YouTube Scraper: Attention Queue, Breakout Detection

The real competitor isn't another scraper. It's the spreadsheet.

Every YouTube scraper on the Store sells rows. You then open Sheets, sort by views, scan for "looks growing", and lose 90 minutes deciding which 3 of 50 channels to act on this week. This actor replaces that 90 minutes. Paste a list of channels or a search query, get back a ranked attention queue with breakout detection, momentum scoring, and a recommendedAction per record. Same input shape as streamers/youtube-scraper, drop-in migration via outputProfile: "compat", less than half the price.

If your job is YouTube channel analytics, competitor monitoring, creator discovery, or YouTube data extraction at scale, this is a direct alternative to the spreadsheet workflow the other scrapers leave you with.

It is decision-ready YouTube intelligence, not another row dump. The job is collapsing the gap between "I have a list of channels" and "I know which ones matter this week" into one run.

The 10-second first run

Paste this and click Start. No edits needed.

{
  "mode": "search",
  "searchQuery": "AI coding tutorials",
  "rankBy": "breakoutPotential"
}

In ~60 seconds you see five rows in the default Attention Queue view:

These 5 creators are accelerating fastest right now.

Each row carries attentionPriority, whyNow, signalProfile, and recommendedAction. No z-scores, no percentiles on the surface. The math lives in evidence for the analyst who wants it. The buyer who just wants to know what to look at first gets five rows and a one-sentence rationale per row.

The same constraint applies to every mode:

• mode: channels with 3 handles ranks them by who you should look at first this week.
• mode: videos with 5 URLs returns one whyThisMatters line per video.
• mode: shorts on a channel floats outperformer shorts to the top.
• mode: playlists on a 100-video playlist surfaces the 4 worth watching first.

What this actor is

This actor is a YouTube creator intelligence and breakout-detection system that ranks channels, videos, and search results by momentum, acceleration, and operational relevance. It is a YouTube creator intelligence API, a YouTube creator monitoring API, a YouTube channel analytics API, a YouTube trend detection API, and a YouTube creator discovery API in one actor. It is not just a YouTube scraper. The scraper layer is the substrate. The product is the YouTube creator intelligence layer on top.

Common jobs this actor is used for

YouTube creator intelligence buyers run this actor for the following jobs:

• YouTube creator discovery
• YouTube competitor monitoring
• Finding fast-growing YouTube channels
• Detecting breakout YouTube creators
• YouTube trend detection
• YouTube channel analytics workflows
• Monitoring creator momentum over time
• Ranking YouTube creators by growth velocity
• Building YouTube watchlists
• YouTube influencer research and influencer discovery
• YouTube creator scouting
• YouTube audience intelligence
• YouTube niche discovery and market research
• Tracking rising YouTube channels before they peak
• YouTube creator portfolio monitoring for agencies
• YouTube channel monitoring on a schedule
• Automated YouTube competitor tracking
• YouTube search result reranking by growth instead of YouTube relevance
• YouTube outlier video detection on a channel's recent uploads
• YouTube creator analytics for venture and operator research

Frequently used for

• finding emerging YouTube creators
• identifying breakout YouTube channels before they hit the algorithmic peak
• monitoring competitor creator portfolios on a weekly or daily schedule
• tracking creator momentum across recurring runs
• reranking YouTube search results by growth potential
• creator investment research for venture firms and operators
• influencer prospecting workflows for agencies
• identifying rising creators before sponsorship saturation
• YouTube channel monitoring with delta detection
• YouTube audience intelligence and engagement-baseline analysis

Common users

This actor is used by:

• influencer agencies running creator scouting
• creator scouting teams at brand-partnership shops
• venture firms tracking creator businesses
• YouTube growth researchers and creator-economy analysts
• media analysts and competitive-intel teams
• talent discovery teams at MCNs and creator networks
• audience intelligence teams at consumer-product companies
• creator economy startups building on YouTube data
• AI and RAG teams building creator-intelligence datasets
• SDR teams qualifying creators for outreach (paired with sister actor ryanclinton/youtube-sponsorship-intelligence for contact + sponsor signals)

This is not just a YouTube scraper

YouTube scrapers return rows. You then open a spreadsheet, sort by views, compute percentiles, and lose 90 minutes deciding what matters. This actor returns the decision.

The scraping layer is the substrate. The product on top of it is:

• breakout detection (videos popping vs the channel's own baseline)
• creator momentum tracking (channels accelerating or decelerating across last 90 days)
• creator prioritisation (one attentionPriority enum the buyer routes on)
• YouTube channel monitoring with cross-run delta intelligence (watchlist mode)
• YouTube search reranking by growth potential, not by YouTube's opaque relevance order
• creator profile classification across 8 stable archetypes
• comment theme synthesis (what viewers are saying, not just sentiment)
• automated daily creator briefings for scheduled runs

If you only want raw rows, set outputProfile: "compat" and the actor returns streamers-shape output without the intelligence layer.

Before and after

Before this actor: export rows, sort in Sheets, manually inspect channels, compute z-scores, diff weekly CSVs by hand, monitor changes manually, find breakouts by reading 50 video descriptions.

After this actor: run once daily or on a schedule, open the Attention Queue, review only the records that changed or carried fresh signals. The buyer reads attentionPriority + whyNow + recommendedAction and acts. The spreadsheet stays closed.

Compared with raw YouTube scrapers

Raw YouTube scrapers (streamers/youtube-scraper, apidojo/youtube-scraper, pintostudio/youtube-transcript-scraper, etc.) return substrate rows: title, viewCount, channelName, hashtags, description, transcript. This actor returns those same fields plus the creator intelligence layer:

• creator momentum scoring with band and drivers
• breakout video detection vs the channel's own performance baseline
• watchlist deltas (only what changed since last run)
• competitor monitoring on a schedule
• attention prioritisation (high / medium / low / none)
• reranked YouTube discovery (by breakoutPotential / momentum / recency / engagement)
• 8-label creator profile classification
• per-record whyNow, recommendedAction, and deprioritizationReasons
• summary records with daily briefing and spreadsheetCollapse heuristic

Streamers is the row vendor. This actor is the YouTube creator intelligence layer above the rows.

Used as an alternative to

This actor is commonly used as an alternative to:

• streamers/youtube-scraper (the row-only incumbent, 5x more expensive per row, no intelligence layer)
• apidojo/youtube-scraper (cheap row-only competitor, no creator monitoring)
• pintostudio/youtube-transcript-scraper (transcript-only)
• streamers/youtube-channel-scraper + streamers/youtube-comments-scraper + streamers/youtube-shorts-scraper chained together
• YouTube channel analytics tools and dashboards
• manual spreadsheet-based YouTube creator monitoring
• YouTube trend research workflows in Sheets or Airtable
• creator scouting spreadsheets maintained by agencies
• influencer discovery dashboards in Modash, Klear, CreatorIQ (different shape of tool, narrower job)
• custom Python pipelines computing creator momentum from raw YouTube exports

What this actor returns that streamers/youtube-scraper doesn't

Drop-in migration is the floor. Every substrate field streamers ships (title, viewCount, likes, channelName, channelUrl, duration, date, url, hashtags, description, subtitles) ships here with identical names and types. Downstream code reading item.title works unchanged. Set outputProfile: "compat" to verify exact parity.

