Card sorting + tree testing case study

Designing a Connected Car App: Card Sort + Tree Test

By UXbeam, information architecture tools and services since 2021 · Updated October 7, 2026

Connected car apps typically organize automotive features by vehicle system, such as climate, security, access, charging and maintenance. An open card sort surfaced another organization based on proximity to the car.

The project team turned both into candidate navigation structures and tested them on eight tree testing tasks of matching intent, with separate participant groups. The proximity structure reached 70.1% task success versus 60.4% for the OEM-style structure. Three of the eight tasks were worded differently for each structure, and the overall gap comes from those three; on the five tasks worded identically, the two structures were close. The study breaks this down below.

Connected car apps: most owners have one, few reach for it

Almost every new car above a base trim ships with a companion app. The J.D. Power 2025 U.S. OEM ICE App Report finds that nearly 80% of owners use their vehicle's app, and 27% use it frequentlyJ.D. PowerJ.D. Power2025 U.S. OEM ICE App ReportDec 2025 · 2,100 ICE vehicle owners. Connectivity and speed take most of the blame. Navigation may contribute too. The audited apps organize features in very similar ways.

App use and satisfaction, non-EV brand apps
nearly 80% use the app · 27% use it frequently

J.D. Power 2025 U.S. OEM ICE App Report. Satisfaction on a 1,000-point scale, four ranked apps shown; 668 is the premium-segment average.

To ground the study, the project team used UXbeam to audit the feature sets of five OEM companion apps plus KeyConnect, a third-party digital-key app that spans dozens of makes. The audit produced 24 features covering access, climate, charging and fuel, security, maintenance, and location. Each of the five OEM apps presents those features grouped by vehicle system.

FordPass app icon FordPassFord HondaLink app icon HondaLinkHonda Tesla app icon Tesla myCadillac app icon myCadillacGM MyHyundai with Bluelink app icon MyHyundai with BluelinkHyundai KeyConnect app icon KeyConnectThird-party digital key

An open card sort found two organizing principles in the same features

The project team ran an open card sort: participants received the 24 feature cards and grouped them however made sense, naming their own categories. Across about 60 US adults, two broad organizing principles appeared. Most participants grouped features by vehicle system, the way every audited OEM app already does. About a quarter organized the same features around access and proximity: what the car does, and what the phone does on the car's behalf when you are somewhere else.

The broad sample is deliberate. Cars get used by far more people than their owners: drivers rent them, borrow one from family, take over a partner's EV for the weekend, or ride along and get asked to warm the cabin from the back seat. Someone who has never opened FordPass may still need to pop the trunk, check the charge, or find where the car is parked. Common vehicle interactions should stay understandable beyond the trained owner of one brand's app, so the card sort started from a broad population of US adults instead of existing connected-car app users.

Card sorting 24 connected-car feature cards: about 60 participants sort them, an OEM-style organizing principle (3 in 4 participants) and a proximity organizing principle (1 in 4) emerge, and each becomes a tree structure headed for a tree test of eight tasks with matching intent.
1 · 24 FEATURE CARDS 2 · ≈60 PARTICIPANTS SORT 3 · TWO PRINCIPLES 4 · TWO STRUCTURES 5 · TREE TEST Tire Pressure Monitoring Heated Steering Wheel Vehicle Status Reports Schedule Maintenance Car Notifications one deck · every participant OEM 3 in 4 participants CLIMATE ACCESS SECURITY CHARGING · FUEL MAINT LOCATION PROXIMITY 1 in 4 participants NEAR AWAY STATUS TREE A CLIMATESECURITYVEHICLE ACCESS CHARGING & FUELLOCATION · MAINT TREE B NEARBYAWAY VEHICLE STATUSSETTINGS TREE TEST BOTH SAME 8 TASKS
Climate Control
Sentry Mode
Fuel Level
Start Engine
Open Trunk
Lock/Unlock Vehicle
24 cards · ≈60 participants · 2 mental models · 2 tested structures

Participants created and named their own categories. Among the companion apps we audited, none used proximity or the user's relationship to the vehicle as the primary navigation structure.

