serpapi-youtube

YouTube Search & Analysis Workflows

YouTube search, video analysis, transcript extraction, comment mining, and content research via SerpApi's YouTube engines. This skill covers workflow logic — load the serpapi foundational skill for API details, parameters, and wrapper script usage.

Dependencies

This skill builds on:

• serpapi — SerpApi wrapper script and full API reference (engines: youtube, youtube_video, youtube_video_transcript)

Core Concepts

Three YouTube Engines

Engine	Purpose	Key Input
youtube	Search videos, channels, playlists, Shorts	search_query
youtube_video	Video details, comments, related videos	v (video ID)
youtube_video_transcript	Video captions/transcript with timestamps	v (video ID)

Search Result Types

A YouTube search returns mixed result types:

• video_results[] — Videos with title, link, channel, views, published_date, description_snippet, length, thumbnail
• channel_results[] — Channels with title, handle, subscribers, link, verified status, description_snippet, thumbnail
• playlist_results[] — Playlists with title, link, video_count, videos preview
• shorts_results[] — YouTube Shorts with title, link, thumbnail, views
• movie_results[] — Movies/paid content when relevant
• ads_results[] — Promoted video results

Search Filters via sp Parameter

YouTube filtering uses the sp parameter with encoded tokens. Common filters:

Sort order:

• Sort by upload date: CAI%3D

Upload date:

• Today: EgIIAg%3D%3D
• This week: EgIIAw%3D%3D
• This month: EgIIBA%3D%3D
• This year: EgIIBQ%3D%3D

Duration:

• Under 4 minutes: EgIYAQ%3D%3D
• 4–20 minutes: EgIYAw%3D%3D
• Over 20 minutes: EgIYAg%3D%3D

Result type:

• Videos only: EgIQAQ%3D%3D
• Channels only: EgIQAg%3D%3D
• Playlists only: EgIQAw%3D%3D

Quality:

• 4K: EgJwAQ%3D%3D
• HD: EgIhAQ%3D%3D

Features:

• Live: EgJAAQ%3D%3D
• Subtitles/CC: EgIoAQ%3D%3D
• Creative Commons: EgIwAQ%3D%3D
• 360°: EgJ4AQ%3D%3D
• VR180: EgPQAQE%3D

Custom filters: Apply any filter combination on YouTube's website, then copy the sp value from the URL. This is the most reliable way to combine multiple filters.

Video Detail Response

The youtube_video engine returns rich data for a single video:

• Metadata — title, channel (name, link, subscribers, thumbnail), views, extracted_views, likes, extracted_likes, published_date
• Description — Full description with content (text) and extracted links
• Chapters — chapters[] with title, start_time, thumbnails (if the video has chapters)
• Key Moments — key_moments[] similar to chapters but auto-generated
• Comments — comments[] with author, text, likes, published_date, replies_count, and reply data
• Comment sorting — comments_sorting_token with tokens for "Top comments" vs "Newest first"
• Related videos — related_videos[] with title, link, channel, views, published_date, length
• Pagination tokens — comments_next_page_token, related_videos_next_page_token for more data

Transcript Response

The youtube_video_transcript engine returns:

• transcript[] — Array of segments with snippet (text), start_ms, end_ms, start_time_text
• chapters[] — Chapter markers with chapter (title), start_ms, end_ms
• available_transcripts[] — List of available transcript languages (with language, language_code, title, type)

Transcript options:

• language_code — e.g., en, es, ja (default: en)
• title — Select a specific transcript by name (e.g., "Twitch Chat - Simple")
• type — asr for auto-generated transcripts

Pagination

• YouTube search — Token-based. Use serpapi_pagination.next_page_token as the sp parameter for the next page
• Video comments — Token-based. Use comments_next_page_token as next_page_token
• Related videos — Token-based. Use related_videos_next_page_token as next_page_token
• Comment replies — Token-based. Use replies_next_page_token as next_page_token

Localization

• gl — Country code (e.g., us, uk, jp) — affects which videos appear and trending content
• hl — Language code (e.g., en, es, ja) — affects UI text and result language

Workflows

1. Topic Research

Find what content exists on YouTube for a given topic. Useful for content planning, SEO research, or general exploration.