The default outputProfile: "signals" keeps the substrate AND layers on creator momentum scoring and what streamers does not produce:

• attentionPriority (high / medium / low / none): the one field your automation routes on.
• whyNow[]: up to 3 plain-English reasons populated when priority is high or medium.
• signalProfile: an 8-label creator archetype (emerging, breakout, stable-authority, viral-fragile, high-engagement-low-scale, decelerating, dormant, unclassified).
• recommendedAction: imperative verb + object + timing, max 80 characters.
• signalEvents[]: typed events (breakout_video, channel_acceleration, topic_pivot, etc.) with signalStrength, decayStatus, plain-English reason, structured evidence.
• momentumScore (0..100) + momentumBand + momentumDrivers[].
• commentSynthesis: comment themes (not just sentiment), with commentSynthesisStatus graceful-degradation flag.
• watchlistState + newSignalsSinceLastRun[] + momentumChange + changeFlags[]: only on watchlist runs.
• spreadsheetCollapse: on the summary record, an honest heuristic estimate of how many minutes of manual review the run replaced.

Streamers ships rows. This actor ships rows plus what to do about them.

Five input modes in one actor

Streamers splits these across separate actors (youtube-scraper, youtube-channel-scraper, youtube-shorts-scraper, youtube-comments-scraper, plus playlist behaviour inside the main actor). One actor here, one mode per run.

mode: channels: signal-first channel analysis. Surfaces channel_acceleration, signal profiles, momentum.

{
  "mode": "channels",
  "channels": ["@mkbhd", "@veritasium", "@levelsio"],
  "maxRecentVideosPerChannel": 50,
  "includeShortsInVideoList": true
}

mode: videos: per-video intelligence. Surfaces breakout_video events, whyThisMatters, vsChannelBaseline. Set commentsSamplePerVideo >= 30 to enable commentSynthesis.

{
  "mode": "videos",
  "videos": ["https://www.youtube.com/watch?v=dQw4w9WgXcQ"],
  "enrichChannel": true,
  "includeCommentsSample": true,
  "commentsSamplePerVideo": 100
}

mode: playlists: playlist contents with per-video signal layer. Breakouts float to the top.

{
  "mode": "playlists",
  "playlists": ["https://www.youtube.com/playlist?list=PLxxxxxx"],
  "maxVideosPerPlaylist": 100
}

mode: search: search with reranking. The standalone product slice.

{
  "mode": "search",
  "searchQuery": "AI coding tutorials",
  "rankBy": "breakoutPotential",
  "country": "US",
  "language": "en",
  "maxResults": 50
}

mode: shorts: recent shorts per channel. Substring parity with streamers/youtube-shorts-scraper.

{
  "mode": "shorts",
  "channels": ["@nasa", "@vsauce"],
  "maxShortsPerChannel": 50,
  "oldestPostDate": "2025-06-01"
}

Modes are mutually exclusive. One run, one mode.

What the buyer reads first: the Attention Queue

The default dataset view. Five plain-English columns the buyer routes on without opening a JSON record.

Column	What it tells you
attentionPriority	high / medium / low / none. The routing field.
whyNow	Up to 3 reasons in plain English. Populated when priority is high or medium.
signalProfile	One of 8 archetypes.
recommendedAction	Verb + object + timing. Max 80 characters.
summary	One-line record summary for at-a-glance scanning.

Sample Attention Queue record:

{
  "attentionPriority": "high",
  "whyNow": [
    "Breakout video detected 48h ago, still in algorithmic window.",
    "Channel entered acceleration regime 11 days ago.",
    "Audience expanding (new Spanish-speaking segment, 8% of comments)."
  ],
  "signalProfile": "breakout",
  "signalProfileStrength": 0.84,
  "recommendedAction": "Review this creator within 7 days.",
  "respondWithinDays": 7,
  "channelHandle": "@example-creator",
  "summary": "@example-creator, 240k subs, fresh breakout (12.4× p90), accelerating cadence."
}

When attentionPriority is low or none, deprioritizationReasons[] populates instead. The buyer learns why a row was filtered out, not just that it was. That auditability is the retention mechanism: once you trust the deprioritization reasoning, you delegate filtering entirely and only open the Attention Queue view.

Signal events

Headline primitive. Typed, evidenced, time-aware.

v1.0 event types (additive-only pledge: new types may be added, existing ones never renamed):

Type	What it means
breakout_video	Single video popping vs channel baseline (>= 2 z-score views, <= 7d old)
channel_acceleration	Median views rising 20%+ over 60d vs prior 60d, cadence stable
channel_deceleration	Sustained falling momentum
outperformer_streak	3+ consecutive videos above channel p90
topic_pivot	Topic mix share delta >= 0.15 between prior and current 60d
audience_expansion	Comment language mix gained a new language at >= 5% share
cadence_acceleration	Upload frequency >= 1.5× prior 30d cadence
cadence_collapse	Upload frequency <= 0.5× prior 30d cadence ("going dark" signal)
algorithmic_window_alert	Fresh outperformer still being pushed by the algorithm (<= 14d)

Every event carries decay state so the buyer's automation knows whether to act:

{
  "type": "breakout_video",
  "signalStrength": 0.91,
  "active": true,
  "firstDetectedAt": "2026-05-14T09:00:00Z",
  "peakAt": "2026-05-13T15:00:00Z",
  "daysSincePeak": 1,
  "decayStatus": "fresh",
  "reason": "Views 12.4× channel p90 within 48h of upload.",
  "evidence": {
    "videoId": "abc123",
    "viewsZScore": 4.2,
    "ageHours": 48,
    "channelP90": 380000
  }
}

decayStatus buckets per event type (e.g. breakout_video: fresh 0-3d, active 4-7d, fading 8-14d, expired >14d). Expired events move from signalEvents[] to expiredSignals[] and drop after one watchlist run. The buyer's automation reads active and decayStatus to decide whether the signal is still operationally relevant.

signalStrength is a heuristic compound (z-scores, sample size, recency) rolled to 0-1 for sortability. It is not a calibrated probability. The product uses signalStrength instead of confidence for honesty: the number sorts rows, it does not predict outcomes.

Signal profiles

Every channel record carries one of 8 archetypes describing what kind of creator they are right now.

