | (Optimal Workshop, UserZoom, Miro, Figma, physical index cards) |
| Participants | (15-20 for open sorts, 20-30 for closed sorts -- need more for statistical reliability) |
| Session length | (Moderated: 30-45 min. Unmoderated: 15-25 min.) |
| Incentive | (Appropriate for session length) |

### Moderated session guide (if applicable)

**Introduction (3 min):**
- "We're trying to understand how you think about [domain]. There are no right or wrong answers."
- "I'm going to show you [N] cards. Each card has a [feature/page/topic] on it."
- (Open sort) "Group the cards in whatever way makes sense to you, then name each group."
- (Closed sort) "Place each card into the category where you'd expect to find it."
- (Hybrid) "Place cards into the categories provided. If a card doesn't fit, you can create a new category."

**During the sort (20-30 min):**
- Let the participant work silently for the first few minutes
- If they pause: "Tell me what you're thinking right now."
- If they struggle with a card: "Where would you go looking for this?"
- Note cards that take a long time to place -- these signal IA friction
- Note cards placed confidently -- these signal strong mental models

**After the sort (5-10 min):**
- "Walk me through your groups. Why did you put these together?"
- "Were any cards hard to place? Which ones?"
- "Are there any groups you'd want to split or merge?"
- (Closed sort) "Were any categories confusing? Would you rename them?"
- "If you were looking for [specific card], where would you go first?"

Step 5: Define analysis method

### Analysis plan

**For open sorts:**

1. **Standardize group labels** -- Participants will name groups differently. Cluster similar group names. (e.g., "Account stuff," "My settings," and "Profile" might all mean the same thing.)
2. **Build a similarity matrix** -- For each pair of cards, calculate how often they were grouped together. Higher similarity = stronger association.
3. **Generate a dendrogram** -- Hierarchical cluster analysis showing which cards naturally cluster. Cut the dendrogram at different levels to explore possible category structures.
4. **Identify outlier cards** -- Cards with low similarity to everything indicate items that don't fit the current mental model or need their own home.

**For closed sorts:**

1. **Calculate placement rates** -- For each card, what percentage of participants placed it in each category? 70%+ agreement = strong signal. Under 50% = the card or category needs work.
2. **Identify mismatches** -- Cards where the "correct" category isn't the most popular choice.
3. **Find confused pairs** -- Categories that participants frequently mix up suggest overlap or poor labeling.

**For hybrid sorts:**

1. **Analyze both dimensions** -- Closed placement rates for existing categories + open analysis for new categories participants created.
2. **Evaluate new categories** -- Did multiple participants create similar new categories? This signals a gap in the proposed IA.
3. **Category rename patterns** -- Did participants rename existing categories? What language did they use?

### Confidence assessment

| Signal | Interpretation |
|---|---|
| 70%+ participants group two cards together | Strong association -- keep them together in IA |
| 50-70% agreement | Moderate association -- consider grouping, but test further |
| Under 50% agreement | Weak or no association -- separate placement likely fine |
| Card placed in 3+ categories evenly | Card label is ambiguous, or the item belongs in multiple contexts |
| Category consistently underused | Category may be unnecessary or poorly named |

Step 6: Plan IA recommendations

### From sort results to IA recommendations

After analysis, produce:

1. **Proposed category structure** -- Based on cluster analysis (open) or validated placement (closed)
2. **Labeling recommendations** -- Use participant language, not internal jargon. Note where participant labels differ from current labels.
3. **Problem areas** -- Items with low agreement, categories with overlap, content that participants couldn't place
4. **Next steps:**
   - Tree test the proposed IA to validate findability (before building)
   - A/B test navigation labels if two strong candidates emerged
   - Use `/usability-test-plan` to test the full IA with realistic tasks

Step 7: Review and validate

Ask the user:

• Are the card labels written in user language? (Test 5 labels with a non-expert.)
• Is the card count manageable? (30-60 is the sweet spot. Over 60 causes fatigue.)
• For closed sorts: are the proposed categories genuinely debatable? (If placement is obvious, the sort won't teach you anything.)
• Do you have enough participants for reliable analysis? (15+ for open, 20+ for closed.)
• What will you do with the results? (Card sorts inform IA, but the IA still needs validation through tree testing or usability testing.)

Card sort design rules

• Card sorts reveal mental models, not usability. They show how users think about content, not whether they can find it. Validate with tree testing or usability testing.
• Open sorts discover. Closed sorts validate. If you don't know how to organize the content, run an open sort first. If you have a proposal, validate with closed.
• Card labels make or break the study. If participants don't understand a card, their placement is noise. Pilot with 3-5 people before launching.
• More participants = more reliable patterns. Open sorts need 15-20 minimum. Closed sorts need 20-30. Unmoderated remote sorts make larger samples feasible.
• The dendrogram is a starting point, not an answer. It shows what clusters naturally -- you still need judgment about what makes a good IA.
• Don't card sort what you can analytics. If you have click data showing where users go, use that. Card sorts are for when you don't have behavioral data or are designing something new.

Output location

Present the card sort plan as formatted text in the conversation. The user adapts it to their tool of choice and shares with the research team.