Step 1: Broad search. Use the serpapi skill with the youtube engine. Search for the topic with search_query. Note total result types returned (videos, channels, Shorts).

Step 2: Filter by recency. Re-search with the upload date filter to see what's been published recently:

• This week (sp=EgIIAw%3D%3D) for trending content
• This month (sp=EgIIBA%3D%3D) for broader recent coverage

Step 3: Identify top performers. From the video results, note which videos have high view counts relative to their age. A video published a week ago with 500K views is more significant than one published 3 years ago with 2M views.

Step 4: Deep-dive top videos. Use the youtube_video engine on 3–5 top-performing videos to get:

• Full description (often contains keywords, links, topics)
• Chapters (reveals content structure)
• Related videos (expands the topic map)

Presentation:

📺 YouTube Research: [Topic]

🔍 Search Overview:
- Videos found: [count]
- Channels found: [count]
- Shorts found: [count]

🏆 Top Videos:
1. "[Title]" by [Channel] — [Views] views ([Date])
   Duration: [Length] | Chapters: [Yes/No]
2. ...

📈 Recent Trend (past week):
- [count] new videos published
- Fastest growing: "[Title]" ([Views] in [Days] days)

📺 Key Channels:
- [Channel] (@handle) — [Subscribers] subscribers, [verified status]
- ...

2. Video Deep Dive

Get comprehensive information about a specific video — metadata, content structure, audience reaction.

Step 1: Get video details. Use the youtube_video engine with the video ID.

Step 2: Extract transcript. Use the youtube_video_transcript engine with the same video ID. Check available_transcripts first for language options. Use language_code matching the video's primary language.

Step 3: Analyze comments. From the video details, review the initial comments. If deeper analysis is needed, paginate through more comments using comments_next_page_token. Switch to newest-first using comments_sorting_token to see recent sentiment.

Step 4: Check related videos. Review related_videos to understand what YouTube associates with this content.

Presentation:

📺 Video Analysis: "[Title]"

📊 Metrics:
- Views: [X] | Likes: [X]
- Published: [Date]
- Channel: [Name] ([Subscribers] subscribers)

📝 Content Structure:
[List chapters if available, or summarize transcript sections]

💬 Audience Reaction ([comment count] comments):
- Sentiment: [Positive/Mixed/Negative]
- Common themes: [list recurring topics from comments]
- Top comment: "[text]" — [likes] likes

🔗 Related Content:
- [Related video 1]
- [Related video 2]

3. Transcript Extraction & Analysis

Extract and work with video transcripts for summarization, content repurposing, or quote extraction.

Step 1: Check available transcripts. Use the youtube_video_transcript engine. The response includes available_transcripts[] showing all languages and types (manual vs auto-generated). Manual transcripts are more accurate; auto-generated (asr type) are available on most videos.

Step 2: Extract the transcript. Request with the desired language_code. If a specific named transcript is needed, use the title parameter.

Step 3: Process the transcript. The transcript returns as timestamped segments. For different use cases:

• Full text summary — Concatenate all snippet values. If chapters exist, summarize per chapter.
• Quote extraction — Search snippets for keywords, note start_time_text for timestamp references.
• Content outline — Use chapters[] as section headers, summarize transcript segments within each chapter's time range.
• Key moments — Find segments where topic changes or key points are made.

Presentation for summaries:

📝 Transcript Summary: "[Video Title]"

⏱️ Duration: [X] minutes | Language: [X] | Type: [Manual/Auto-generated]

📑 Chapters:
1. [Chapter Title] (0:00) — [Summary]
2. [Chapter Title] (3:45) — [Summary]
...

🔑 Key Quotes:
- "[Quote]" — [Timestamp]
- "[Quote]" — [Timestamp]

📋 Full Summary:
[Paragraph summary of the entire video]

4. Comment Mining

Extract insights from video comments — audience questions, sentiment, feature requests, feedback patterns.

Step 1: Get initial comments. Use the youtube_video engine. The response includes the first page of comments and sorting options.

Step 2: Choose sort order.