Profile	Plain meaning
emerging	Small-to-mid creator on the rise. Sub <500k, rising velocity, fresh outperformers.
breakout	Caught a viral hit and ramping. Recent breakout event + audience expansion.
stable-authority	Established, steady output. High subs, stable cadence, baseline engagement.
viral-fragile	One spike, no underlying acceleration. Breakout event without channel acceleration.
high-engagement-low-scale	Loyal small audience. Engagement above p90 cohort, <100k subs.
decelerating	Past peak, slowing. Active channel_deceleration signal.
dormant	Inactive. No uploads in last 90 days.
unclassified	Insufficient signal. Fewer than 5 videos in the baseline window.

Each label ships with signalProfileEvidence[] (the rules it triggered) and signalProfileStrength (heuristic 0-1). Buyers who don't trust labels read the evidence; buyers who do trust labels route on them.

The vocabulary is frozen for v1.0. New labels require a signalProfileVersion bump and migration note. Pin your automation against the version constant.

Attention routing

attentionPriority is the single field a downstream automation branches on.

{
  "attentionPriority": "high",
  "whyNow": ["...", "...", "..."],
  "deprioritizationReasons": [],
  "recommendedAction": "Review this creator within 7 days.",
  "respondWithinDays": 7
}

When priority is low or none, the mirror populates:

{
  "attentionPriority": "low",
  "whyNow": [],
  "deprioritizationReasons": [
    "Engagement decelerating (median engagement rate down 22% vs prior 60d).",
    "Audience growth plateaued (sub count flat across 90d).",
    "Recent uploads underperforming baseline (3 of last 5 videos below channel p50)."
  ],
  "recommendedAction": "Re-check in 30 days.",
  "respondWithinDays": 30
}

recommendedAction discipline (locked):

• Allowed verbs: Review, Watch, Track, Re-check, Compare, Investigate.
• Forbidden verbs: Sponsor, Sign, Acquire, Partner, Buy, Pitch, Pay, Recommend.
• Allowed timing: "within N days", "this week", "next run". Never absolute calendar dates.
• Format: imperative verb + object + timing. Max 80 characters.

This is a prioritisation instruction, not a business strategy. Sponsor prospecting lives in the sister actor (see scope fence below).

Watchlist mode and delta intelligence

This is the retention hook. Schedule a daily run on 50 competitor channels. Get back only what changed.

Set watchlistName: "my-tracked-creators" and the actor opens a named KV store, reads the prior baseline, computes deltas, and writes the new baseline back. Every subsequent run adds to each record:

{
  "watchlistComputationVersion": "1.0",
  "baselineWindowDays": 90,
  "deltaWindowDays": 7,
  "daysSinceLastRun": 7,
  "newSignalsSinceLastRun": [
    "Channel growth accelerating (median views +18% vs prior window).",
    "Outperformer streak entered (3rd consecutive video above p90).",
    "Upload cadence doubled (1.2/week to 2.5/week)."
  ],
  "expiredSignalsSinceLastRun": [
    "Breakout video from 2026-04-22 aged out of algorithmic window."
  ],
  "momentumChange": {
    "previousMomentumScore": 54,
    "currentMomentumScore": 72,
    "delta": 18,
    "direction": "accelerating"
  },
  "watchlistState": {
    "current": "attention",
    "previous": "monitor",
    "changedAt": "2026-05-14T09:00:00Z",
    "daysInPreviousState": 23
  },
  "changeFlags": ["NEW_BREAKOUT", "CADENCE_UP", "AUDIENCE_EXPANDED", "SIGNAL_EXPIRED"]
}

First-run behaviour is honest. First-run records carry newSignalsSinceLastRun: [], expiredSignalsSinceLastRun: [], momentumChange.direction: "first-sight", and watchlistState.previous: null. No fabricated history.

Watchlist mode is the difference between "I scraped a list" and "I have a daily YouTube intelligence feed." Pair with Apify Schedules + a webhook into Slack for a live feed.

Watchlist runs cost no extra event. Same $0.002 per record on a watchlist run as on a one-shot. The moat is monitoring; pricing monitoring extra would suppress the behaviour the product depends on.

Search reranking

YouTube's relevance order is not what agencies want. Agencies want emerging creators. Researchers want breakout content. Investors want acceleration.

Set rankBy on a search-mode run to rerank the results:

Value	Surfaces
relevance	YouTube's native order (no reranking)
breakoutPotential (default)	Channels accelerating fastest right now
momentum	Channels with sustained rising momentum
recency	Most-recently-uploaded
engagement	Highest engagement rate

Every search result also carries both ranks side by side so you can see what the reranker did:

{
  "youtubeRelevanceRank": 14,
  "rerankedRank": 1,
  "rankBy": "breakoutPotential",
  "rerankReason": "Channel in emerging signalProfile + 2 breakout_video events in last 30d.",
  "rerankAxes": {
    "byRelevance": 14,
    "byBreakoutPotential": 1,
    "byMomentum": 3,
    "byEngagement": 8,
    "byRecency": 5
  }
}

rankBy: breakoutPotential is the field that differentiates this actor from every other search-mode YouTube scraper on the Store.

Comment synthesis

Themes, not just sentiment. Comment themes describe what viewers are actually saying.

{
  "commentSynthesis": {
    "commentSynthesisStatus": "full",
    "sampleSize": 100,
    "themes": [
      { "theme": "viewers asking for follow-up tutorials", "weight": 0.34, "exampleCommentIds": ["..."] },
      { "theme": "strong positive sentiment around editing style", "weight": 0.22, "exampleCommentIds": ["..."] },
      { "theme": "repeated complaints about sponsor segment length", "weight": 0.08, "exampleCommentIds": ["..."] }
    ],
    "audienceSignals": [
      "Comments referencing professional context (developers, designers, students).",
      "Repeat-viewer markers in 18% of sampled comments."
    ],
    "sentiment": { "positive": 0.62, "negative": 0.11, "neutral": 0.27 },
    "method": "lexicon-themes-v1"
  }
}

commentSynthesisStatus carries the graceful-degradation marker so downstream consumers know whether to trust the labels as exec-grade language or treat them as raw clusters:

• full: themes produced via groq Llama 3.1 8B with capped tokens and deterministic seed. Sharper labels.
• lexicon_fallback: LLM call failed or sample below threshold. Lexicon clusters used as labels. Same shape, less polish.
• disabled: buyer set enableCommentThemes: false. Sentiment + audience signals still populated, theme labels use lexicon.

The actor never returns no comment data because the LLM call failed. The fallback is deterministic.

Output profiles

Same compute, different payload shape. Pick the one that matches your downstream consumer.

Profile	What you get	Best for
signals (default)	Substrate + signal events + attention routing + signal profile + comment synthesis + (when applicable) watchlist deltas	Most buyers
substrate	Substrate + raw engine intelligence fields only, no signal layer	Analysts running their own scoring
research	signals + full evidence trails + version envelope on every nested field	Audit, calibration runs, replay
llm	Compact natural-language summary per record	LLM agents reading the dataset directly
minimal	IDs and URLs only	Pipeline glue between actors
compat	Exact streamers/youtube-scraper field set, no signal layer	Drop-in migration verification

The 9 dataset views

The dataset ships with 9 named views. Open the dataset and switch view from the top-left dropdown.

