skills/youtube-script/references/tech-review.md

# YouTube Tech Review Scriptwriter

Write an original spoken-word script for a tech video. Patterns below come from the 20 most-viewed videos each of six leading channels in the genre (~120 scripts). They describe craft, not personas — never copy any real creator's rituals, sign-offs, or catchphrases; invent originals. The genre's core contract: **the presenter performs work the viewer can't or won't do — spending, breaking, benchmarking, or reading — and reports back in first person, skeptical by default, excited only when the product forces a change of mind on camera.**

## Step 1 — Intake

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

1. **Format**:
   - **Single-product review/impressions** (10–25 min, on camera).
   - **Gauntlet/ranking** (list of products/items with per-item verdicts).
   - **News-analysis/roundup** (VO-dense, recency-driven).
   - **Tutorial/demo** (screen-recording narration).
2. **The labor proof** — what did/will the creator actually do? (spent $X, tested for N days, read all N pages, built it, broke it). Hands-on review hooks need a number + a truthful first-person effort verb. If there is no labor proof, do not fake a hands-on review: ask for it or switch to a clearly sourced news-analysis/tutorial lane that discloses the lack of testing.
3. **The honest verdict** — what does the creator actually think? Reviews are written backwards from the verdict.
4. **Scope hedge** — what they did NOT test (the verdict must state it).
5. **Target length**; talk-pace differs by format: 160–200 wpm on-camera, 200–235 wpm VO-dense. If the target is ≤4 minutes, apply **Short-form mode** (below) instead of the long-form beat map.

## Step 2 — Hook (offer 3 options, ≤3 sentences / ≤15 seconds each)

Rule for hands-on formats: the hook must contain **at least one number and a truthful first-person effort verb** (spent / read / bought / broke / tested). For a sourced non-hands-on lane, use a named evidence action and never imply a test that did not happen.

| Hook type | Formula |
|---|---|
| Stunt ledger | "I'm about to spend $[X] in the next [N minutes] on the most [superlative] [category] — then we test whether it's [dream] or [scam]." |
| Fresh intel | "[N hours] ago, [event]. I've [read/tested] all [N pages/units]. Here are the [N] things that matter." |
| Verdict-first tease | "[One-line verdict]. The real question is [the deeper why]." |
| Disaster cold-open | "[Failure shown live]. I did everything right. So what's going on?" |
| Forbidden object | "[Gatekeeper] refused to [sell/fix/send] this. I got one anyway." |
| Countdown bomb | "[N] worst/best [things] — and #1 will [visceral promise]." (tease the top item, never name it) |
| Contrarian flip | "Everyone says [X]. That's not what we're doing today." |
| Payoff montage | "[Rapid results display] — I'll show you exactly how." (tutorials) |

## Step 3 — Structure by format

### Single-product review
- **0–5%**: hook + credential ("I've been using it for a week") + roadmap ("everything you need to know").
- **5–12%**: what-is-it frame. State price now if price IS the news; otherwise hold it.
- **12–70%**: walkthrough — **never spec-sheet order**. Choose: *curiosity order* (the questions a normal person would ask, in sequence), *physical-tour order* (walk the object end to end), or *usage chronology* (unbox → set up → real task). Pull the single most impressive thing forward and flag it as such ("the most impressive thing is X — I'm not kidding"). Plant 3+ named deferrals ("we'll get to the battery").
- **70–85%**: downsides block, explicitly numbered.
- **85–95%**: **verdict** — bottom line + who should buy it + honesty hedge ("this isn't the full review; I haven't [X]").
- **95–100%**: one-line sign-off + named next video.

### Gauntlet/ranking
Announce the sort rule in sentence two (price ascending / countdown to #1 / tiers) — the announced sort IS the retention spine. Item loop: setup → live demo → score or verdict per item. Keep a running tally (spend counter, score count) with a total reveal at the end.

### News-analysis
Recency hook → 30-second context primer → numbered findings (strongest first, or wildest explicitly held for last: "the crazy stuff is coming in a few minutes") → per-finding mini-verdicts with caveats interleaved (never a separate limitations section) → final calibration paragraph → sign-off.

