skills/youtube-script/references/commentary.md

# YouTube Commentary & Video Essay Scriptwriter

Write an original spoken-word script in the commentary/essay genre. The patterns below come from analyzing the 20 most-viewed videos each of four defining channels in the genre (~80 scripts). They describe *craft*, not personas: never copy any real creator's catchphrases, greeting rituals, audience names, or in-jokes — those belong to specific people; invent originals. The user's own voice fills the persona slot; these rules shape the writing around it.

## Step 1 — Intake

Establish before writing (ask only for what's missing; infer what you can):

1. **The artifact** — what is being reacted to or analyzed? (a video, trend, film, creator, product, artwork). Commentary cannot exist without a concrete subject; if the user gives only a vague topic ("cringe TikToks"), ask for or propose specific artifacts.
2. **The lane** — pick one of three sub-styles; each has its own beat template:
   - **Comedic commentary** (15–30 min): host reacts to found footage/phenomena with jokes. Pick when the goal is entertainment through mockery-with-affection.
   - **Deadpan satirical explainer** (2–8 min): a mock-guide delivered in calm textbook voice. Pick for topic-based comedy with no footage.
   - **Earnest video essay** (6–10 min): analytical close-reading of a work. Pick when the goal is insight, not laughs.
3. **Target length** in minutes (drives beat timings; ~140–150 spoken words/minute). If the target is ≤4 minutes, apply **Short-form mode** (below) instead of the long-form beat map.
4. **The host's stance** — what do they actually think of the artifact? The stance must appear in the first 30 seconds.
5. **Persona facts** (optional but valuable) — self-deprecation material, running bits, audience name.

## Step 2 — Hook (offer 3 options)

Draft three hook options from different rows of this menu, tuned to the artifact. By 0:30 the viewer must know **the artifact and the host's stance** — no generalized throat-clearing.

| Hook type | Formula | Best for |
|---|---|---|
| Cold-open sketch | Act a character/scene that obliquely stages the topic for 10–20s, then snap to normal register | Comedic commentary |
| Artifact-first | Play/read the weird thing raw for ~10s → react → "has anyone else noticed…" | All lanes (essay default) |
| Discovery story | "People kept sending me [X], so I finally looked into it" | Comedic commentary |
| Experiment stakes | "I [spent/did costly irreversible thing] to answer [question]" | Commentary with an experiment premise |
| Verdict-first | "[X] is [bad/brilliant]. Let me explain." | Commentary, essay (positive polarity) |
| Deadpan definition | Textbook definition of an everyday thing, played straight; first undercut within 20s | Satirical explainer |
| Fake-service | "I know a lot of people struggle with [X], so I made this guide" → anti-help | Satirical explainer |
| Question / superlative | "Of all [category], this one [extreme claim]" — then immediately count something concrete | Essay |
| Personal confession | "[Embarrassing autobiographical fact], which is why I can explain [topic]" | Explainer, commentary |

## Step 3 — Structure by lane

### Comedic commentary (15–30 min)
- **0–3%**: cold-open bit or greeting ritual.
- **3–10%**: origin story (how the artifact found the host) + premise promise. Optional ≤15s announcement, self-aware and cut off fast.
- **~10%**: explicit spoken chapter turn ("Anyway — let's get into it.").
- **10–80%**: the **clip-react loop**, the atomic unit (30–90s per cycle):
  1. ≤20 seconds of subject material — never let footage run longer uninterrupted.
  2. Literal restatement of what happened.
  3. Logic interrogation (ask the clip the questions the audience is thinking).
  4. Act-out: voice the on-screen subtext as dialogue, or inflate into a hypothetical scene.
  5. Tag, then next clip.
  Order segments by **ascending absurdity** and join them with labeled escalation pivots ("Even worse though…", "And somehow it gets weirder").
- **40–60%**: one mid-video pivot — a real setback folded into the plot, a thesis deepening, or a format twist.
- **80–90%**: verdict plus **one sincere zoom-out paragraph** (why this actually matters). Irony never closes the video.
- **90–100%**: light outro in persona — subscribe-as-lore gag and/or a next-video teaser joke, under 20 seconds total.

### Deadpan satirical explainer (2–8 min)
Definition open → announce a scaffold (tiers/stages/steps) → 12–18 setup-twist units at ~15–25s each → final tier reverses onto the speaker or viewer (the taxonomy's top tier turns out to be the audience itself) → stop dead, no outro.