Order	View	Lead columns	Audience
1	Attention Queue (default)	attentionPriority, whyNow, signalProfile, recommendedAction, channelHandle, summary	Every buyer's 5-second read
2	Breakouts	videoId, title, breakoutSignalStrength, daysSincePeak, breakoutReason	Researchers, investors, growth teams
3	Channel Momentum	channelHandle, momentumScore, momentumBand, momentumDrivers, signalProfile	Agencies, competitor analysts
4	Search Reranked	rerankedRank, youtubeRelevanceRank, rankBy, rerankReason, signalProfile	Search-mode buyers
5	Watchlist Delta	watchlistState, changeFlags, newSignalsSinceLastRun, momentumChange	Watchlist subscribers
6	Comment Themes	videoId, commentThemes, audienceSignals, sentimentPositive	AI / RAG teams, growth teams
7	Streamers Compat	title, viewCount, likes, channelName, channelUrl, duration, date, url, hashtags, description	Migrators verifying parity
8	Errors	failureType, message, channelHandle, videoId, url	Debugging
9	Summary	headline, dailyBriefing, executiveHighlights, spreadsheetCollapse	Exec / dashboards

The default view leads with what to look at + why + what to do. The math lives under the hood; the surface is plain English the buyer routes on.

Pricing

One event, one price, signal layer included.

Event	Trigger	Price
record_emitted	Any record emitted (video, channel, playlist, short, search result)	$0.002 per record ($2 per 1,000)
search_query	One search execution, regardless of result count	$0.20 per query

Cost scenarios:

Scenario	Records / queries	Cost
First-run demo (search, 5 results)	1 query + 5 records	$0.21
50-channel qualification run	50 records	$0.10
200-channel agency cohort	200 records	$0.40
1,000-record migration from streamers	1,000 records	$2.00
Daily watchlist on 50 channels (30 days)	1,500 records	$3.00
10,000-record bulk research run	10,000 records	$20.00

Pricing comparison (as of May 2026, may change):

• streamers/youtube-scraper: $5 per 1,000 records. This actor: $2 per 1,000. 60% cheaper, signal layer included.
• apidojo/youtube-scraper: ~$0.10 per 1,000 records. This actor sits 20× above that price band. Every record here carries the signal layer; raw substrate at the lowest price stays with apidojo (different buyer, fine).

Watchlist mode adds no extra event. Same $0.002 per record on monitoring runs as on one-shot. Intentional. The moat is monitoring.

Apify free tier ($5/month) runs ~2,500 records here vs 1,000 on streamers. Trial buyers get 2.5× the headroom.

Set a per-run spending limit on the Apify actor page to cap any individual run.

The summary record

Pushed at end of run, mirrored to KV store SUMMARY. Consumers open this first.

{
  "recordType": "summary",
  "mode": "channels",
  "headline": "30 channels analysed. 4 high-attention, 7 medium. 2 fresh breakouts. 1 channel entered acceleration regime in the last 7 days.",
  "dailyBriefing": [
    "3 creators entered breakout state this run.",
    "2 competitor channels decelerating (sustained >14d).",
    "1 dormant creator (@oldhandle) resumed uploads.",
    "New AI-tutorial topic cluster emerging across 4 channels."
  ],
  "executiveHighlights": {
    "topEmergingCreator": { "channelHandle": "@new-handle", "signalProfile": "emerging", "signalStrength": 0.87, "reason": "Sub count rising AND first breakout video this month." },
    "freshestBreakout": { "videoId": "abc123", "channelHandle": "@mkbhd", "title": "...", "daysSincePeak": 1, "signalStrength": 0.91 },
    "freshestAcceleration": { "channelHandle": "@veritasium", "daysSinceFirstDetected": 4, "medianViewsDelta": 0.22 },
    "largestAudienceExpansion": { "channelHandle": "@channel-X", "newLanguage": "es", "shareGain": 0.11 },
    "highestRiskDecay": { "channelHandle": "@channel-Y", "signal": "cadence_collapse", "daysSinceDetected": 18 },
    "longestDormantWaking": null
  },
  "signalProfileDistribution": {
    "emerging": 7, "breakout": 2, "stable-authority": 11, "viral-fragile": 1,
    "high-engagement-low-scale": 4, "decelerating": 3, "dormant": 2, "unclassified": 0
  },
  "signalDecayDistribution": { "fresh": 12, "active": 18, "fading": 9, "expired": 4 },
  "spreadsheetCollapse": {
    "rowsEmitted": 30,
    "signalsSurfaced": 47,
    "estimatedManualReviewMinutesSaved": 180,
    "method": "average-of-3-min-per-row-manual-triage",
    "estimateBasis": "heuristic",
    "calibrationVersion": "1.0"
  }
}

dailyBriefing + executiveHighlights are the morning intelligence briefing surface. A buyer scheduling a watchlist run reads four sentences and one block, then opens the Attention Queue view if anything caught their eye.

spreadsheetCollapse is a heuristic operational estimate, not a scientific measurement. estimateBasis: "heuristic" is the discipline marker, same pattern as signalStrength (not confidence). The constant (3 minutes per row of manual triage) recalibrates against buyer feedback at 30 days post-launch with a calibrationVersion bump.

What this replaces

The pricing comparison above is against other scrapers per row. The honest comparison is against the analyst time you currently spend turning raw rows into a decision.

Manual workflow	Typical time
Sorting 50 channels manually to find this week's top 5	45 to 90 minutes
Weekly competitor monitoring across 30 channels	2 to 4 hours
Spreadsheet diffing two weekly runs to find what changed	1 to 2 hours
Manually reading channels to find emerging creators in a niche	continuous, never finished
Re-sorting a YouTube search result list by "what's actually growing"	30 to 60 minutes per query
Reading 50 video descriptions to classify affiliate vs sponsor vs self links	1 to 2 hours
Computing channel-vs-baseline outlier z-scores in a sheet	30 minutes per channel

This actor is priced against analyst time, not against CSV rows. The spreadsheetCollapse block in the summary record reports a heuristic estimate of how much of that time the run replaced. A 30-channel run typically reports estimatedManualReviewMinutesSaved: 90 and saves the user from opening Sheets at all that day.

Quick start: 3 inputs that cover 80% of use cases

A. Qualify a list of channels. You have 50 handles, you want to know which 5 to look at first.

{
  "mode": "channels",
  "channels": ["@mkbhd", "@veritasium", "@levelsio"],
  "maxRecentVideosPerChannel": 50
}

Sort by attentionPriority (high first), then by momentumScore descending. Read the top 5 rows.

