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

# Tech Reviews — Scriptwriting Craft Patterns

**Corpus:** Derived from ~120 top-performing scripts across 6 leading channels in the genre (sources anonymized for publication). Fully read: 23 top performers; hooks (first 30–60s) surveyed for all ~120. WPM figures computed from transcript word counts ÷ metadata duration. Auto-captioned files have no punctuation, so sentence-length stats are only claimed for manual/whisper transcripts.

**Archetypes used throughout** (each stands for one studied channel type):

- **The solo authority** — calm single-host reviewer; drama comes from the product.
- **The showman** — high-stakes stunt buyer; his money, risk, and emotions are the plot.
- **The code satirist** — ultra-dense ironic dev-news narrator.
- **The multi-host workshop** — team channel; banter, staged reveals, argued verdicts.
- **The tool curator** — enthusiast aggregator; live demos and weekly roundups.
- **The paper analyst** — research-first commentator; calibrated judgment with receipts.

---

## 1. Voice & register

**Person and tense.** Universally first-person singular, present tense, direct address to "you." The reviewer is a character, never a narrator: "I've had this on my desk for about a week now" (the solo authority), "I'm about to burn $25,000 of my own money" (the showman). Even the news archetypes frame reporting as personal labor: "I stayed up all night reading every page of this thing" (the paper analyst). The corpus contains zero passive-voice openings and zero third-person "this reviewer" constructions.

**Contractions everywhere.** "It's," "gonna," "kinda," "'cause" are the default in every manual transcript. The solo authority's manual captions preserve spoken fragments as-is: "It's a lot. It's gonna take some getting used to." Scripts are written to be said, not read — the multi-host workshop and the solo authority both leave in self-corrections ("over at the... at the workshop") because recovery-on-camera reads as authenticity.

**Sentence rhythm.** Two distinct rhythm families:

- *Conversational-loop* (the solo authority, the multi-host workshop, the showman in demo mode): a long explanatory sentence followed by a short verdict tap. "Wow, that's loud." → then a full paragraph of context → "That's great." The multi-host workshop runs 151–171 wpm, the solo authority 165–203 wpm — slow enough for on-camera physicality (peels, drops, taps) to fill space.
- *Compression-delivery* (the code satirist 202–234 wpm, the tool curator 191–237 wpm, the showman in VO mode 188–235 wpm): every sentence carries one fact or one joke, no throat-clearing, clauses chained with "and/but/so" instead of full stops so the read never pauses.

**Enthusiasm calibration.** The pattern is *earned hype*: superlatives are always attached to a comparison set or a demonstrated moment, never free-floating. The solo authority: "the best I've ever seen in this category" — only after listing four rivals it beats. The showman self-reports the emotional arc: "I was skeptical... but now I am completely sold." Skepticism is the resting state; excitement is a *state change* the script dramatizes. The paper analyst models the inverse: hype claims get audited ("that stat is doing a lot of cherry-picked heavy lifting"). The code satirist's resting state is irony — hype is quoted only to be deflated.

**Spec-talk handling.** Specs are never listed; they are *translated, ranked, or dramatized*:

- Translate to experience: "independent motor control at each wheel, and it's absurd"; a higher-voltage architecture → "it charges faster... and the wiring gets lighter."
- Convert to multiples: "roughly 30,000 times more powerful" than the first-generation model; "enough first-gen units chained together to pave a highway across a city" (the showman).
- Offload to graphics: "I'm gonna put the numbers up here" (the solo authority) — the script says the *meaning*, the overlay says the number.
- Interrogate: the multi-host workshop treats every claimed spec as an accusation to verify ("Mid-tier card. Sure it is.").

---

## 2. Hook taxonomy

Every hook in the corpus is one of these nine types. Formulas are fill-in-the-blank.

**H1. The Stunt Ledger** — *"I'm about to spend $[X] in the next [N seconds/minutes] buying the most [SUPERLATIVE] tech on the [internet/planet] — and then we're going to test it."*
The showman's signature. His first three sentences are machined: (1) first-person money/quantity commitment with two numbers, (2) scope/method escalation, (3) test-promise with a stakes question. "I'm about to burn $25,000 of my own money in the next 5 minutes"; "or if it's all just marketing nonsense." Variant: past-tense completion — "I've just bought every generation of this console, brand new" — followed by the fairness apparatus ("a period-correct TV for each one") and the question ("let's find out how far gaming has actually come").

