skills/youtube-script/references/true-crime.md

# YouTube True-Crime Scriptwriter

Write an original spoken-word script for a true-crime or dark-story video. Patterns below come from the most-viewed videos of three defining channels in the genre (~50 scripts). They describe craft, not personas — never copy any real channel's running bits, ritual jokes, or sign-offs; invent originals for the user. This genre carries real ethical weight: the ethics rules in Step 6 are retention devices AND obligations — top channels prove they're the same thing.

## Step 1 — Intake

Establish before writing (ask only for what's missing):

1. **The case** — facts, timeline, verified sources. If the user provides only a case name, research or ask for: victims, perpetrator, timeline, investigation beats, legal outcome. Never invent case facts; mark uncertain details for the user to verify.
2. **The lane**:
   - **Oral storytelling** (20–45 min, narrator to camera): one continuous chronological story, mystery withheld.
   - **Footage analysis** (any length, VO over interrogation/bodycam/court footage): killer revealed early, the collapse of their story is the engine. Only viable if real footage exists.
   - **Conversational storytelling** (30–60 min, host with parallel activity): cradle-to-conviction chronology, host personality forward.
3. **The reveal-ordering law** — decide what to withhold: the mystery you run is NEVER the one the title answered. Title names the killer → run how-they-were-caught. Title hides them → run who/how. Footage shows them → run why-the-lie-collapses.
4. **Legal status** — pre-verdict cases need "allegedly" discipline throughout.
5. **Target length** (~145 spoken words/min). If the target is ≤4 minutes, apply **Short-form mode** (below) instead of the long-form beat map.

## Step 2 — Hook (offer 3 options, ≤45 seconds each)

| Hook type | Formula |
|---|---|
| Ranked superlative | "Of the [N] stories I've covered, this is top-[n] most [adjective]." |
| Located warning | "There's a [graphic sequence] at the END — discretion advised." (warning = promise + address) |
| Artifact tease | "They found a [video/photo]. When you realize what it shows… we'll play it near the end." |
| Victim clock | "At exactly [time] on [date], [name] [did an ordinary thing] — and never [came back]." |
| Uniqueness thesis | "What makes [case] unlike anything in true crime is [unusual dataset/behavior]." |
| Normalcy trap | "A [perfect family / ordinary night] — until [date], when [someone in their midst had a secret]." |
| Cold scene + rewind | 30s of the scariest moment, cut at the peak, ritual break, restart at chronological beginning. |
| Raw audio cold open | Unnarrated 911/jailhouse audio plays first; identify the voice only afterwards. |

The content warning doubles as the strongest retention device in the genre: it tells the viewer exactly what they forfeit by leaving, and where it lives.

## Step 3 — Structure (single-case template)

1. **0–2%** Hook + warning.
2. **2–5%** Fixed ritual (greeting/theme, an original one) ending in a clear boundary — same every episode, trains "story starts now."
3. **5–20%** Normalcy backstory. Victims get personhood before peril: marriages, jobs, dreams. Victim-first if identity withheld; perp-first (childhood → red flags) if the title reveals them.
4. **20–45%** Anomaly + investigation, told strictly from the discoverers' POV — the barista who found the door unlocked, not an omniscient camera. A dead end or lead every 2–4 minutes. State the police's wrong theory sincerely before it breaks.
5. **45–55%** A clean act break: a calm procedural beat that releases tension before the escalation resumes. **Nothing interrupts the story's second half.**
6. **55–80%** Escalation ladder: each segment ends on a contradiction, partial reveal, or "it would turn out…" pivot. Release reveals partially: name → method → location.
7. **80–90%** THE payoff: second-pass retelling from the inside (confession-based), or the capture. **Dilate time here** — beat-by-beat sensory detail lives only in this zone; everywhere else time compresses.
8. **90–97%** Fates and sentences — fast, cold, precise. Land one exact number here (exact seconds, dollars, dates).
9. **97–100%** Outro ritual, identical every episode. No recap — the story just told IS the summary.

**Footage-analysis variant**: reveal crime + perpetrator fully by 8% → chronological footage spine sequenced as escalating institutional encounters (statement → interrogation: rapport, fatigue, confrontation → arrest) → teach each interrogation tactic just before the footage shows it working → trial excerpts with one devastating cross-examination → cold verdict card, no outro.

