Every spoken word transcribed. Every caption read. Every claim checked against live sources. The mechanics underneath, written down so you can run them tonight.
Drop your email and the whole file opens right here on this page. A copy of everything lands in your inbox too.
free · delivered on this page the second you hit the button
Best for: making an invisible cost or waste feel physical, then giving away the free fix.
Invented meter UI. Borrowed-authority footage. Star-counter tick-up. Icon-per-second capability montage. One rule for the honest version: the meter must show true numbers.
Best for: authority through curation. Each item is a name, one-line benefit, and a quantified on-screen receipt.
Cream chapter cards with serif ordinals. README highlight swipes. Star-history charts. Ending the list on the most authoritative item.
Best for: proof of capability plus a lead magnet. The reel demonstrates the exact thing it gives away.
Progress bar across the demo (tells the viewer the walkthrough is finite). Disembodied pointing-hand cutout. Guide-page flash at one page per second.
Best for: "most people don't realize you can do X" education that gates the setup, not the knowledge.
The escalation midpoint. Hand-drawn annotation arrows over a collage. A logo-chain animation that explains the integration with zero words.
Best for: replacing a known expensive service with one install. The most sales-native shape in the corpus.
"It does X" anaphora. One-claim-one-receipt editing. The twist differentiator. Only the ask changes pill color. Warning from the audit: this shape sells hardest and lies easiest. Your version has to survive the claim check.
Best for: cold-audience beginners. A negative-command hook plus the safety of an official source.
Receipts rhythm: repo cards land word-for-word with the voice. The deliberate slow-down to full frame for the trust line. Lighting dims onto the keyword.
Best for: speed-to-trend. Riding a fresh release with a price-anchor payoff.
Blur-then-reveal withhold. The pricing-page receipt. The "not a typo" credibility beat. Pixel-glitch CTA slab.
Best for: the strongest cold-audience hook shape in the corpus. Curiosity plus relatability in 4 seconds.
Filmed-monitor authenticity (webcam bezel and glare visible). The empty search result as the hook's proof. The creator's own analytics as social proof inside the demo.
Best for: "one connector and one prompt" simplicity claims. The finished artifact is visible behind you from second zero.
Proof in the mise-en-scene. Time-boxed promise. Handheld pans across the payoff. Chapter change via background swap.
Best for: saves, dwell time, and comment velocity. Not a reel: 10 individually rendered looping video slides uploaded as a carousel.
| the job | call |
|---|---|
| Cold-audience trust, lowest production | ONE SMALL THING |
| Make an invisible cost hurt | THE METER |
| Curation authority | TOP N STACK |
| Prove you can build | BUILD LOG · 30-SECOND BUILD |
| Tease a capability, gate the setup | HIDDEN COMMAND |
| Sell against an expensive alternative | AGENCY KILLER |
| Ride a fresh release | NEWS DROP |
| Maximum cold hook strength | FAKE FAIL |
| Saves and comment velocity | CAROUSEL ENGINE |
The tool and the per-page token cost are real. But "a 20 page document burns up to 70,000 tokens" fails its own math (20 x 3,000 = 60,000), and the reel never tells you scanned pages and charts should stay PDFs. Nobody in 90 thousand comments checked.
"Cutting your usage by over 50% without losing any accuracy": the repo's own stats say about half the code and about 20% cheaper. One terminal command shown on screen does not exist. And the official directory's own trust warning never comes up.
The differentiator it sells is a reciprocal backlink exchange, which Google's spam policy names directly. The "when someone asks ChatGPT, you're the answer" proof was a prompt phrased in the product's own category language.
"A dashboard that auto updates" is really "rebuilds when you ask." The revenue and ROAS numbers shown can't be audited from the reel. And every metric shown except payment-processor revenue is vanity-adjacent.
"70 times fewer tokens, that's not a typo." Their own benchmark table: 71.5x on a 52-file project, 5.4x on four files, roughly nothing on six. Big codebase, monster. Small project, skip it. That is the difference between a claim and a benchmark.
Same pattern everywhere: a real tool at the core, stretched at the edges. "Just two free tools" hides the paid plan one of them needs. "The source code for Claude Code itself" is not open source. Staged batch screens repeat the same URL four times.
Every reel in the corpus is true at the core and exaggerated at the edges. None of them lie about existence; all of them lie about ease. Which means the account that keeps the identical viral structure but makes the edges true (current numbers, honest caveats in one clause, receipts that survive scrutiny) loses nothing on hook strength and compounds the one thing none of them can copy without breaking their back catalog: trust.
The next step after your reel runs: a free first agent that runs in the Claude you already pay for.
get your first agent →