## Step 4 — Register & opinion rules

- First person, present tense, contractions mandatory. Zero passive-voice openings.
- Rhythm: long explanatory sentence → short verdict tap ("That's dope."). At VO speed, flat syntax — one idea per sentence, analogy within 5 seconds of any hard concept.
- **Specs are never listed — they're translated** ("that means…"), converted to multiples ("30,000× the PS1"), or offloaded to on-screen overlays while the script says the meaning.
- **Skepticism is the resting state; excitement is a state change the script dramatizes** ("I was skeptical… now I'm completely sold").
- Demo before judgment, always. Opinions prefixed with ownership ("I think", "to me"). Cite your own limits ("I'm not an expert on this, but…").
- Every negative gets a fairness beat ("in fairness…") and an escape valve (who it's still for / what would fix it). Target decisions, never the people who bought the thing.
- Every hype claim gets a comparison class or an audit; manufacturer claims trigger a test or an explicit "I hope that's true".
- Hedge in three tiers: future-proof ("we'll see"), sample-size ("one unit, one week"), epistemic ("I can't confirm this yet").

## Step 5 — Retention rules

- One **named deferral** per 2 minutes ("we'll get to the battery") — specific feature, never vague "more later".
- Reference only deferrals the script actually planted: never write "as I promised" or "I said I'd get to this" unless the earlier line exists in this script — a false callback is an instant credibility break.
- One "but here's the thing" pivot per section.
- Hold the biggest number (total cost, top rank, wildest finding) for the final 20%; tease it once at the top.
- Close the hook's loop verbatim at the end (open "I never got it" → close "now I get it").
- Running games serialize reveals cheaply: a price-interrogation gag, a running poll, a per-item score the audience can argue with.

## Step 6 — 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 keep it outside any running test, demo, or reveal. Otherwise end on a consistent one-line sign-off (an original ritual, never a real creator's) or a forward hook. If a subscribe ask is wanted, it's one line, themed to the moment, and tied to a named future payoff ("subscribe for the [X] comparison next") — boilerplate "smash that like button" is absent from every top performer.

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

- No spec-sheet-order walkthroughs. No invented monetization beats; only include a specific integration when explicitly requested.
- No unowned opinions ("some say…") — replace with "I tested / I read / I bought".
- No free-floating superlatives without a comparison class or demo.
- No verdict dump at 0:00 unless a deeper "why" mystery is immediately re-established.
- No press-release language except quoted for audit or mockery.
- No apologizing for length — justify it with promised density.
- No generic mid-roll CTAs of any kind.

## 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 (labor proof, specs-as-translations, earned hype, fairness beats):

- Budget: short tech runs fast — 160–200 wpm → a 3-minute video is ~500–580 spoken words.
- **Verdict-first is the default hook**: one-line verdict + the number + the effort verb in sentence one ("I used it for two weeks — it's the best $400 I've spent this year. One problem.").
- Structure collapses to: verdict hook (0–10%) → 3 proof beats, strongest first (10–70%) → the one catch (70–85%) → who it's for + honesty hedge in one line, hard stop.
- **No named deferrals** — there's nothing to defer to; the "held reveal" becomes at most one held number (price or score) paid in the final line.
- Timestamps every 15–30 seconds; retention map 3–5 lines; outro one line tied to a named payoff, or cut.
- If vertical (Shorts): the product must be on screen doing something in the first 2 seconds; hook must work muted; consider a loop-close.

## Output format

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

## Hook options (pick one)
[3 hooks, each ≤3 sentences, labeled by type, each with a number + effort verb]

## Script
[Timestamped sections: 【0:00】…]
[VO as flowing spoken prose; verdict taps on their own lines]
[Cues in brackets: [DEMO: …] [B-ROLL: …] [OVERLAY: spec numbers] [SCORE CARD: 7/10]]

## Retention map
[One line per named deferral / catch pivot / held reveal / tally beat, 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 (9-hook taxonomy with illustrative examples, density techniques for 200+ wpm scripts, opinion architecture), read `references/tech-review-patterns.md`.