Consumption Pattern Analysis

Overview

Analysis of learning content consumption across 9,996 tracked videos and 91 channels.

Content Categories

By Watch Time

By Purpose

Speed Patterns

Distribution

Speed    % of watches    Content type
────────────────────────────────────────
1x       10%             Complex, novel
1.5x     25%             Lectures, detailed
2x       40%             Standard tutorials
2.5x     20%             Familiar territory
3x+      5%              Skimming, known content

Speed Selection Factors

Go slower (1-1.5x):

• New/unfamiliar topic
• Complex explanations
• Non-native speaker presenter
• Following along (coding)

Go faster (2-3x):

• Familiar topic, seeking specific info
• Redundant explanation
• Slow speaker
• Review/refresher

Optimal Speed Research (Needed)

Open questions:

• Does 2x speed affect retention vs. 1x?
• Is there a "comprehension cliff" at certain speeds?
• Does speed preference correlate with topic familiarity?

Temporal Patterns

Time of Day

Hour        Consumption    Content Type
──────────────────────────────────────────
06-09       Low           News, short updates
09-12       Medium        Tutorials (work-adjacent)
12-14       Low           Background/podcast
14-17       Medium        Problem-solving searches
17-20       Low           Transition time
20-24       High          Deep learning, long content
00-03       Medium        Night owl learning

Day of Week

• Weekdays: Work-related, shorter, more searches
• Weekends: Longer sessions, deeper topics, new interests

Channel Analysis

Top Categories (91 channels tracked)

1. Development/Programming (28 channels)

   • Tutorials, framework updates, best practices

2. AI/ML (19 channels)

   • Papers explained, tool demos, news

3. Tech News (15 channels)

   • Product launches, industry trends

4. Productivity (12 channels)

   • Tools, workflows, systems

5. Domain Specific (17 channels)

   • Various specialized topics

Channel Engagement Patterns

Search vs. Browse

Discovery Mode

How new content is found:

• Algorithm recommendation: 40%
• Direct search: 30%
• Subscription feed: 20%
• External links: 10%

Retrieval Mode

How previously-watched content is re-accessed:

• Search transcript database: 60%
• Browser history: 20%
• Memory + manual search: 15%
• Never re-accessed: 5%

Key insight: Searchable transcripts change the retrieval pattern dramatically. Before: rarely re-access. After: frequent queries.

Retention Indicators

High Retention Signals

• Took notes during/after
• Searched transcript later
• Applied within 24 hours
• Discussed/explained to others

Low Retention Signals

• Background listening
• 3x+ speed
• No follow-up action
• No transcript search

Hypothesis

Retention correlates more with active retrieval (searching later) than with consumption parameters (speed, attention during watch).

Infrastructure Impact

Before Transcription Pipeline

• 5+ hours/day watching
• Low retrieval of past content
• Re-watching common
• "What was that tutorial?" frequent

After Transcription Pipeline

• Still ~5 hours/day watching
• High retrieval via search
• Re-watching rare (search instead)
• "What was that tutorial?" answered in seconds

Measured Savings

• Estimated 4 hours/day not re-watching
• Query resolution: seconds vs. 10-30 minutes
• Cross-video insights newly possible

Open Questions

1. Causation: Does the pipeline improve learning, or just retrieval?
2. Optimal consumption: What combination of speed + attention maximizes retention?
3. Diminishing returns: Is there a "too much" threshold for video learning?
4. Active vs. passive: How much of consumption is truly educational vs. entertainment?

Contribute your own consumption data or challenge these patterns.