youtube-script

name

youtube-script

description

Draft spoken video scripts for YouTube longs (8-12 min), YouTube shorts (60s vertical), or talking-head clips. Spoken cadence (read aloud, ear-tested · not eye-read), beat-by-beat structure with timing cues and B-roll suggestions, hook-in-the-first-five-seconds discipline, retention pulls every 30s. Voice anchored to your LinkedIn corpus but rewritten for spoken delivery (shorter sentences, no em dashes anywhere, no list-stacking). Hands off to /video-use for editing the recording, or to /launch-video if synthetic motion graphics. Logs the row in the Notion Content DB. Use when asked to "write a youtube script", "draft a video script", "shorts script for X", "talking-head script", "scripted video about Y", or "/youtube-script". For written long-form (newsletter / blog), use /long-form instead.

youtube-script

Spoken long-form. Talking-head video and shorts in your voice · written for the ear, not the eye. Different cadence than /long-form even though the topic might be the same.

When to use

Trigger on:

• "Write a YouTube script for X"
• "Draft a shorts script"
• "Talking-head script about Y"
• "Video script · 8 minutes"
• "Voiceover for the launch video" (talking copy, not narration beats)
• User explicitly types /youtube-script

Skip for:

• Written newsletter / blog → /long-form
• LinkedIn / X posts → /linkedin-copywriter · /x-copywriter
• Synthetic motion-graphic narration → /launch-video (Remotion has its own beat-anchored narration system)
• Editing existing recorded footage → /video-use

Inputs

The brief should include:

• Type: long (8-12 min YouTube) · short (60s vertical · YT Shorts / IG Reels / TikTok) · clip (90s-3min talking-head clip)
• Topic + angle
• Target duration (default: 8 min long, 60s short, 2 min clip)
• Audience: cold YouTube viewer · email list opt-in · existing followers
• Optional source: Notion row URL if invoked from /repurpose
• Optional CTA: subscribe · grab a lead magnet · book a call · reply on LinkedIn · default by type below

If anything is missing, ask in one batch, then draft.

Voice anchor (spoken)

Same author voice as /linkedin-copywriter, but rewritten for the ear. Read ../linkedin-copywriter/corpus.md before drafting, then translate it for spoken delivery:

Spoken differences:

• Shorter sentences. Average ~12 words. The reader's eye can hold a long sentence; the listener's ear cannot.
• No em dashes anywhere. (Hard rule for everything <YOUR_BRAND>, but especially obvious here · em dashes don't have an audible equivalent. Just use a period.)
• Avoid stacked lists. "First, second, third" works on the page; three items spoken in a row sound like a robot. Break with beats.
• Contractions are mandatory. "It's", "you're", "we'll" · not "it is", "you are", "we will". Spoken voice contracts.
• Numbers are spoken. "Forty-seven" not "47" if you want it to land. "47" is fine on the teleprompter and reads aloud as "forty-seven". Just be aware.
• Read every line aloud before declaring it done. If your tongue trips, the camera will catch the trip.

Structure by type

YouTube long (8-12 min)

00:00-00:05   HOOK            One sentence. The friction. The promise.
00:05-00:25   COLD OPEN       Set the stakes. What you'll prove.
00:25-00:50   INTRO + IDENT   Brief who-you-are. Subscribe pulse 1.
00:50-...     BEAT 1          First insight. Concrete example.
              [retention pull every 90-120s · "but here's the weird part" / 
               "and that's where it gets interesting"]
              BEAT 2          Second insight. Counter-take or extension.
              BEAT 3          Third. Proof / data / customer example.
              [BEAT 4-5 if 12 min]
~12:00        PAYOFF          What changes if you act on this.
              CTA             Subscribe + lead magnet + (optional) book a call.

5-7 beats total. Each beat ~60-120 seconds spoken. Word count target: ~150 words/min spoken pace · 8 min = ~1200 words · 12 min = ~1800 words.

YouTube short (60s · vertical)

00:00-00:02   HOOK            Two seconds. ONE sentence. Stakes-loaded.
00:02-00:08   PROBLEM         What's broken. (One sentence.)
00:08-00:35   PAYOFF          The insight / framework / number.
                              No more than 2 supporting points.
00:35-00:50   PROOF           One specific example or data point.
00:50-00:60   CTA             "Save this. / Follow for more. / Comment X."

~150 words total. Every word earns its place. No "subscribe" intros. No "today we're talking about". Just the hook.

Talking-head clip (90s-3min)

00:00-00:05   HOOK            One sentence.
00:05-...     ARG             One argument, fully developed.
              [no retention pulls · clip is short enough]
              CTA             Single line. Often "let me know what you think".

3-5 beats. Used for clipped LinkedIn / X video, weekly customer shoutouts, etc.

Workflow

1. Brief

Inputs above. If invoked from /repurpose, treat the master Notion row's content as raw material to translate to spoken cadence.

2. Beat outline

Output the beats with timestamps and one-line descriptions before drafting the script. Show to the user, wait for green light. As with /long-form, don't draft the full script until the outline is approved.