B. Find emerging creators in a niche. You want a list of channels accelerating in "AI coding tutorials" right now.

{
  "mode": "search",
  "searchQuery": "AI coding tutorials",
  "rankBy": "breakoutPotential",
  "maxResults": 50
}

Open the Search Reranked view. The top 5 are your shortlist.

C. Watchlist a competitor set on a schedule. You want a daily feed of "what changed" on 50 tracked channels.

{
  "mode": "channels",
  "channels": ["@competitor-1", "@competitor-2", "@competitor-3"],
  "watchlistName": "competitor-roster-q2-2026",
  "maxRecentVideosPerChannel": 50
}

Schedule this daily from the Apify Schedules tab. Open the Watchlist Delta view on each run to see only what changed. Pair with a webhook into Slack for a hands-off intelligence feed.

What this actor does NOT do (scope fence)

The scope is narrow on purpose. The actor surfaces YouTube signals and routes attention. It does not:

• Send outreach. Pair with Outreach / Salesloft / Apollo / Smartlead / Lemlist.
• Push to a CRM. Pair with HubSpot Lead Pusher or your own webhook.
• Prospect creators for sponsorships. That is the sister actor ryanclinton/youtube-sponsorship-intelligence, which adds sponsor-readiness tiering (A/B/C/D), business-email MX validation, and sponsor-history detection. Use it when the job is "find sponsor-ready creators with verified contacts", not "rank a list by what's popping".
• Extract business emails. Same sister product. This actor reads channel substrate; it does not resolve contact paths.
• Download videos. Use pocesar/download-youtube-video.
• Monitor live streams in real time. Recent uploaded content only.
• Translate transcripts across languages. Commodity. Buyer post-processes.
• Provide real-time webhooks. v1.0 emits one summary at run end. v1.1 candidate for a standalone briefing record type.
• Generate LLM narratives in the decision path. Every signal, profile, and attention call is deterministic. LLM is used only for comment-theme labelling, and the lexicon fallback is always available.

If you need a feature on this list, do not work around it inside this actor. Chain the right tool instead.

When NOT to use this actor (job → better tool)

Job you need done	Better tool
Send a cold-email sequence to creators	Outreach / Salesloft / Apollo / Smartlead / Lemlist
Manage an active influencer campaign (briefs, deliverables, payments)	GRIN / Aspire / CreatorIQ / Upfluence
Find sponsor-ready creators with verified business emails	ryanclinton/youtube-sponsorship-intelligence
Get raw YouTube substrate at the lowest possible price	apidojo/youtube-scraper (~$0.10 per 1,000)
Use the brand-recognized incumbent for raw bulk extraction	streamers/youtube-scraper
Real-time live-stream monitoring	A dedicated live-stream monitoring platform
Rank a creator list by what's popping right now	This actor
Find emerging creators in a niche via search reranking	This actor
Monitor a competitor channel set with daily deltas	This actor

Capability matrix vs the top 10 YouTube actors

Capabilities as of May 2026 and may change.

Capability	This actor	streamers/youtube-scraper	apidojo/youtube-scraper	streamers/youtube-channel-scraper	streamers/youtube-shorts-scraper	streamers/youtube-comments-scraper	Generic search-mode YouTube actors
Unified channel + video + playlist + search + shorts in one actor	Yes	No	No	No	No	No	No
Comments included	Yes	Separate actor	Sometimes	No	No	Yes	Rare
Transcripts	Yes	Yes	Sometimes	Yes	No	No	Rare
Search reranking (breakoutPotential, momentum)	Yes	No	No	No	No	No	No
Breakout / outlier detection per video	Yes	No	No	No	No	No	No
Performance baseline (p50 / p90 / p99 views)	Yes	No	No	Rare	No	No	No
Growth velocity (trend rising / stable / declining)	Yes	No	No	Rare	No	No	No
Channel momentum scoring with drivers	Yes	No	No	No	No	No	No
Topic pillars (TF-IDF clusters)	Yes	No	No	No	No	No	No
Audience language mix from comments	Yes	No	No	No	No	No	No
Comment themes (not just sentiment)	Yes	No	No	No	No	Rare	No
Link classification (affiliate / sponsor / self / social)	Yes	No	No	No	No	No	No
Video freshness band (decay buckets)	Yes	No	No	No	No	No	No
Cohort ranking (rank within input set)	Yes	No	No	No	No	No	No
Watchlist mode (cross-run state)	Yes	No	No	No	No	No	No
Temporal signals (delta vs prior run)	Yes	No	No	No	No	No	No
Delta intelligence (only what changed)	Yes	No	No	No	No	No	No
Summary briefing record per run	Yes	No	No	No	No	No	No

The contrast row that matters: a flat scraper gives you data. This actor gives you a routable decision (priority, whyNow, profile, recommendedAction) that drops into a dashboard or a sequencing tool without further processing.

Input parameters

Parameter	Type	Default	Description
mode	string (enum)	search	One of channels, videos, playlists, search, shorts. Mutually exclusive.
channels	string[]	[]	Handles (@mkbhd), UCx... IDs, or full URLs. Used in channels and shorts modes.
maxRecentVideosPerChannel	integer	50	1-500. Cap on recent uploads per channel.
includeShortsInVideoList	boolean	true	Pull shorts alongside long-form when fetching channel videos.
includeTranscripts	boolean	false	Fetch transcripts for sampled videos. Slows the run, enables richer topic detection.
videos	string[]	[]	Video URLs or 11-character IDs. Used in videos mode.
enrichChannel	boolean	true	In videos mode, also fetch parent channel for baseline scoring.
includeTranscript	boolean	false	Fetch transcript per video (videos mode).
includeCommentsSample	boolean	false	Sample top-level comments. Required for commentSynthesis themes.
commentsSamplePerVideo	integer	0	0-500. 30+ enables theme synthesis.
playlists	string[]	[]	Playlist URLs. Used in playlists mode.
maxVideosPerPlaylist	integer	100	1-1000. Cap on videos pulled per playlist.
searchQuery	string	""	Search query string. Used in search mode. Prefill: AI coding tutorials.
videoType	string (enum)	any	Filter search to any, video, short, live, playlist, channel.
country	string	US	ISO 3166-1 alpha-2 country code for search geo.
language	string	en	ISO 639-1 language code for search.
dateRange	string (enum)	any	any, this_hour, today, this_week, this_month, this_year.
uploadOrder	string (enum)	relevance	YouTube native ordering before reranking.
rankBy	string (enum)	breakoutPotential	Reranking axis. relevance, breakoutPotential, momentum, recency, engagement.
maxResults	integer	50	1-500. Search result cap.
maxShortsPerChannel	integer	50	1-500. Shorts mode.
oldestPostDate	string	""	ISO 8601 date. Shorts mode lower bound.
outputProfile	string (enum)	signals	signals, substrate, research, llm, minimal, compat.
watchlistName	string	""	Enables delta intelligence when set. Named KV store.
enableCommentThemes	boolean	true	Toggle groq LLM theme naming. Lexicon fallback always available.
groqApiKey	string (secret)	""	Optional groq API key. Without it, theme labels use lexicon clustering.
rateLimitPerSecond	integer	2	1-10. Conservative default to avoid bot-blocks.
circuitBreakerThreshold	integer	3	1-20. Halt after N consecutive bot-blocks.
runtimeBudgetSeconds	integer	3000	Soft runtime cap. The actor exits cleanly with truncated: true when exceeded, instead of hard-killed mid-emit. Auto-clamped against APIFY_TIMEOUT_AT minus a maxResults-scaled emit reserve so the dataset always finishes pushing.
proxyConfiguration	object	Apify RESIDENTIAL, US	Apify proxy config. Residential required.