## Step 4 — Register & retention rules

- Speak, don't write: contractions, additive chaining ("and so… and then…"), paraphrased dialogue ("he's like…"). No "moreover/furthermore".
- **Tense is a camera**: past tense for background; slide into present tense the moment someone discovers something.
- Address the viewer directly at least once per act: brief them ("before we continue, you need to understand…"), quiz them, or hand them a choice.
- Foreshadow with *time marker + unspecified shift*: "then in [month], everything changed."
- Withhold identities behind generic labels ("a man in his 30s", "the woman on the tape") until the reveal beat.
- **Establish exact numbers early; collide the lie with the number late** (the alibi says 10 minutes; the camera metadata says 62 seconds).
- Promissory notes with addresses: "which we'll come back to", "she asks this question nine times in this interrogation" — every note paid on screen.
- Reference only setups the script actually planted: never write "as I promised" or "like I said earlier" unless the earlier line exists in this script — a false callback is an instant credibility break.
- **Atrocity never touches atrocity**: alternate horror ↔ procedure, analysis, or (in conversational lane) humor. Jokes live in the seams, never inside a violence beat.
- Payoff cadence by lane: oral storytelling = long fuse, micro-payoff every 2–4 min, one bomb at 80–90%; footage analysis = a rung every 5–10 min (contradiction → amended story → partial admission → arrest → verdict); conversational = a commentary beat every 20–40 seconds.

## Step 5 — Ending & outro

Do not invent sponsor reads, brand integrations, product plugs, or other monetization beats. Include a specific monetization integration only when the user explicitly requests it, and place it before the story's second half at a clean act boundary. Otherwise, oral storytelling closes with a short fixed outro ritual, identical every episode (an original recurring sign-off, never copied from a real channel), placed only after the fates/sentences beat; footage analysis ends cold on the verdict card with no outro at all. One optional plain line (next-episode pointer or story-suggestion invitation) may sit at the very end — nothing ever interrupts the story body.

## Step 6 — Ethics rules (non-negotiable, and they double as trust devices)

- Victims get personhood before peril; never invent their interior monologue at death. Reconstruct only from confessions, testimony, and records — and say so.
- Fence all speculation: "allegedly" pre-verdict, "it's believed", "according to [source]". Grade confidence out loud.
- Name omissions and their reasons on camera ("I won't name the victims, and here's why") — stating the reason is itself a trust device.
- Render killers pathetic, never mythic. No killer glamor, no victim-blaming.
- Name institutional failures where documented.
- Narrate gore mechanism briefly, cut to consequence — never linger.
- Content warnings are contractual: give them verbally, with the payoff's location.

## Step 7 — Anti-patterns (hard avoid list)

- No unresolved endings — verdict/fate always delivered in-video; retention is intra-video, never cross-video bait.
- No mystery about the thing the title already answered.
- No mid-climax ads or CTAs. No recap padding at the end.
- No fake experts or "criminologists say" vagueness — cite named sources, techniques, documents.
- No omniscient dramatization, no tonal whiplash into comedy during atrocity, no written-English register.

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

The beat maps above were measured on long-form; under ~4 minutes the physics change. Apply these overrides — **the ethics rules in Step 6 do not compress away**:

- Budget: ~150 wpm → a 3-minute video is ~420–470 spoken words.
- **The first line IS the hook** — victim clock or normalcy trap in one sentence. No ritual section, no located-warning preamble (a needed warning becomes one clause: "fair warning — this gets dark").
- Single case, single reveal: normalcy in one or two sentences (victims still get personhood — it compresses, it doesn't vanish), anomaly by 20%, one "it would turn out…" pivot, fate/sentence in the final line.
- **One loop only**: the withheld identity or detail, opened at the top, paid at 80–90%. No second-pass retelling, no chapter foreshadow lines.
- Timestamps every 15–30 seconds; retention map 3–5 lines; outro one plain line or cut.
- If vertical (Shorts): hook must work muted; consider a loop-close (final line feeds the first).

## Output format

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

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

## Script
[Timestamped sections: 【0:00】…]
[Narration as flowing spoken prose; dialogue paraphrased in-voice]
[Cues in brackets: [FOOTAGE: …] [PHOTO: …] [MUSIC STING] [RE-ENACTMENT B-ROLL: …]]
[⚠ VERIFY tags on any fact the user must confirm against sources]

## Retention map
[One line per located promise / withheld identity / foreshadow line / number-collision, with timestamp]
```

Claims about the script belong in the script: no self-reported pacing or compliance boasts in titles or headers — the retention map is the only meta-commentary, and it lists devices, not grades.

For full pattern detail (15-hook taxonomy with illustrative examples, per-lane beat maps, the two-pass structure, ethics conventions), read `references/true-crime-patterns.md`.