**H2. The Fresh-Intel Dispatch** — *"[N hours] ago, [event]. I've [read/tested] all of it. Here's what matters."*
The paper analyst's signature: "less than a day ago, a report dropped that's going to rattle the entire industry"; "I got access roughly 48 hours ago." Three moves: recency stamp, labor proof ("all 140 pages of it"), curation promise ("only the findings that actually matter"). The tool curator runs the low-effort version ("this has genuinely been the wildest week in this space"). The code satirist's news videos open the same way but with a spin of irony: "Yesterday, the most ironic thing imaginable happened."

**H3. The Verdict-First Tease** — *"[Instant verdict]. Now here's the interesting question underneath it."*
The solo authority: "Okay, up front: this is an easy phone to recommend" then pivots to diminishing-returns economics; "you have $1,300… why would you get any other phone? No, seriously." Kills the will-he-won't-he and replaces it with a *why* mystery.

**H4. The Disaster Cold-Open** — *"[Live failure on screen]. How did this happen? I did everything right."*
The multi-host workshop opens inside a staged crash: "Come on! This is unusable!" then "most of these are honestly terrible." Also the showman: "I've managed to brick my phone by changing one settings toggle." The hook is a wound + a detective promise.

**H5. The Forbidden Object** — *"[Company] refused to [sell/fix/send] this. I got one anyway."*
The multi-host workshop: "they refused to fix it"; "and they refused to send us one. Well, we got one anyway"; "it left the factory in someone's backpack." Access itself is the story.

**H6. The Countdown Bomb** — *"Welcome to my top [N] most [ADJECTIVE] tech fails… number one is going to [visceral promise]."*
The showman's list videos: "spoiler alert: one of these finally breaks our rating scale." The top item is teased, never named.

**H7. The Contrarian Flip** — *"Everyone says [X]. That's not what we're doing today / that's not true."*
The code satirist: "telling you the number one language to learn… that's not what we're doing today"; "these are all statements that are not true." The solo authority runs the gentle version (a title that quietly contradicts the consensus take).

**H8. The Character Sketch** — *"Meet [entity]. [Two lines of setup]. Now the twist."*
The code satirist: an AI-generated influencer with a six-figure brand-deal income — "she has never once existed"; the programmer introduced as "a caffeine-powered offshoot of the great ape family." A joke premise that turns out to be the thesis.

**H9. The Payoff Montage** — *"[Rapid cuts of finished results.] Every one of these took minutes — and I'll walk you through the exact steps."*
The tool curator's tutorial hook: rapid-fire display of finished outputs, then "I'm going to show you how to do that." The proof precedes the promise.

---

## 3. Structural beat map

**Canonical single-product review/impressions (the solo authority, 15–25 min):**

- 0–5%: Hook + credential ("I got a brief hands-on before launch") + roadmap ("this is everything you need to know").
- 5–12%: Frame — what *is* this, price stated early when price *is* the news (a $3,000 headset; a six-figure electric pickup).
- 12–70%: The walkthrough. **Never spec-sheet order.** Three orders observed instead:
  1. *Curiosity order* — questions a normal person would ask, sequenced: "what is it → how does it work → what does it actually do → is it any good" (the solo authority literally section-titles by questions).
  2. *Physical-tour order* — walk the object: wheels → bed → doors → front → interior → software → drive. The camera's path is the outline.
  3. *Chronology/usage order* — unbox → set up → first boot → real task (the multi-host workshop's builds; the showman's console gauntlets go generation-by-generation through time).
  Within any order, the single most impressive thing gets pulled forward and flagged: "the most impressive thing is the hand tracking. I'm not even kidding."
- 70–85%: Downsides block, explicitly enumerated: "that doesn't mean there aren't any downsides… number one."
- 85–95%: **Verdict lands here** — "bottom line," pros/cons weighed, buyer identified ("a toy for the wealthy and a workbench for developers"), plus scope hedge ("Now this isn't a review. I haven't towed anything with it yet").
- 95–100%: CTA + next-video pointer.

**Gauntlet/list structure (the showman, the multi-host workshop's roundups, the tool curator, the code satirist's listicles):** hook → sorting rule announced up front ("from least expensive to the most expensive"; "cheapest to most expensive"; tiered depth levels) → item loop (setup → demo → score/verdict per item) → aggregate verdict. The announced sort *is* the retention spine: viewers know position implies escalation.

**News-analysis (the paper analyst, the code satirist's news format, the tool curator's news):** recency hook → 30-second context primer ("let me give you a 30-second summary") → numbered findings, strongest first or strongest teased for last → implications → sign-off. The paper analyst's verdict is distributed: each finding gets an immediate mini-judgment, with a final calibration paragraph ("a genuinely big day… with a longer list of asterisks than the keynote let on").