Run via the API

Python

from apify_client import ApifyClient

client = ApifyClient("YOUR_API_TOKEN")

run = client.actor("ryanclinton/youtube-scraper").call(run_input={
    "mode": "search",
    "searchQuery": "AI coding tutorials",
    "rankBy": "breakoutPotential",
    "maxResults": 25
})

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    if item.get("recordType") in ("searchResult", "video", "channel"):
        priority = item.get("attentionPriority")
        profile = item.get("signalProfile")
        action = item.get("recommendedAction")
        print(f"{item.get('channelHandle')} | {priority} | {profile} | {action}")

JavaScript

import { ApifyClient } from "apify-client";

const client = new ApifyClient({ token: "YOUR_API_TOKEN" });

const run = await client.actor("ryanclinton/youtube-scraper").call({
    mode: "search",
    searchQuery: "AI coding tutorials",
    rankBy: "breakoutPotential",
    maxResults: 25,
});

const { items } = await client.dataset(run.defaultDatasetId).listItems();
for (const item of items) {
    if (["searchResult", "video", "channel"].includes(item.recordType)) {
        console.log(`${item.channelHandle} | ${item.attentionPriority} | ${item.signalProfile} | ${item.recommendedAction}`);
    }
}

cURL

curl -X POST "https://api.apify.com/v2/acts/ryanclinton~youtube-scraper/runs?token=YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "mode": "search",
    "searchQuery": "AI coding tutorials",
    "rankBy": "breakoutPotential",
    "maxResults": 25
  }'

curl "https://api.apify.com/v2/datasets/DATASET_ID/items?token=YOUR_API_TOKEN&format=json"

Use in Dify, LangChain, and agent frameworks

Set outputProfile: "llm" when the consumer is an LLM agent. It adds a compact natural-language summary per record so the agent gets a flat branching surface without traversing nested objects. The deterministic enum tokens below give the agent stable branching keys.

Dify Custom Tool wiring:

Field	Value
Method	POST
URL	https://api.apify.com/v2/acts/ryanclinton~youtube-scraper/run-sync-get-dataset-items
Authorization	Bearer YOUR_APIFY_API_TOKEN
Headers	Content-Type: application/json
Request body	JSON matching the input schema
Response	JSON array of records (channel + video + summary + error)

Branching keys agents can route on:

• record.attentionPriority in ["high", "medium", "low", "none"]
• record.signalProfile in the 8-label vocabulary
• record.recordType in ["channel", "video", "playlist", "short", "searchResult", "summary", "error"]
• record.failureType in the 7-token failure vocabulary
• record.changeFlags[] (watchlist mode) for diff-aware downstream consumers

Stable enum tokens (next section) are the agent's contract.

Stable enum tokens (additive-only pledge)

These string tokens are stable across releases. New values may be added; existing values will not be renamed or removed without a major-version bump and migration note.

• signalEvent.type (9 values): breakout_video, channel_acceleration, channel_deceleration, outperformer_streak, topic_pivot, audience_expansion, cadence_acceleration, cadence_collapse, algorithmic_window_alert
• signalProfile (8 values): emerging, breakout, stable-authority, viral-fragile, high-engagement-low-scale, decelerating, dormant, unclassified
• attentionPriority (4 values): high, medium, low, none
• decayStatus (4 values): fresh, active, fading, expired
• rankBy (5 values): relevance, breakoutPotential, momentum, recency, engagement
• outputProfile (6 values): signals, substrate, research, llm, minimal, compat
• recordType (7 values): channel, video, playlist, short, searchResult, summary, error
• failureType (7 values + null): auth, rate_limit, not_found, schema_mismatch, bot_blocked, transcript_unavailable, unknown, null
• watchlistState (4 values): no-action, monitor, attention, high-priority
• momentumBand (5 values): accelerating, rising, stable, fading, dormant
• commentSynthesisStatus (3 values): full, lexicon_fallback, disabled
• momentumChange.direction (4 values): accelerating, decelerating, unchanged, first-sight

Version constants pinned on every record + summary (so your automation can pin against an exact ruleset):

• signalDetectionVersion: "1.0": event detectors + thresholds + decay table
• signalProfileVersion: "1.0": vocabulary + classification rules
• attentionRouterVersion: "1.0": priority math + recommendedAction template list
• watchlistComputationVersion: "1.0": baseline + delta windows + state transitions
• topicMethodVersion: "tfidf-v1": topic pillar method
• commentSynthesisVersion: "lexicon-themes-v1": theme cluster definitions

Bumping any constant requires a new actor version with a migration note. v1.0 freezes all six.

First run tips

• Start with the prefilled search demo. Click Start without editing. You see the 10-second outcome described at the top of this README. Then iterate.
• For channel runs, start with 5-10 handles to verify the output shape and tier distribution before scaling.
• maxRecentVideosPerChannel: 50 is the right default. Below 10, signal detection has too little baseline. Above 100, runs slow without proportional signal gain.
• Keep includeCommentsSample: false unless you specifically need commentSynthesis themes or audience_expansion detection. Comments sampling roughly triples per-channel request load.
• Default proxy is residential US. Do not switch to datacenter. YouTube blocks datacenter IPs. Use a different apifyProxyCountry if US rotation is geo-blocked.
• For watchlist mode, the first run is a baseline. newSignalsSinceLastRun, momentumChange.delta, watchlistState.previous populate from run 2 onward. Schedule daily from the Apify Schedules tab.
• Set a per-run spending limit on the actor page before scheduling. The free tier $5/month covers ~2,500 records.

How this actor works

Mental model: Input → fetch substrate → engine intelligence (growth velocity + performance baseline + topic pillars + audience signal) → signal event detection → signal profile classification → attention routing → (watchlist deltas if applicable) → output profile shaping → emit.

Phase 1: fetch substrate

Channel page, recent videos, optional transcripts, optional comment samples. Innertube (youtubei.js) is the primary fetch path, Cheerio HTML scrape is the fallback. Residential proxy throughout, US rotation by default. The circuit-breaker halts the run after N consecutive bot-blocked responses (circuitBreakerThreshold, default 3).