• Top comments (default) — most liked/relevant comments, best for sentiment analysis
• Newest first — use comments_sorting_token to switch; best for recent reactions or tracking response to events

Step 3: Paginate for volume. Use comments_next_page_token as next_page_token to get more comments. Each page returns ~20 comments. For thorough analysis, collect 3–5 pages (60–100 comments).

Step 4: Analyze patterns. Look for:

• Questions — Comments with ? or asking for help → content gap indicators
• Feature requests — "Wish you'd cover..." or "Can you make a video about..."
• Complaints — Negative sentiment → pain points
• Praise — Specific compliments → what resonates
• Timestamps — Comments referencing specific moments (e.g., "5:32 was hilarious") → highlights

Step 5: Check replies. High-engagement comments often have reply threads. Use replies_next_page_token to expand them. Creator replies are especially valuable (marked by channel author badges).

Presentation:

💬 Comment Analysis: "[Video Title]"

📊 Overview:
- Total comments (estimated): [X]
- Analyzed: [X] comments across [X] pages
- Sort: [Top / Newest]

😊 Sentiment Breakdown:
- Positive: ~[X]%
- Neutral: ~[X]%
- Negative: ~[X]%

❓ Top Questions Asked:
1. "[Question]" — [likes] likes
2. "[Question]" — [likes] likes

💡 Common Themes:
- [Theme 1] — mentioned [X] times
- [Theme 2] — mentioned [X] times

🔥 Most Engaged Comments:
- "[Comment excerpt]" — [likes] likes, [replies] replies

5. Channel Research

Research a YouTube channel's content strategy, top videos, and audience.

Step 1: Find the channel. Search with the youtube engine using the channel name. Look for channel_results[] to find the exact channel with subscriber count and verification status.

Step 2: Get their content. Search for videos from the channel: use search_query with the channel name or include site:youtube.com/@handle patterns. Filter by recency to see recent uploads.

Step 3: Analyze top videos. Use the youtube_video engine on 5–10 of their videos (mix of top-performing and recent) to extract:

• View counts and engagement (likes)
• Content structure (chapters)
• Description links and calls-to-action
• Related videos (what YouTube associates with their content)

Step 4: Review audience via comments. Check comments on 2–3 videos to understand audience demographics, sentiment, and what they request.

Presentation:

📺 Channel Analysis: [Channel Name] (@handle)

📊 Channel Overview:
- Subscribers: [X]
- Verified: [Yes/No]
- Description: [excerpt]

🎬 Content Overview (from [X] analyzed videos):
- Average views: [X]
- Average likes: [X]
- Publishing frequency: ~[X] videos/month
- Typical video length: [X] minutes

🏆 Top Videos:
1. "[Title]" — [Views] views ([Date])
2. "[Title]" — [Views] views ([Date])

📈 Recent Performance (last month):
- Videos published: [X]
- Total views: [X]
- Best performer: "[Title]"

💡 Content Themes:
- [Theme 1] — [X] videos
- [Theme 2] — [X] videos

👥 Audience Insights (from comments):
- [Key insight]
- [Common request or question]

6. Content Brief from YouTube Research

Build a content brief for a new video or article based on what's already on YouTube.

Step 1: Research the topic. Follow Workflow 1 (Topic Research) to find existing content.

Step 2: Extract transcripts from top videos. Use Workflow 3 (Transcript Extraction) on the 3–5 best-performing videos. Focus on chapter structures and key points.

Step 3: Mine comments for gaps. Use Workflow 4 (Comment Mining) on top videos. Look for unanswered questions — these are content opportunities.

Step 4: Map related content. From the related_videos of top performers, identify adjacent topics that could be covered.

Step 5: Synthesize the brief.

📋 Content Brief: [Topic]

🎯 Opportunity:
- [Why this topic is worth covering]
- Search volume signal: [X] videos in the past [timeframe]

📺 Existing Top Content:
1. "[Title]" by [Channel] — [Views] views
   Key points: [summary]
   Gap: [what it doesn't cover]
2. ...

❓ Audience Questions (from comments):
1. [Question not answered in existing videos]
2. ...