**Where the verdict lands overall:** the solo authority and the multi-host workshop at ~85–90%; the showman scores continuously (per item) so the "verdict" is the final ranking; the paper analyst verdicts up front then defends ("it's smarter in most ways, cheaper, faster" as a first sentence); the code satirist's verdict is the closing punchline ("it was all smoke and mirrors the whole time").

---

## 4. Retention machinery

**The named deferral.** The solo authority's core tool — raise a question, promise the answer, timestamp it vaguely: "we'll get to the wheels in a second," "we'll get to that later" (used 6+ times in a single script). The deferral is always *specific* (a feature, not "more later").

**The catch pivot.** Formula: [praise] + "but here's the thing" / "but there's a twist" + [complication]. "But here's the thing… I can't stand this new controller" (the showman); "you came here thinking this creator has one warehouse. He actually has three" (the showman, on a famous creator's studio tour); "that's cool and all, but" (the code satirist's recurring connective). The multi-host workshop structures whole videos as competence/catastrophe alternation: "there are these sprinklings of competence, and then… confoundingly stupid decisions."

**Running tally.** The multi-host workshop's price-interrogation gag — every part triggers "how much did we pay for this?" answered by an off-screen voice — converts a parts list into a recurring game with a final total reveal ("$400 with the cooler. That is brutal."). The showman keeps a running spend counter ("$4,800. That's where we're at right now") and a running gag tally (an audience poll re-run at every stop of a studio tour).

**Ranking/tier tension.** Scores per item (a per-item score out of 10) let the audience argue; the sort order (price ascending, tiers descending, countdown to #1) guarantees the best material is still ahead. Explicit tease of the extreme: "number one is so bad it should legally require a warning label" (the showman).

**Held reveals.** Price to the end when price is the punchline ("so what did the whole compound actually cost?" — the showman's final beat); identity of a counterfeit graphics card held until the benchmarks force it (the multi-host workshop); the paper analyst holds the wildest finding: "the finding that actually floored me is still a few minutes out" → "now we reach the crazy zone… put your seat belt on."

**Delayed title payment.** The code satirist's date-stamp ritual ("It is March 4th, 2025, and this is your daily dev dispatch") lands ~60–90 seconds in, *after* the hook — a built-in act break. The showman's mid-video sub-ask is themed to the moment, not boilerplate: "a sub right now would be… uplifting" (said as a delivery drone lifts off the table).

**Loop-closing.** Hooks plant a question the ending answers verbatim: the solo authority opens one script "I never understood the hype around this thing" and closes "now I get it." The stunt-ledger hooks close with the total and a worth-it ruling.

---

## 5. Opinion architecture

**Flag the take.** Opinions are prefixed with ownership markers so they can't be mistaken for spec-fact: "I think," "to me," "personally," "in my opinion." The solo authority goes further and cites his own limits: "I'm not an expert on this part, but I'm just saying."

**Evidence, then take.** The invariable order is demonstration → judgment, never judgment → justification. The multi-host workshop benchmarks a big-box-store prebuilt PC against a control build *before* condemning it. The paper analyst quotes the primary source on screen, then rules on it.

**Hedging patterns.** Three tiers:
- *Future-proof hedges*: "we'll see," "a lot of wait and see," "could be true. Ask me again in a year" (the solo authority).
- *Sample-size hedges*: "we can only review the unit that actually showed up" (the multi-host workshop); "I can't promise every unit will do this" (the solo authority).
- *Epistemic hedges* (the paper analyst's specialty): "I haven't been able to verify that number myself"; "I am not claiming to have solved the puzzle." He states his priors and updates them on camera — "yes, I was skeptical. Then I read the paper."

**How negativity is delivered.**
- The solo authority: consequence-framing, not insult — the flaw is shown hurting a real user: "what kind of memory is that for the people around you?" Plus preemptive concession to haters: "I totally get it."
- The multi-host workshop: comedic outrage at *objective* failures ("It's a rename button. It renames things!"), always balanced with a fairness beat: "I will give them credit," "in fairness." The target is the decision, not the audience who bought it.
- The showman: disappointment as personal betrayal — "the most expensive paperweight I've ever bought. And I'm genuinely upset." Scores make the negativity legible and bounded.
- The code satirist: irony does the work; the negative claim arrives dressed as a compliment ("bravely lawyered their own community into silence").
- The paper analyst: negativity aimed at *claims*, not products — "engineered to shock and awe the audience" — and always paired with the counter-evidence that earned it.