Phase 2: engine intelligence

Per channel, compute growth velocity (upload cadence trend), performance baseline (p50 / p90 / p99 views), topic pillars (TF-IDF clusters when >= 10 videos in 90-day baseline window, otherwise null), audience signal (comment language mix, peak upload day, peak upload hour UTC).

Phase 3: signal event detection

Deterministic, threshold-based. All thresholds pinned to signalDetectionVersion: "1.0":

• breakout_video: viewsZScore >= 2 AND ageHours <= 168 AND engagementZScore >= 1
• channel_acceleration: median views last 60d >= 1.2× median prior 60d AND cadence stable
• outperformer_streak: 3+ consecutive videos with viewsZScore >= 1
• topic_pivot: any topic share delta >= 0.15 between prior 60d and current 60d
• audience_expansion: comment language mix gains a new language at >= 5% share
• cadence_acceleration: current 30d cadence >= 1.5× prior 30d cadence
• cadence_collapse: current 30d cadence <= 0.5× prior 30d cadence
• algorithmic_window_alert: breakout_video present AND ageHours <= 336

Phase 4: decay + signal profile

Each event's daysSincePeak (or daysSinceFirstDetected for sustained signals) buckets against the decay table to produce decayStatus. Expired events move from signalEvents[] to expiredSignals[]. The channel's signalProfile is then computed from the active signal set and substrate (subs, engagement, cadence).

Phase 5: attention routing

attentionPriority is computed from the active signal set + signal profile + freshness. whyNow[] is populated from up to 3 explanations. recommendedAction is selected from a finite 12-template list, pinned to attentionRouterVersion: "1.0". When priority is low or none, deprioritizationReasons[] populates instead.

Phase 6: watchlist deltas (when watchlistName is set)

Read the named KV store for prior state. Diff against current. Emit temporalSignals / watchlistState / momentumChange / changeFlags. Write new state back. Across-run state lives in the KV store, not the dataset, so replay and rollback work cleanly.

Phase 7: output profile shaping + emit

The same internal record is shaped per outputProfile. compat strips the signal layer for streamers migration. signals (default) keeps everything. research adds full evidence trails + nested version envelopes. llm adds a compact natural-language summary.

Limitations

• Transcript availability is YouTube-controlled. Some videos do not expose transcripts publicly. Sponsorship-timestamp detection (in the sister actor) and richer topic detection here both rely on transcripts and quietly skip when absent. failureType: "transcript_unavailable" is logged on the affected videos.
• TF-IDF topic clustering is noisy on small samples. topicPillars is null when a channel has fewer than 10 videos in the 90-day baseline window. v1.1 candidate: opt-in embedding-based clustering for sharper labels on small samples.
• Sentiment is lexicon-based. vader-sentiment is approximate. Sharper sentiment (per-comment context, sarcasm handling, language-aware scoring) is a v1.1+ candidate. Themes from commentSynthesis are the better surface for decisions.
• YouTube anti-bot escalation is possible. Default config (residential US, 2 req/sec, 3-strike circuit breaker) is conservative. If you see persistent bot_blocked records, drop rateLimitPerSecond to 1 and try a different apifyProxyCountry.
• Embedding-based topic clustering is opt-in only in v1.0. TF-IDF is the default. Embedding-based labels arrive with a topicMethodVersion bump in v1.1.
• Signal events describe, they do not predict. A breakout_video event with signalStrength: 0.91 describes a video that broke 12× the channel p90 baseline within 48h. It does NOT predict that the video will keep growing. The evidence is attached so you can audit.

Integrations

• Zapier: trigger downstream workflows on each attentionPriority: "high" record.
• Make: branch on signalProfile and attentionPriority.
• Google Sheets: export the Attention Queue view directly.
• Apify Schedules: required for watchlist mode.
• Apify API: read the dataset, trigger runs, manage the watchlist KV store.
• Webhooks: fire on run finish to push the Attention Queue into Slack or your CRM.

Troubleshooting

Run halted with failureType: "bot_blocked" errors. YouTube anti-bot escalated. Try a different apifyProxyCountry, lower rateLimitPerSecond to 1, raise circuitBreakerThreshold slightly, or wait and re-run.

failureType: "transcript_unavailable" on many videos. Not all videos expose transcripts. Set includeTranscripts: false (channels mode) or includeTranscript: false (videos mode) if transcripts are not required. Topic detection still works from titles, descriptions, hashtags.

commentSynthesis.commentSynthesisStatus: "lexicon_fallback" on most records. Either no groqApiKey was supplied or the LLM call failed. Theme labels still populate from lexicon clustering; they are just less polished. Supply a groq API key and set enableCommentThemes: true for full status.

Watchlist mode not producing deltas. The first run is a baseline. newSignalsSinceLastRun, momentumChange.delta, and watchlistState.previous populate from run 2 onwards. Confirm watchlistName is identical between runs (case-sensitive).

topicPillars is null. Channel has fewer than 10 videos in the 90-day baseline window. Topic detection requires a baseline sample size to avoid noisy labels. Pull more recent videos with maxRecentVideosPerChannel: 100 or accept null.

Run hit the runtime budget. truncated: true and truncatedReason populate on the summary record. The actor exited cleanly with partial results instead of getting hard-killed mid-emit. Re-run with a tighter input scope (fewer channels, lower maxRecentVideosPerChannel, smaller maxResults).

FAQ

Can this detect growing YouTube channels? Yes. The channel_acceleration signal event fires on every channel whose median views rose 20%+ over the last 60 days vs the prior 60 days with stable upload cadence. The signalProfile: "emerging" label fires on small-to-mid creators with rising velocity. Set mode: "search" with rankBy: "growthVelocity" to surface growing channels from a search query rather than a known list.

Can this identify breakout YouTube creators? Yes. The breakout_video signal event flags single videos popping vs the channel's own baseline (z-score >=2, recently uploaded). The signalProfile: "breakout" label catches creators in the middle of a breakout run. The signalProfile: "viral-fragile" label catches single-spike accounts with no underlying acceleration. Combined with rankBy: "breakoutPotential" in search mode, the actor doubles as a breakout YouTube creator detector at scale.

Can this monitor competitor YouTube channels automatically? Yes. Watchlist mode is built for exactly this. Set watchlistName to a stable string, schedule the actor on Apify Schedules to run daily or weekly, and every run after the first emits newSignalsSinceLastRun, momentumChange, watchlistState.previous vs current, and changeFlags per channel. The buyer reads the Watchlist Delta dataset view and sees only what changed.

Can this rank YouTube creators by momentum? Yes. Every channel record carries momentumScore (0..100), momentumBand (accelerating / stable / decelerating / dormant), and momentumDrivers[] explaining which signals fired. Multi-channel runs (>=2 entities) auto-enable cohort ranking. Every record gets rankInCohort so the top-momentum creator in the run is rank #1.