The tree test: eight tasks, separate groups

We tested both structures with eight tasks of matching intent, using separate participant groups. Five tasks were worded identically for both structures; three (cabin temperature, security monitoring and range) were worded differently, which matters when reading the results below. Before participant testing, the final versions were cleaned up to remove duplicate destinations and obvious wording cues.

Each group had 18 participants in the clean set, drawn from the same US panel. Participants saw labels and hierarchy only, one task at a time. The eight tasks as worded for the proximity structure (B); where structure A's wording differed, it follows in brackets:

  1. You're in the parking lot and want to start your car remotely before you get to it.
  2. You want the cabin temperature set to 72°F before you get in. (A: You want the inside of your car at 72°F before you get in.)
  3. Your hands are full. You need to get into the boot without putting anything down.
  4. You're leaving your car overnight in an unfamiliar area. You want to turn on the security monitoring. (A: You're leaving your car overnight in an unfamiliar area. You want to make sure it's protected while you're gone.)
  5. You're at home and realize you may have left your car unlocked at the station.
  6. You have a meeting at 9am tomorrow 12 miles away. You want the car ready and warm when you leave at 8:30.
  7. You parked somewhere unfamiliar two hours ago and can't remember exactly where.
  8. You're about to leave on a long drive and want to check your fuel level. (A: You're about to leave on a long drive and want to make sure you have enough range to get there.)

Overall: 70.1% task success for proximity, 60.4% for the convention

Across 144 attempts per structure, the proximity tree reached 70.1% task success (101 of 144) versus 60.4% (87 of 144) for OEM-style. Direct success, reaching the destination without backtracking, was 43.8% (63 of 144) in both. Overall success was higher for the proximity structure, while Direct success was identical. Because three tasks were worded differently, the overall success gap cannot be attributed to the structure alone. The gap sits in the three tasks worded differently for each structure: 44 of 54 successes for proximity against 26 of 54 for OEM-style. On the five tasks worded identically, OEM-style had 61 of 90 and proximity 57 of 90, with 42 of 90 direct for each.

Task success by structure
A · OEM-style
60.4%
B · Proximity
70.1%

18 participants per structure · 8 tasks each. Share of attempts reaching a correct destination. Direct success was 43.8% in both structures.

The overall scores conceal large differences between individual tasks.

Where each structure wins and loses

Task success per structure · n=18 each
T1 · Start the car remotelyThe parking-lot context maps cleanly onto Nearby
A
72%
B
89%
T2 · Cabin to 72°FWorded differently for each structure; similar success, Direct favored A (56% vs 28%)
A
61%
B
67%
T3 · Open the boot, hands fullPhysical actions have an obvious home in both
A
78%
B
72%
T4 · Turn on security monitoring overnightThe widest gap, on a task worded differently for each structure: A read "protected while you're gone", B "turn on the security monitoring"
A
33%
B
83%
T5 · Unlock check from homeIdentical wording; the clearest like-for-like difference is directness: B 61% vs A 17%
A
72%
B
67%
T6 · Schedule tomorrow's departureWeak in both; scheduled departure still had no clear home
A
33%
B
28%
T7 · Find the parked carA ahead, direction only at this sample size; Location may be a cleaner label than Away
A
83%
B
61%
T8 · Check range before a long driveWorded differently for each structure; B's task said "fuel level" and its destination is labelled Fuel Level
A
50%
B
94%

Percent of attempts reaching a correct destination, per task. Readings summarize the paths participants took.

The two largest success gaps favored the proximity structure, but both occurred on tasks worded differently for each structure. Security monitoring from afar reached 83% success with the proximity structure versus 33% with the OEM-style structure, where A's task read "protected while you're gone" and B's "turn on the security monitoring". Checking range before a long drive reached 94% versus 50%, where B's task said "fuel level" and its destination is labelled Fuel Level. On identically worded tasks the clearest difference was directness rather than success: unlocking the car from home went direct for 11 of 18 with the proximity structure against 3 of 18. Finding a parked car under Location favored the OEM-style structure, 83% versus 61%, a difference this sample size cannot confirm. Neither structure gave participants a clear place for scheduled departure.

