YouTube Channel Playbook: Titles, Thumbnails & Video Ideas

You already scrape YouTube channels. This Actor tells you which titles, thumbnails, topics, formats and upload days beat each channel's normal reach, where your channel has gaps next to the others, and writes original title + thumbnail + hook ideas for your channel.

Give it the dataset of a YouTube Scraper or YouTube Channel Scraper run on one or more public channels: yours and the ones you compete with. For every video it:

1. compares its views with the median of the same channel's videos of the same format and age, so a 2-day-old Short is not measured against a 2-year-old tutorial, and a 10K channel compares fairly with a 1M channel,
2. labels the title: number, question, how-to wording, curiosity gap, negative framing, length, ALL-CAPS, talks to "you", and its angle (tutorial, listicle, review, story, explainer, challenge…),
3. looks at the thumbnail: face or no face, facial expression, text on it, contrast, main subject, dominant colour, arrows/circles,
4. reads the first 60 seconds of subtitles when your dataset has them, to label the opening hook (videos are never downloaded),
5. groups all videos into topic clusters across channels.

Then, per channel, it shows what over-performs and under-performs, Shorts vs long-form, upload cadence and best days, and gaps: patterns that win on the other channels but that this channel rarely uses (with one channel: winners it under-uses). Finally it writes N video concepts for your channel brief.

Concepts take the pattern, never the words. The writer only gets labels, statistics and generic title templates (like "[EXPERT] reacts to [N] [TOPIC] mistakes"), not real titles. Every concept is checked against all input titles, thumbnail text and subtitles: a concept that shares any 5-word sequence, or that mostly restates an existing title, is dropped and not charged.

How to use

1. Run the YouTube Scraper on the channels you want to learn from (paste channel URLs in Direct URLs; set Maximum videos and Maximum Shorts). 30–100 videos per channel give useful medians. Optional: set Transcription & subtitles to Only download subtitles to also get the opening-hook analysis.
2. Open this Actor, pick that run's dataset in YouTube scraper dataset, type your channel's @handle in Your channel (if it is in the dataset), paste your channel brief and choose how many concepts you want.
3. Start. The Actor reads the dataset, checks the price against your run's maximum cost, and only then begins.

No dataset? Paste items in Items (YouTube Scraper JSON format) or link a JSON or CSV export in Items URL. The prefilled input is a 9-video synthetic demo of two fictional cooking channels, so the form runs as-is.

Input

Field	Required	Notes
datasetId	one of these three	Dataset of a YouTube Scraper or YouTube Channel Scraper run (picker in Console; via API pass the run's defaultDatasetId). Read-only access
items		Items as JSON: title, url, thumbnailUrl, viewCount, likes, commentsCount, date, duration, type (video / shorts / stream), channelName, channelUsername, optional subtitles
itemsUrl		Link to a JSON array or CSV export with the same fields
myChannel	no	Your channel's @handle, name or URL. Turns on the "your channel vs the others" view and aims the concepts at it
maxVideos	no	Default 200, max 1,000. Taken evenly across channels
channelBrief	when concepts > 0	Niche, audience, tone, what you can film, goal, what to avoid (up to 4,000 characters)
conceptsCount	no	0–30, default 10. 0 = analysis only
language	no	Language of concepts and takeaways. Default English

{
  "datasetId": "<your YouTube Scraper run's dataset ID>",
  "myChannel": "@yourchannel",
  "maxVideos": 200,
  "channelBrief": "Weeknight Meals: 30-minute one-pan dinners for busy people. Host cooks on camera at home. Friendly, practical, no chef jargon. Avoid clickbait we cannot deliver.",
  "conceptsCount": 10
}

Output

Dataset: one row per video (from the demo run):