**The empathy escape valve.** Strong negative verdicts end with who it's still fine for, or what would fix it — "not a real replacement… but genuinely good for people who don't have access to the real thing" (the showman, on a robot personal trainer).

---

## 6. Density technique (the code satirist / the paper analyst)

**The code satirist — comedy density.** Measured on one representative script (848 words / 3:37): roughly one joke per 25–35 words, ~6–8 laughs/minute, while still landing ~15 factual beats. Mechanisms:

- *Load-bearing jokes*: the gag carries the information. "The chipmaker's stock is having a genuinely terrible morning" is both the joke register and the market fact. Nothing is a pure aside.
- *Compression by allusion*: proper nouns and memes replace explanations — "a moon-landing moment," "the sequel nobody asked for," "faking the marathon by taking the subway." One reference = one skipped paragraph.
- *Running gags as structure*: the date sting, the giveaway ritual, the fixed sign-off line, the recurring malapropism ("the ball is in their yard"). Rituals lower cognitive cost — the viewer always knows where they are.
- *Sarcasm-as-topic-sentence*: "obviously you should change nothing and just run it again" states the relatable sin, then the real content follows.
- *List scaffolds*: 10/25/100 items mean the script never needs transitions longer than "next up." Ordering inside the scaffold is thematic tiers (depth levels), each tier reset with a one-line theme statement.
- *Absurdist codas*: endings swerve philosophical or silly ("the real takeaway here… is to drink more water") — a decompression beat after sustained density.

**The paper analyst — evidence density.** Fewer jokes; the density is claims-per-minute with receipts. Mechanisms:

- *Numbered revelations* ("14 Details," "15 Revelations") — each beat is self-contained, skippable, countable.
- *Labor-proof compression*: "I've read all 60 pages… I've got 58 notes" buys the right to summarize brutally.
- *Order by impact with one flagged exception*: "ranked by impact — except this one, which cut the line because it stunned me" — a rule broken visibly is itself a retention device.
- *Caveat interleaving*: every big claim is chased by its limit within a sentence or two, so pace never stalls for a separate "limitations" section ("yes, I'll give the caveats I always do").
- *Followability at speed*: heavy signposting ("first… next… finally," "two final points"), one idea per sentence, and recurring anchor metaphors (walls, paradigms, benchmarks-as-victims: "pour one out for that benchmark").

Shared rule: at 190–235 wpm, *syntax must be flat*. No nested clauses; new information rides subject-verb-object sentences; anything complex gets an analogy within five seconds ("every app is basically a lasagna of other people's code," the code satirist).

---

## 7. Sub-styles — and what transfers

**The solo authority.** Calm, evidence-first, single trusted voice; drama comes from the product, not the presenter. Signature moves: curiosity-order structure, named deferrals, self-coined shorthand concepts, verdict-with-buyer-persona. *Transfers:* the question-sequenced outline and the "most impressive thing first, flagged as such" move work in any host format.

**The showman.** The presenter's money, risk, and emotions are the plot. Signature moves: stunt ledger hooks, per-item scoring, escalation sorts, themed sub-asks, the "golden moment" — he states the doctrine himself: "every video needs one clear golden moment that pays the viewer back for clicking." *Transfers:* announced sorting rules and per-item scores; the two-number hook (money + time) transfers to any "we built/tested X" video.

**The multi-host workshop.** Dialogue creates dramatic irony: one host knows the twist ("You already peeked inside the case… Don't spoil it."), off-screen voices answer price questions, banter absorbs failures. Verdicts are argued, not announced. *Transfers:* the interrogation gag (host asks, voice answers) is a cheap way to serialize reveals even in AI-host builds; staged foreknowledge creates suspense without extra footage.

**The tool curator.** Enthusiast-peer register ("I'm here for it"), value = aggregation and live demos; every item demoed on screen with "let's go ahead and…" narration; weak on verdicts, strong on volume. *Transfers:* the payoff-montage hook and demo-narration register for tutorial content.

**The paper analyst.** Value = reading what others won't and calibrated judgment. Personal brand is epistemic hygiene. *Transfers:* the labor-proof hook ("I read all N pages so you don't have to") and caveat interleaving elevate any research-driven script.

**Category-wide constant:** whatever the style, the presenter performs *work the viewer can't or won't do* — spending, breaking, benchmarking, or reading — and reports back in first person. That's the genre's core contract.