3. Draft the script

Format:

# <Title>

**Type**: long / short / clip
**Target duration**: <m:ss>
**Hook**: <one sentence · the cold open>

---

[00:00-00:05] HOOK
<spoken line>

[00:05-00:25] COLD OPEN
<spoken paragraph · 50-70 words>
[B-ROLL: <visual cue if applicable>]

[00:25-00:50] INTRO
<spoken paragraph>
[B-ROLL: <cue>]

[00:50-02:00] BEAT 1 — <one-line description>
<spoken paragraph · 100-150 words>
[B-ROLL: <cue>]

... etc

[12:00] CTA
<spoken line · single sentence>

Conventions:

• B-roll cues in [B-ROLL: ...] brackets after the line they reference. Cues describe what should be on screen during that voiceover (screen recording / cut-in / chart / quote card / etc.). Don't B-roll every paragraph · only when the visual genuinely adds.
• Retention pulls are sentences like "but here's where it gets weird", "the thing nobody tells you is", "and this is where most people get it wrong". Use one every 90-120s in longs. Drop them for shorts and clips.
• Subscribe pulses in longs only. Once at 0:25-0:50 ("if you're new here, hit subscribe · I drop one of these every Friday"), once in the CTA. No mid-video subscribe interruptions.
• No filler intros. "Today we're going to talk about..." is dead on arrival. Hit the hook, set the stakes, move.

4. Read aloud check

For every script before declaring done:

1. Read it aloud (or have the user do it). Mark any line that trips your tongue.
2. Time it. If the read is more than ~10% over target duration, cut not extend.
3. Check the hook works in the first two seconds for shorts and first five seconds for longs. If you can't say it in that window, the hook is wrong.

5. Slop scrub (mandatory · brand checklist)

Same 4 copy checks as every text-producing skill:

1. Em-dash scan · zero — (especially important here · spoken cadence doesn't tolerate them)
2. Banned-words scan · the full brand-doc list
3. Negation-pivot scan · "not just X, but Y" patterns
4. Numerals · use real numbers where possible

6. Save the script

Long-form scripts live in the Notion Content DB page body. Don't write to a file in the repo (text drafts stay in Notion · the repo is for assets).

Create a row via mcp__claude_ai_Notion__notion-create-pages with data source <YOUR_NOTION_CONTENT_DB_ID>:

• Title: video title (e.g. "Why GTM dashboards are dead · YouTube")
• Status: Scripting
• Pillar: pick the multi-select that matches the angle
• Format: Video
• Channel: YouTube (default) · or LinkedIn if it's a clip going there

Page body:

Type: long / short / clip
Target duration: <m:ss>
Source: <master URL if from /repurpose>

— Beat outline —
1. HOOK
2. COLD OPEN
...

— Script —
[full timed script with B-roll cues]

— Production notes —
- Camera setup: <if specific>
- Lower-thirds: <if any>
- End-card link: <where the CTA points>

If invoked from /repurpose, the orchestrator may handle row creation · check the source context to avoid duplication.

7. Hand off to production

After the script lands in Notion, suggest the next step:

• For a recorded video: record → run /video-use to cut + grade + subtitle.
• For a synthetic / motion-graphics version: run /launch-video (different beat structure, but the script's beats can map to its Remotion scenes).
• For a clip going on LinkedIn: chain into /linkedin-copywriter for the post copy that wraps the video.

Don'ts

• Don't pad to hit duration. If 8 min feels thin, propose a 60s short or a clip and stop. Long videos that feel padded retain worse than tight 60s shorts.
• Don't draft before the beat outline is approved. Spoken-script rewrites at the line level cost 5-10x outline rewrites.
• Don't write essays. This is spoken. Sentences are short. If a sentence reads like an essay sentence, it'll sound like one. Trim.
• Don't stack lists. "First X. Second Y. Third Z." is the fastest way to make a viewer click off. Break each item into its own beat with a B-roll change.
• Don't borrow YouTube guru tropes. "Buckle up" / "Stay tuned" / "Drop a like and subscribe" / "What's up everybody" all read like AI slop in your voice. Pretend those don't exist.
• Don't bolt the CTA onto the end of the last beat. Give it its own beat. You say the line, look at camera, ask. Pause matters.
• Don't write the video title here. Title generation is its own art (and YouTube SEO matters). Put a working title at the top of the script and flag for separate optimization.
• Don't fabricate quotes / metrics / customer names. Same rule as every other content skill.

See also

• Brand reference: <YOUR_BRAND_DOC> · voice rules + skill checklist (if you have a brand doc with voice rules, link it here)
• Voice corpus (shared with /linkedin-copywriter and /long-form): ../linkedin-copywriter/corpus.md
• Sibling skill (written long-form): /long-form
• Production hand-off: /video-use (edit recorded footage) · /launch-video (synthetic motion graphics)
• Notion Content DB data source: <YOUR_NOTION_CONTENT_DB_ID>
• Upstream skill: /researcher (sources ideas), /repurpose (orchestrates a master into spoken-script variants)