📝 Recommended Structure:
1. [Section] — Cover [topic], which top videos miss
2. [Section] — Address [common question]
3. [Section] — Include [unique angle]

🔑 Must-Include Points:
- [Point from transcript analysis]
- [Point from comment mining]

🎯 Differentiation:
- [How to stand out from existing content]

7. Multi-Video Comparison

Compare multiple videos on the same topic — useful for benchmarking, review analysis, or finding the best resource.

Step 1: Find competing videos. Search the topic with the youtube engine. Identify 3–5 videos covering the same subject.

Step 2: Get details for each. Use the youtube_video engine on each video to collect metadata.

Step 3: Extract transcripts for each. Use the youtube_video_transcript engine to get content for comparison.

Step 4: Compare across dimensions.

📊 Video Comparison: [Topic]

| Metric | Video A | Video B | Video C |
|--------|---------|---------|---------|
| Title | [title] | [title] | [title] |
| Channel | [name] | [name] | [name] |
| Views | [X] | [X] | [X] |
| Likes | [X] | [X] | [X] |
| Duration | [X] min | [X] min | [X] min |
| Chapters | [Y/N] | [Y/N] | [Y/N] |
| Published | [date] | [date] | [date] |

📝 Content Coverage:
- Video A covers: [topics]
- Video B covers: [topics]
- Video C covers: [topics]
- Unique to A: [topics]
- Unique to B: [topics]
- Common to all: [topics]

🏆 Best For:
- Quick overview: [Video X] (shortest, most concise)
- Deep learning: [Video Y] (most comprehensive)
- Most engaging: [Video Z] (highest like ratio)

Common Patterns

"Find YouTube videos about [topic]"

1. Search with youtube engine, search_query=[topic]
2. Present top 5–8 video results with views, date, channel
3. Offer to dive deeper into any specific video

"What does [video] cover?"

1. Get video details with youtube_video
2. Get transcript with youtube_video_transcript
3. Summarize using chapters (if available) or transcript sections

"Get me the transcript of this video"

1. Use youtube_video_transcript with the video ID
2. Check available_transcripts for language options
3. Return formatted transcript with timestamps

"What are people saying about [video]?"

1. Get video details (includes first page of comments)
2. Paginate if needed for more comments
3. Summarize sentiment and recurring themes

"Find recent videos about [topic]"

1. Search with youtube engine
2. Filter with sp=EgIIAw%3D%3D (this week) or sp=EgIIBA%3D%3D (this month)
3. Sort by upload date with sp=CAI%3D for chronological order

"What videos are similar to [this one]?"

1. Get video details with youtube_video
2. Present related_videos[] from the response
3. Offer to analyze any of the related videos

"Research [channel name] on YouTube"

1. Search with youtube engine for the channel name
2. Identify the channel from channel_results[]
3. Follow Channel Research workflow (Workflow 5)

"Build me a content brief about [topic]"

1. Follow Content Brief workflow (Workflow 6)
2. Combine topic research, transcript analysis, and comment mining

Tips

• Video ID extraction — The video ID is the v parameter from any YouTube URL: youtube.com/watch?v=VIDEO_ID. Also works from short URLs: youtu.be/VIDEO_ID.
• sp filter stacking — To combine filters, apply them on YouTube's UI and copy the resulting sp value from the URL. Manually combining encoded tokens is unreliable.
• Views vs age — Always contextualize view counts with publish date. Calculate views-per-day for fair comparison across videos of different ages.
• Transcript availability — Most videos have auto-generated (ASR) transcripts. Manual transcripts are higher quality but less common. Check available_transcripts first.
• Comment sampling — Top comments (default sort) are biased toward early, popular responses. For recent sentiment, switch to newest-first sorting.
• Localization matters — Set gl and hl to match your target audience. YouTube serves different trending content and search results per region.
• Shorts vs videos — Shorts appear in shorts_results[], separate from video_results[]. They have different engagement dynamics (higher views, lower engagement depth).
• Pagination costs — Each page of results costs one SerpApi search credit. Be strategic — 2–3 pages of comments is usually enough for analysis.
• Channel content — YouTube search doesn't have a direct "list all videos by channel" API. Search with the channel name + topic keywords, or use multiple searches to build a picture.
• Chapters as structure — Videos with chapters are easier to analyze. Use chapter titles as section headers when summarizing transcripts.