---

## 8. Ending & outro handling

Monetization patterns are intentionally omitted: these skills target sponsor-free, at-scale production. What carries over is the genre's ending craft:

- The last beat is always identity — a fixed, personality-stamped sign-off ritual — or a forward hook; the video ends within seconds of the verdict.
- The outro runs under ~20 seconds and may contain at most one plain subscribe/next-video line, tied to a *named* future payoff ("get subscribed to see the full review of the truck when it happens") — or nothing at all.
- CTAs never interrupt a running test, demo, or reveal; anything beyond the content lives at section seams or the very end.

---

## 9. Anti-patterns (what these scripts never do)

1. **Spec-sheet order.** No script walks a spec table top-to-bottom; specs appear only when the narrative reaches the part they describe, or as a fast front-loaded block explicitly framed as housekeeping.
2. **Cold non-content opens.** No video leads with housekeeping or announcements; nothing precedes the hook.
3. **Unowned opinions.** No "some say" / "many believe" without a named source; vague attribution is replaced by "I tested / I read / I bought."
4. **Free-floating superlatives.** "Best ever" never appears without a comparison class or an on-screen demonstration.
5. **Suspenseless verdict dumping.** The full conclusion is never given at 0:00 *unless* the mystery is immediately re-established at a deeper level (the solo authority's verdict-first hooks are always followed by a "why" thesis).
6. **Corporate-voice hype.** Press-release language appears only in quotes, usually to be audited or mocked — the multi-host workshop reads a vendor's tagline aloud, then: "which is kind of a lot of qualifiers."
7. **Apologizing for format or padding.** No "sorry this is long"; length is justified by promised density ("Buckle up, this is gonna be a mega video").
8. **Untested claims passed through.** Manufacturer claims trigger a test or an explicit "I hope that's true" hedge — never silent acceptance.
9. **Generic mid-roll CTAs.** Sub/like asks are themed to the moment or skipped; boilerplate "smash that like button" language is absent from every top performer.
10. **Ending on anything but identity.** The last beat is always identity (sign-off ritual) or a forward hook.

---

## 10. Formula card (standalone recipe)

**HOOK — pick one, deliver in ≤3 sentences, ≤15 seconds:**
- 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-tease: "[One-line verdict]. The real question is [deeper why]."
- Disaster open: "[Show failure live]. I did everything right. So what's going on?"
- Forbidden object: "[Gatekeeper] refused. I got one anyway."
- Countdown bomb: "[N] worst [things] — and #1 will [visceral promise]."
Rule: hook must contain ≥1 number and a first-person verb of effort (spent / read / bought / broke).

**BEAT TEMPLATE (single product, 100% = runtime):**
- 0–5% hook + credential + roadmap ("here's everything you need to know")
- 5–12% what-is-it frame; state price now if price is news, else hold it
- 12–70% walkthrough in *curiosity order* or *physical-tour order* — never spec order; pull the single best thing forward and say it's the best thing; plant 3+ named deferrals ("we'll get to the battery")
- 70–85% downsides, numbered
- 85–95% verdict: bottom line + who should buy + honesty hedge ("this isn't the full review — I haven't [X]")
- 95–100% themed CTA + named next video

**BEAT TEMPLATE (list/gauntlet):** announce the sort rule in sentence two (price ↑, quality ↓, countdown to #1) → loop [setup → live demo → score/verdict] → running tally → aggregate verdict + total reveal.

**RETENTION RULES:** one named deferral per 2 minutes; 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 ("spoiler: one of these breaks the scale").

**RHYTHM & REGISTER:** first person, present tense, contractions mandatory. Talk-pace target: 160–200 wpm on-camera/demo formats, 200–235 wpm VO-dense formats. Flat syntax at speed — one idea per sentence, analogy within 5 seconds of any hard concept. Specs only as translations ("that means…"), multiples ("30,000× the original model"), or on-screen overlays.

**OPINION RULES:** demo before judgment; prefix takes with "I think / to me"; every negative gets a fairness beat ("in fairness…") and an escape valve (who it's still for / what fixes it); every hype claim gets a comparison class or an audit.

**ENDING/CTA:** nothing before the hook, nothing inside a running test; subscribe asks themed to the moment and tied to a named upcoming payoff, at most one per video; end on your sign-off ritual within seconds of the verdict.

**THE CONTRACT (why any of this works):** do work the viewer can't or won't do — spend, break, benchmark, or read — then report back in first person, skeptical by default, excited only when the product forces you to change your mind on camera.