{
  "status": "analyzed",
  "channel": "Demo Weeknight Meals",
  "isMyChannel": true,
  "title": "Why Your Stir Fry Is Soggy",
  "format": "long-form",
  "publishedAt": "2026-10-08T17:00:00.000Z",
  "views": 27500,
  "ageBucket": "7-30d",
  "uploadDay": "Thu",
  "baseline": "long-form, 7-30d old",
  "baselineMedianViews": 14000,
  "performanceIndex": 1.96,
  "performance": "over",
  "topicCluster": "stir fry cooking problem",
  "titleAngle": "explainer-why",
  "titleTemplate": "Why Your [FOOD] Is [NEGATIVE_OUTCOME]",
  "isQuestion": true,
  "curiosityGap": true,
  "negativeFraming": true,
  "titleLength": "≤30 chars",
  "thumbFace": "none",
  "thumbHasText": true,
  "thumbText": "SOGGY?",
  "thumbContrast": "high",
  "thumbFocus": "single-object-product",
  "thumbColour": "orange",
  "hookType": "no-transcript",
  "priceUsd": 0.004
}

With subtitles in the dataset, transcriptFirst60s, hookType (question, bold-claim, preview-payoff, story-open, problem-pain, surprising-fact, direct-address, cold-open-action…) and hookPattern are filled too.

Key-value store

Record	Content
playbook.md	The playbook: key takeaways, a channel table, per channel Shorts vs long-form, cadence and best days, over- and under-performing patterns, gaps, top videos; topic clusters; your concepts; pattern tables for every label
concepts.json	Takeaways and concepts as JSON, each with the pattern evidence behind it
channels.json	All channel summaries and pattern statistics as JSON
OUTPUT	Counts, amount charged and links to the files above

A concept (from the demo run):

{
  "id": "C3",
  "title": "Meal Prep 3 Weeknight Dinners in One Go",
  "format": "long-form",
  "topicCluster": "meal prep for the week",
  "titleAngle": "how-to-tutorial",
  "titlePatterns": ["number", "curiosity-gap"],
  "thumbnail": {
    "description": "Three dinner components in containers beside a single pan on a small kitchen counter. Bright, high-contrast text labels the batch-cooking payoff.",
    "text": "ONE COOK, THREE DINNERS",
    "face": "none",
    "focus": "collage",
    "colour": "red",
    "contrast": "high"
  },
  "hookLine": "I'll show you how to prep three weeknight dinners in one cooking session, using supermarket ingredients and a single pan.",
  "whyItShouldWork": "A concrete number and useful meal-prep promise offer a clear plan, while the thumbnail makes the one-session payoff easy to understand.",
  "evidence": { "thumbText": { "value": "text on thumbnail", "videos": 3, "medianIndex": 1.22, "scope": "Demo Weeknight Meals" } }
}

Pricing

Pay per event. Failed videos and dropped concepts are free.

Event	Price
Video analysed	$0.004
Concept written (title + thumbnail + hook line)	$0.01

Examples:

Run	Price
3 channels × 50 videos, 10 concepts	150 × $0.004 + 10 × $0.01 = $0.70
200 videos, 10 concepts (defaults)	$0.90
100 videos, analysis only	$0.40

The Actor computes the maximum price before it starts and stops, free of charge, if your run's maximum cost is lower.

Available on paid Apify plans.

Limits and good to know

• On the free Apify plan the Actor returns a short preview with an example result and prices instead of running.
• Up to 1,000 videos and 30 concepts per run. Long datasets are capped at maxVideos, taken evenly across channels.
• Numbers (medians, indexes, lift, cadence) are computed in code from the scraper's public view, like and comment counts at scrape time, not by the AI. With few videos per pattern the results are hints: the playbook shows the number of videos behind every figure and a confidence level.
• Cadence is measured per format, because scrapers cap Videos and Shorts separately. Upload days and hours are in UTC. Relative dates from the YouTube Channel Scraper ("3 weeks ago") are approximate, so those videos are left out of the upload-day analysis; that scraper also has no likes, so engagement is empty.
• The opening hook is read only from subtitles that are already in your dataset (first 60 seconds). No video or audio is downloaded.
• Thumbnail and title labels come from an AI model and can be wrong on unusual images; thumbnail text is read from the image and can contain small errors.
• Concepts are ideas to test, not guarantees. They only claim what your brief supports, but check them before you film.
• Only public channel-level and content-level data is used and output: no comments, commenter names or channel contact details.
• Use it on public channels you are allowed to analyse and keep to YouTube's terms. Thumbnails are viewed in memory and never stored or re-hosted.