Card sorting opened a second design direction

The card sort revealed a radically different structure. Tree testing showed people could navigate it about as well as the convention on the five identically worded tasks, and more directly on one of them. Its larger wins on security monitoring and range checks came on tasks worded differently for each structure, so they cannot be credited to the structure alone. A pattern from roughly a quarter of card sort participants was enough to produce a testable information architecture; a like-for-like re-test of the three mismatched tasks is what would settle the comparison.

What a comparison like this can and cannot show, and how to read tree-test results in an assessment, is covered in the information architecture assessment guide.

Reading multiple mental models

Most participants share one mental modelRefine the convention
A coherent minority mental model appearsBuild it as a candidate IA
Two candidate structures existTree test both, same tasks

Two tested structures, plus an industry example

Mercedes-Benz has tested a third approach inside the vehicle. MBUX Zero Layer brings contextually relevant functions to the top level instead of requiring drivers to find them through fixed submenus. Zero Layer changes what the in-car screen shows; the proximity structure found with UXbeam changes how a companion app groups actions. Among the companion apps we audited, none uses proximity or the driver's relationship to the vehicle as the primary navigation structure.

Three approaches side by side: an OEM-style companion app organized by vehicle system (Climate, Security, Access, Fuel); the UXbeam proximity nav concept organized by Near, Away and Status; and Mercedes-Benz Zero Layer, an in-car interface organized by what is relevant right now.

Tested here: the two companion-app structures. Industry example: Mercedes-Benz MBUX Zero Layer, a case of a similar IA restructuring. Illustrative reconstructions, not product screenshots.

2021Mercedes-BenzMBUX Zero Layer launches with the EQS, bringing contextually relevant functions to the top level.
Feb 2023Mercedes-BenzZero Layer expands to C-Class and S-Class vehicles through an over-the-air update.
Jan 2025BMWBMW presents Panoramic iDrive and Operating System X at CES, with widgets that adapt to driver habits.

See Zero Layer in motion

A short demo showcases how Mercedes-Benz MBUX Zero Layer surfaces relevant actions directly in the current interface instead of requiring submenu navigation.

For the next design stage, see moving from a sitemap to low-fidelity wireframes.

Method note

  • Open card sort: run in UXbeam. The deck grew from 24 to 26 cards during fieldwork. Analysis snapshot at about 60 participants; recruitment later continued to 133.
  • Participants in both studies were US-based adults recruited through online panels, deliberately not restricted to current connected-car app owners. The research question is whether participants can find common vehicle interactions under a given structure, so general-population findability is the right measure.
  • A third, smaller mental model pointed toward a non-navigational concept and was explored separately; it is not part of this tree test comparison.
  • Tree tests: the OEM-style tree (A, version 2, with decoy destinations) fielded April 12 to 19, 2026; the proximity tree (B, version 3) April 7 to 11, 2026. Leading terms audited before launch. The exports do not record task order, so randomization is not claimed.
  • Clean analysis set: 18 participants per structure after excluding internal/test sessions, one outlier session, and one participant exposed to both trees.
  • Rates are observed percentages of completed attempts (144 per structure). Per-task gaps of a few points are noise. The large gaps on T4 and T8 fall on tasks worded differently for each structure; the T7 gap (15 of 18 against 11 of 18) is direction only at this sample size.
  • Five of the eight task instructions were identical for both structures. T2, T4 and T8 differed in wording (listed above), so per-task results on those three describe the combined structure and wording, not the structure alone.

Sources

Study data: UXbeam card sorting and tree testing sessions shown on this page.

Find the structures your users would propose

Run a card sort, take each pattern to a tree test, and compare structures on the same tasks.

Run a free card sort