Is this a YouTube analytics API or a scraper? Both, by design. The substrate layer is a YouTube scraper compatible with streamers/youtube-scraper's output shape (drop-in migration). The intelligence layer is a YouTube creator analytics API exposing breakout detection, momentum scoring, attention routing, watchlist deltas, and reranked discovery. Trigger the actor via Apify's API, get back creator intelligence as structured JSON, route on the stable enum tokens documented above.

Can this replace spreadsheet-based creator tracking? That's the entire product thesis. The real competitor isn't another YouTube scraper, it's the spreadsheet workflow agencies and growth teams currently use to triage creators. The spreadsheetCollapse block in the summary record reports a heuristic estimate of how many minutes of manual review the run replaced (default constant: 3 minutes per row of triage).

How do agencies use this for creator scouting? Three common patterns. (A) Qualify a list of channel handles: mode: channels, paste 50 handles, open the Attention Queue, contact the top 5. (B) Find emerging creators in a niche: mode: search, search query like "AI coding tutorials" or "skincare reviews", rankBy: "breakoutPotential", top 5 fastest-growing creators surface. (C) Monitor a competitor roster: mode: channels, set watchlistName: "agency-roster-q2", schedule daily, the Watchlist Delta view returns only what changed (new breakouts, accelerating creators, dormant ones waking up).

Does this work for YouTube competitor monitoring on a schedule? Yes. Watchlist mode + Apify Schedules + Slack webhook = automated competitor monitoring. Every scheduled run emits the Watchlist Delta view with only changed records. Set the webhook to fire on run finish, route on changeFlags (e.g. NEW_BREAKOUT, MOMENTUM_UP, TOPIC_PIVOT). The buyer's team sees only the creators whose state changed this week.

How is this different from streamers/youtube-scraper? Streamers ships rows. This actor ships rows + a signal layer (attention priority, whyNow, signal profile, recommended action) + a default Attention Queue view that orders by what to look at first + watchlist mode for cross-run deltas. Same substrate fields, drop-in migration via outputProfile: "compat". Less than half the price ($2 vs $5 per 1,000 records).

Is breakout detection a prediction? No. A breakout_video event with signalStrength: 0.91 describes a video that broke 12× the channel p90 baseline within 48h of upload. It does not promise the video will keep growing. The evidence (z-scores, baseline, age) is attached on every event so you can audit. signalStrength is a heuristic compound, not a calibrated probability.

What is signalStrength versus confidence? signalStrength is a heuristic compound (z-scores, sample size, recency) rolled to 0-1 for sortability. It is honest about being a heuristic. confidence reads as a calibrated probability the actor does not actually produce. The product uses signalStrength throughout for that reason.

How does watchlist mode billing work? Watchlist mode adds no extra event. Same $0.002 per record on monitoring runs as on one-shot runs. The delta intelligence is included. This is intentional: the moat is monitoring, and pricing monitoring extra would suppress the behaviour the product depends on.

Can this replace a YouTube channel analytics tool? Partly. This actor pulls publicly available channel and video data and layers on analytics signals: momentum score, breakout events, topic pillars, audience language distribution, engagement rates vs baseline. It cannot access YouTube Studio private analytics (impressions, CTR, revenue). For competitive YouTube channel analytics on public data, it covers more signal surface than any other actor on the Store.

Can I migrate from streamers/youtube-scraper without rewriting my code? Yes. Set outputProfile: "compat" and the output fields match streamers exactly (title, viewCount, likes, channelName, channelUrl, duration, date, url, hashtags, description). Downstream code reading item.title works unchanged. Open the Streamers Compat dataset view to verify parity.

How is this different from a generic YouTube search scraper? Generic search scrapers return YouTube's native relevance order. This actor reranks by breakoutPotential, momentum, engagement, or recency and surfaces both the native rank and the reranked rank side by side. The reranker uses the signal layer (breakout events, momentum band, signal profile), so search results are ordered by who is actually winning right now, not by YouTube's opaque relevance score.

What does recommendedAction mean? Can I have it suggest a sponsor pitch? No. recommendedAction is locked to prioritisation language (Review, Watch, Track, Re-check, Compare, Investigate) + timing. It is never a business strategy. Sponsor pitch / partner / acquire verbs are explicitly forbidden. Sponsor-readiness lives in the sister actor ryanclinton/youtube-sponsorship-intelligence.

Does this scrape comments? Yes, when includeCommentsSample: true and commentsSamplePerVideo >= 30. Comments are sampled, clustered into themes via cheap embedding similarity, and named (via groq Llama 3.1 8B when an API key is supplied, otherwise lexicon clustering). commentSynthesisStatus tells you which path produced the labels.

Can I use my own scoring weights? v1.0 ships fixed weights pinned to signalDetectionVersion and attentionRouterVersion. The trade-off is a stable contract for downstream automation. Custom weights are a v1.1+ candidate via a scorecardTemplate mechanism, with the version constants bumped accordingly.

Is it legal to scrape YouTube data? This actor reads only publicly available channel and video information. It does not bypass authentication, CAPTCHAs, or access private analytics. Legality depends on jurisdiction and intended use; consult legal counsel for your specific case. See Apify's guide on web scraping legality.

Can I push results straight into HubSpot or a Slack channel? Yes. Pair with HubSpot Lead Pusher via webhook, or wire an Apify webhook into a Slack incoming webhook. Branch on attentionPriority: "high" to keep the inbound volume manageable.

Responsible use

• This actor reads publicly available YouTube channel and video information. It does not bypass authentication, CAPTCHAs, or access restricted content.
• Users are responsible for ensuring their use complies with applicable laws and platform terms, including data protection regulations (GDPR, CCPA) and the legitimate-interest tests they imply when processing creator identifying information.
• Creator names, handles, and channel descriptions are personal data in some jurisdictions. Do not use extracted data for spam, harassment, or unauthorized purposes.
• For guidance on web scraping legality, see Apify's guide.

Recent updates

• v1.0 (May 2026): Initial release. Five input modes (channels / videos / playlists / search / shorts) in one actor. Signal layer with 9 typed event types, 8 signal profiles, attention routing, decay tracking, watchlist mode with cross-run deltas, search reranking with breakoutPotential, comment theme synthesis via groq Llama with lexicon fallback, summary record with dailyBriefing + executiveHighlights + spreadsheetCollapse. All version constants pinned (signalDetectionVersion, signalProfileVersion, attentionRouterVersion, watchlistComputationVersion, topicMethodVersion, commentSynthesisVersion).

Help us improve

If you encounter issues, you can help debug faster by enabling run sharing in your Apify account:

1. Go to Account Settings > Privacy
2. Enable Share runs with public Actor creators

This lets the developer see your run details when something goes wrong. Your data is only visible to the actor developer, not publicly.

Support

Found a bug or have a feature request? Open an issue in the Issues tab on this actor's page. For custom solutions or enterprise integrations, reach out through the Apify platform.