### Earnest video essay (6–10 min)
Cold artifact → thesis as one hedged "I think" sentence by 15% → **evidence ladder**: (1) quantify something concrete first (word counts, shot counts, seconds of pause), (2) name the technique with a fast inline definition, (3) corroborate externally, (4) only then interpret → **replay the key moment** so the audience feels their own perception upgrade → widen to human stakes in the last 20% → second-person closer → warm outro.

## Step 4 — Register rules (all lanes)

- First person, present tense, contractions, spoken diction. Narrate clips in narrative present ("So then he goes…").
- **Peer, never lecturer.** Hedge big claims through the self ("I think", "for me"); use comic precision only for clearly invented hypotheticals. Every factual number still needs evidence.
- Perform confusion before resolving it. Ask the artifact questions out loud.
- **Self-deprecation at three fixed points**: the hook, immediately before any punch at someone else (it licenses the mockery), and inside the CTA.
- Ethics check before punching: verify real harm, echo affected groups, target institutions/output/behavior — never appearance or fans.
- Keep seams visible: mistakes and production problems stay in as bits.
- Escalate in threes: literal contradiction → hypothetical scene → world-scale consequence.
- Callbacks: plant the weirdest found line early, re-drop unannounced 2–3 times, label the final callback (labeling it is the joke).
- Reference only setups the script actually planted: before writing "as I promised" / "like I said earlier" / "callback!", verify the earlier line exists in this script. A false callback is an instant credibility break.

**Joke density targets**: deadpan explainer — a reversal every 15–25s, twist as the *final clause* of a calm sentence, never telegraphed; commentary — big laugh every 30–60s with micro-jokes (tags, act-outs) between; essay — no jokes required, but one delight beat per minute (a count, a replay, a "did you see that?"). Lists: N sincere items + 1 absurd in identical grammar.

## Step 5 — Ending & outro

Do not invent sponsor reads, brand integrations, or other monetization beats. Include a specific monetization integration only when the user explicitly requests it, placed outside the clip-react loop and before the final sincere payoff. Otherwise, after the sincere zoom-out, use a light outro in persona: subscribe framed as joining an original running bit (never a plea), and/or a next-video teaser gag. Under 20 seconds total, or cut entirely.

## Step 6 — Anti-patterns (hard avoid list)

- No "in today's video we'll cover…" previews — structure is enacted, not announced (unless the preview is itself a joke).
- No neutral recap: every description of the artifact carries stance or joke.
- No manufactured hype or ad-copy voice anywhere.
- No cliffhanger abuse: retention comes from density and escalation, not withheld information.
- No vibes-only superlatives in essays: every "best/perfect" is paid for with a count, a named technique, or a replay.
- No explaining jokes (except a deliberately labeled callback).
- Never imitate a real creator's identity markers: greetings, audience names, outro songs, catchphrases.

## Short-form mode (target ≤ 4 minutes)

The beat maps above were measured on long-form; under ~4 minutes the physics change. Apply these overrides and keep the register rules (peer voice, self-deprecation licenses the punch, ethics check before punching):

- Budget: ~150 wpm of host speech → a 3-minute video is ~350–420 host words plus clip air (clips get even shorter: ≤8 seconds each).
- **Artifact + stance in the first sentence** — no cold-open sketch, no greeting ritual, no origin story.
- At most 2–3 clip-react cycles, ordered by ascending absurdity, each cycle compressed to: clip → one restatement-or-interrogation line → one act-out or tag.
- The sincere point compresses to ONE closing line — it still exists (irony alone doesn't close a video, even a short one), but it's a sentence, not a paragraph.
- **One loop or callback max**; no announced scaffold, no mid-video pivot.
- Timestamps every 15–30 seconds; retention map 3–5 lines; outro one line or cut.
- If vertical (Shorts): open ON the artifact (clip playing in the first 2 seconds); the stance line must work muted; consider a loop-close.

## Output format

Deliver the script as markdown:

```
# [Working title] — [lane], ~[N] min

## Hook options (pick one)
[3 hooks, each 40–80 words, labeled by type]

## Script
[Timestamped sections every 60–90s of runtime: 【0:00】…]
[VO lines as flowing spoken prose]
[Production cues in brackets on their own lines: [CLIP: …] [ACT-OUT: …] [B-ROLL: …]]

## Retention map
[One line per escalation pivot / callback plant / loop, with timestamp]
```

Claims about the script belong in the script: no self-reported pacing or compliance boasts in titles or headers (word counts, "stance lands by 0:30") — the retention map is the only meta-commentary, and it lists devices, not grades.

For deeper pattern detail (12-hook taxonomy with illustrative examples, per-archetype mechanics, full anti-pattern rationale), read `references/commentary-patterns.md`.