YouTube Trails
Designing for learners on a platform built for watchers
Summary
A concept exploration into intentional content discovery
It began as a structured academic exercise in Social Interaction Design, grounded in user research and IA testing. After graduating, I returned to it as a personal sandbox, using it to explore Cursor.ai and push the interaction concept further, generating a wider set of ideas.
Product Design
Interaction Design
Solo · Academic + Self-directed
Systems thinking
Interface Architecture
Flow Design
Why
YouTube is remarkably good at keeping you watching. For learners, hobbyists, and professionals, that same energy is an untapped opportunity to help people not just watch more, but build on what they've watched. A way to track a topic over time, revisit related content intentionally, and curate around a goal, not just a channel.
The Problem
How might we empower YouTube users to organise and revisit topic-specific content without disrupting the discovery experience that makes YouTube valuable?
Project story at a glance
This project has two honest layers:
a research-grounded concept (Phase 1) and
a self-directed building experiment (Phase 2).
3
Features directly driven by user data
85%
Tree test task success on core flows
10
Features built across both phases
0
Phase 2 features tested with users
Research Findings
Persona
Emily
25
Social Media Manager
Seeks variety and inspiration.
Uses YouTube casually for ideas and trends.
Avoids rigid structure.
David
42
Professor
Seeks focus and efficiency.
Uses YouTube for lectures and professional enrichment.
Feels overwhelmed by algorithmic clutter.
What I did
I ran a two-part research process: a behavioral survey (n=~15) to understand how users balance search and browsing, followed by tree testing via ProvenByUsers to validate whether new features could sit intuitively within YouTube's existing navigation.
Wireframes
Trail concept Integration,
Early wireframes
Dual-nav concept exploration,
Early wireframes
Users x Survey
Easier access to related content
86%
Hobby/interest based recommendations
71%
Organize hobbies / interests
57%
Felt overwhelmed sometimes in current feed
89%
Interested in Trails concept
71%
Unsure how they'd use it
14%
Users x Tree test Task Success Rate
Find Shorts
87.5%
Existing YouTube pattern
Use search bar
77.8%
Fast, avg 6.7s
Trail history
75%
Paths were inconsistent
Find a playlist
55.6%
44% went to search instead
Continue a trail
37.5%
Most searched or used history
Find community
37.5%
Avg 34s, explored unrelated
Customize / filters
28.6%
Avg 34s, explored unrelated
Temp watch later
22.2%
Most chaotic paths, 42s avg
0% Focused topic feed
100% went to Main Feed or Search Bar. The concept was completely invisible in the IA.
The most important finding of Phase 1.
What users could find
Standard YouTube actions (search, Shorts) were fast and direct. Trail history was findable once inside the system. Familiar patterns held.
What the IA buried
Every new concept, focused feed, filters, temp saves, community was invisible or deeply nested. The more novel the feature, the worse it performed.
Create & Curate Flow
Iteration lvl #2
Share & Collaborate Flow
Iteration lvl #2
Data - Decision - Feature Map
Testing based
Survey
71% want smarter recs + 57% want to organize their own content
Both, not either/or- the core insight
Design question
Two modes without switching apps?
User controls which experience they're in
Tested
Feed toggle - Regular Trail
iOS-style switch in header; nav changes automatically between modes
Wireframe tabs → toggle switch → dual nav system
Survey + user comment
Users want to control how much recommendation vs. topic content they see
"Volume of recommendation should be user-controlled" direct quote
Design question
Adjust algorithm without jargon?
Expanded in Cursor
Content focus slider
Single axis → dual-topic balance (React 34% ↔ 66% Node.js). Topic selector with main topic + meta tags.
Open: single axis was simpler. Does dual-topic + tag selector add too much complexity?
22% success · avg 42s
"Temporary watch later" completely failed
Chaotic paths, users looped through History, Playlists, Trails
Design question
Save without committing to a trail?
Lighter-weight than playlists
Tested
Parked videos
Inline Park + Trail buttons on every feed card. Parked tab: To Watch / Watched / time stats / per-video notes.
Most fully realized feature in the prototype
0% success
Focused topic feed completely invisible
100% went to Main Feed or Search Bar, concept didn't exist in users' mental model
Design question
Entry point or label problem?
Naming and discoverability are the real issue
Expanded in Cursor
Full onboarding flow
8 contextual cards triggered on first use. Welcome → Feed Toggle → Slider → Parked → Trail Path → Community → Notes.
Open: does onboarding fix discoverability or is the label / entry point still wrong? Not retested.
28.6% success · avg 28s
Customize / filters deeply buried
0 direct successes, users had to wander to find it
Design question
Controls at feed level, not settings
Surface inline, not nested
Tested
Inline focus ratio display
"React 34% ↔ 66% Node.js" visible in header at all times. Tap to expand Content Balance panel directly over feed.
Concept-driven, no direct data
HCI framework
Community curation over algorithmic discovery
Social interaction design theory, curators not just consumers
Design question
What if the best playlist was made by a person?
Human intent over algorithmic suggestion
Assumed
Collaborative playlists
3 complete screens: Your Collection, Create (email invite), Shared With You. "Curated by Mom / Professor Konow."
Open: creator consent, email inviting model, platform trust, all unknown.
The Afterthought · Lateral Thinking · Serendipity
How might we empower YouTube users to move from passive watchers to active curators, without requiring them to become creators?
Cursor building forced this
"Where does all of this live inside YouTube?"
Building made IA decisions that felt vague in Figma
Design question
New IA that doesn't break YouTube's existing structure?
Expanded in Cursor
Dual nav + full IA rebuild
5 primary destinations, 4 modal types, 4-tab Trail Details hub, conditional nav switching between feed modes.
Open: biggest unknown in the project. Original IA had 7 tree-tested tasks. This IA has never been tested.
After graduating, I returned to this project using Cursor.ai, treating a concept I already understood as a low-stakes environment to learn a new tool.
The prototype Git-link
The prototype is a working thinking artifact, focused on interaction logic, not visual polish. It contains more ideas than a real v1 would ship. Treat it as a thinking artifact, not a finished product.
The Build Sequence
Interface Architecture & Static Screens
Interactive artifact:
Click on Phases pill
Click on Rectangular-colored-nodes
How scope grew
Ph. 1
4 features · user tested · limited flows
v1 - 2
+ dual-topic slider · topic coloring
v3 -4
+ onboarding · tooltips · split nav · trail details hub
v5.5
+ contextual help · Trail Path · creator cards · collab playlists
Reflection
On building with AI tools
Cursor reduced the friction of turning ideas into interactions, which was both useful and dangerous. It's easy to keep building when building is fast. The harder discipline, deciding what not to build, didn't come from the tool. It had to come from me. And I didn't always apply it.
One key flip
The plan was to group parked videos under History & Playlists to reduce navigation. But parked Videos became its own prominent bottom tab. Neither decision has been user-tested. That's the most honest open question in the prototype.
On scope
The second phase grew too wide too fast. A new IA, onboarding flow, integration guidelines, and micro-interaction systems all running in parallel, none developed to the depth they deserved. The right approach: pick one flow, test it, then expand.
Next steps
1
Narrow to one flow (trail creation) and run structured usability testing with at least 5 participants
2
Run a second tree test on the expanded IA, the original had 7 tasks, the current IA is ~3× larger
3
Test whether onboarding fixes the 0% focused-feed discoverability failure, or if the label is still the problem
4
Explore curator side incentives. Do Trails help or hurt channel-level discovery?
Data synthesis glimpse
Observation-to-design-change synthesis table, 15 findings mapped across survey and tree test data.
