Card sorting before you build: test feature hierarchy

You have a list of features for your product. Maybe it's on a whiteboard, in a Notion doc, or living rent-free in your head. You know roughly what the product does. The question you probably haven't asked: do your users know where to find things?

Most founders ship a product with a navigation structure that mirrors how they think about it — by technical module, by development sprint, by the order they built things. Users don't think this way. They think in tasks, in goals, in the job they're trying to do. The mismatch creates "navigation debt" that's expensive to fix after launch.

Card sorting is the fastest way to prevent this. It takes 20 minutes to set up and 10 users to get a clear signal. Here's exactly how to do it.

What card sorting actually tells you

A card sort gives participants your feature list (as "cards") and asks them to group those features into categories that make sense to them. It answers:

Where do users expect to find each feature? Not where you'd put it — where they'd look for it.
Which features feel related to users? You might separate "Analytics" from "Reports." Users might group them together.
What language do users use for categories? You might call it "Settings." Users might call it "Account" or "Preferences."
Which features are orphans? Items that participants can't categorize consistently are features they don't understand — a signal worth knowing before you build them.

When to run a card sort (hint: now)

The ideal time is after you have a feature list and before you commit to a navigation structure. That's usually:

Before building your MVP's interface
Before a major redesign or feature reorganization
Before migrating to a new product tier structure (free vs pro vs enterprise)
Before adding a batch of new features to an existing product

If you're a founder with a feature list and a prototype in Figma, you're in the perfect window. If you've already shipped and your support inbox has "I can't find X" tickets, you're overdue but it's not too late.

The 20-minute setup

Here's how to go from nothing to a live card sort in 20 minutes:

Step 1: List your features (5 minutes)

Write every feature, page, or section your product will have as a short phrase. Don't group them. Just list them.

Rules:

2–4 words per card. "Usage Analytics" not "A dashboard showing usage analytics over time."
12–20 cards total. Fewer than 10 is too easy to sort. More than 25 causes fatigue.
One concept per card. "Billing" and "Invoices" are separate cards, even if they'll live on the same page.
Use user language, not internal jargon. "Activity Feed" not "Event Stream."

Example for a project management SaaS:

My Tasks
Team Calendar
Time Tracking
Project Settings
Notifications
File Storage
Client Portal
Invoicing
Reports Dashboard
Team Chat
Integrations
Billing
User Permissions
Templates
Knowledge Base

That's 15 cards. Perfect scope.

Step 2: Create the study (5 minutes)

Go to ValidateThat and create a free account
Create a new card sort study
Choose "Open" sort (let users create their own categories — this gives you the richest data)
Add your cards
Write a one-line welcome message: "Help me organize [Product Name]'s features. Group them however makes sense to you."

Step 3: Recruit 10 participants (10 minutes)

You don't need 50 people. Research shows card sort patterns stabilize at 15 participants, but with 10 you'll see the dominant groupings clearly enough to make a decision.

Fastest ways to get 10 responses:

DM 10 people in your target user profile individually. Personal asks convert at 50%+.
Post in one Slack/Discord community where your target users hang out.
Use Prolific for $35 total (~$3.50/response) if you don't have an audience yet.

Don't recruit your team. They know too much about your product's internal structure. You want outsiders who represent your users' mental models.

Step 4: Wait 24 hours

Responses typically come in within 24 hours for self-recruited studies, or within 2–4 hours for Prolific.

Reading the results

Once you have 10 responses, here's what to look at:

The similarity matrix

This shows which cards were grouped together most often. Look for:

Strong clusters (70%+ agreement) — these are natural groupings. Your navigation should respect them.
Weak associations (30–50%) — these items could go either way. Test placement with a tree test after.
Orphan cards (low similarity to everything) — these are features users don't understand or can't categorize. Either your label is confusing or the feature itself needs rethinking.

Category names (open sorts only)

If you ran an open sort, look at what participants named their groups. Common themes:

If 7 out of 10 people create a "Settings" group containing the same 4 cards, that's your settings page.
If people use 4 different names for the same group ("Tools," "Utilities," "Apps," "Add-ons"), you have a labeling challenge — test a few options.
If one card appears in 3+ different categories across participants, it might need to live in multiple places (or be renamed for clarity).

The "surprising" groupings

The most valuable findings are usually the non-obvious ones:

Users group "Time Tracking" with "Invoicing" (because time → billable hours → invoice). You had it under "Productivity."
Users group "Knowledge Base" with "Templates" (because both are reference material). You had them in different sections.
Users can't agree where "Integrations" belongs — some put it with Settings, others with Tools. This tells you it might need prominent top-level placement.

From card sort to product decisions

A card sort doesn't give you a complete navigation. It gives you constraints — groupings that your users' mental models expect. Turn them into decisions:

Lock in the clusters with 70%+ agreement. These are non-negotiable groupings.
Test ambiguous items with a follow-up tree test. Give users tasks ("Find where to track time") and see if your proposed structure works.
Name your categories using participants' language. If 8/10 people called it "My Workspace," don't call it "Dashboard."
Kill or rename orphan cards. If nobody can categorize "Workflows," the concept isn't clear to users yet.

What this prevents

Teams that skip IA validation and ship based on internal mental models typically hit these problems 3–6 months post-launch:

"Where is X?" support tickets — users can't find features that exist
Low adoption of new features — they're buried in unintuitive places
Navigation redesigns — expensive, disruptive, and avoidable
Feature confusion — users don't understand what a feature does because its placement gives no context

A $0–$35 card sort before you build prevents thousands of dollars of redesign work later. It takes less time than one team meeting about navigation. And the data is unambiguous — you don't need to interpret body language or read between the lines of user feedback.

Next steps

Create a free card sort → (takes 5 minutes)
See sample templates for card sort ideas by industry
Run a follow-up tree test to validate your structure once you've built it

Or if you'd rather skip the DIY and get a full validation report with 30 participants, methodology selection, and analysis done for you — see how it works.

Want the full breakdown on analyzing your results? Read How to analyze card sort results once your responses come in.