# UGC SCRIPT WRITING SYSTEM — VERSION 2

*Universal Framework for AI-Generated UGC Scripts*

Niche-Agnostic • Brand-Agnostic • Production-Ready

Built for AI video generation tools and chunked production workflows

## Contents

1. Two Script Formats
2. Non-Negotiable Principles
3. The Universal Script Structure
4. Universal Direction Blocks
5. Voiceover Architecture
6. Variation Types
7. Pain Point Selection
8. Camera and B-Roll Direction Standards
9. Common Failure Modes
10. Brand Input Variables
11. Chunked Production Workflow
12. Chunk Breakdown Templates by Script Length
13. Captions for Organic Reach
14. Testing and Scaling Logic
15. How to Use This Document

## 1. Two Script Formats

The system produces two formats. Pick the one that matches the funnel stage.

| Attribute | Full Stack Format | Mid-Funnel Punchy |
| --- | --- | --- |
| Runtime | 28–32 seconds | 18–22 seconds |
| Word count | 150–180 words | 55–70 words |
| Beat structure | Five-beat: hook, reframe, mechanism, payoff, CTA | Three-beat: sharp hook with reframe folded in, mechanism with analogy, soft payoff with close |
| Default scene count | 3 scenes | 2 scenes (3 only if voiceover requires it) |
| Default chunk count (chunked workflow) | 5 chunks | 3 chunks |
| Use for | Cold audiences who need the full education arc to convert | Warmer audiences who already know they have the problem and need a fast credibility signal |

## 2. Non-Negotiable Principles

These rules apply to every script, every format, every niche. They are not preferences. They are the core ruleset that separates a winning UGC script from a losing one.

### 2.1 The Hook Owns the First Five Words

The opening line identifies the exact person being spoken to and confirms the specific frustration they are carrying. Never narrates the creator's morning, weekend, or life. Calls out the right person and repels everyone else.

### 2.2 Sentence Flow Over Choppy Phrasing

No two-word stutter patterns. Sentences connect, breathe, and lead into each other. Use connectors and complete thoughts.

### 2.3 Keep the Mechanism Simple

Name the proprietary mechanism the way the brand names it, but only if a 12-year-old would understand. Always pair the mechanism with a one-second tactile analogy the viewer can picture. The analogy is the most important sentence in the entire script — it is what the viewer repeats to a friend later.

### 2.4 Soft CTAs Outperform Direct Sales Pitches

The locked competitor formula works for cold traffic but reads as ad-coded for warmer audiences. Default to suggestion-style closes.

### 2.5 No Em Dashes in Voiceover. No Bolding in Voiceover.

Em dashes break natural speech rhythm when read aloud. Bolding has no function in audio. Use commas, periods, and short connectors.

### 2.6 Avoid Words AI Voiceover Tools Mispronounce

Common offenders: collagen, hyaluronic, niacinamide, keratosis, brand names with non-standard spelling. Either replace with plain-language alternatives or include a phonetic pronunciation note above the voiceover. When in doubt, swap the word.

### 2.7 Brand Pronunciation Lock (When Brand Is Spoken)

If the brand name appears in voiceover, include a pronunciation block: exact spelling, dictionary phonetics, and instruction that no shortcuts or variations are permitted. If the brand is consistently mispronounced even with phonetics, remove brand mentions from voiceover.

### 2.2 Example: Sentence Flow

:::box WRONG RHYTHM
"No chemicals. No scents. No replacing."
:::

:::box RIGHT RHYTHM
"There are no chemicals, no scents, and nothing to replace ever again."
:::

### 2.4 Soft CTA Bank

- "I'll leave a link to the one I'm using below."
- "I'll drop a link below if you want to look into it."
- "Linking it below so you don't have to go searching for it."
- "Leaving a link in the comments / bio so you can check it out yourself."

## 3. The Universal Script Structure

Every script output follows the same skeleton. Variables fill in. Sections never reorder. No meta-commentary about process.

### 3.1 Standard Production Skeleton

Use this when generating a single video as one continuous prompt.

```
Create a [STYLE] ad for [PRODUCT CATEGORY]. The creator is a [GENDER] in their [AGE RANGE], [VIBE — 2 to 3 words], [HAIR DESCRIPTION], wearing [OUTFIT]. Natural look, [MAKEUP LEVEL]. They're in [SETTING WITH LIGHTING]. [SKIN DIRECTION BLOCK] [APPLICATION DIRECTION BLOCK] @Image 1 is the product, [PRODUCT NAME — what it is]. Use it as a reference for what the [PACKAGE TYPE] looks like when held or interacted with, not as a static image. [B-ROLL SEQUENCING BLOCK] [UGC REALISM DIRECTION BLOCK] Camera style: [CAMERA DIRECTION]. [B-ROLL DIRECTION WITH EXPLICIT NO-PRODUCT OPENING CUTS] [SCENE COUNT AND PACING NOTE] Here's the full script: "[VOICEOVER SCRIPT]"
```

### 3.2 Chunked Production Skeleton

Use this when generating clips in chunks for post-production stitching using the same model and character across multiple generations. See Section 12 for full chunk breakdown templates.

```
[CHARACTER LOCK — paste into every chunk]
[UNIVERSAL DIRECTION BLOCKS — paste into every chunk]
  • Skin Direction Block
  • Application Direction Block
  • B-Roll Sequencing Block
  • UGC Realism Direction Block
[CHUNK-BY-CHUNK BREAKDOWN]
For each line of voiceover:
  • Line number and exact voiceover text
  • Estimated runtime in seconds
  • Scene-specific visual direction
  • Continuity notes
  • Whether @[product asset] is included or not
```

## 4. Universal Direction Blocks

These four blocks lock in production quality and prevent the most common AI generation failures. Paste them into every script. Adapt the variables to the specific product.

### 4.1 Skin Direction Block

Prevents the AI from generating the "before" condition on the present-tense creator who is supposed to be showing the result.

:::box SKIN DIRECTION BLOCK — paste verbatim
*Important skin direction: The creator has naturally beautiful, smooth, clear skin with an even tone and a soft healthy glow. Her complexion is clean and radiant throughout the entire video — no visible [CONDITION THE PRODUCT TREATS] at any point. She has the kind of skin that looks like she has already been using the product for months, because she has. This applies to every talking-to-camera scene, every product application scene, and every close-up.*
:::

**Adaptation for Other Categories**

- Hair: "The creator has naturally beautiful, healthy, full hair with no visible [thinning/breakage/frizz] at any point."
- Teeth: "The creator has naturally bright, white, healthy teeth with no visible [staining/yellowing] at any point."
- Home/cleaning: "The creator's home is clean, lived-in, and beautiful — no visible [stains/clutter/damage] at any point."

*Principle: lock the creator visually into the after condition the viewer is being shown the path to.*

### 4.2 Application Direction Block

Prevents the AI from generating product as a visible matte or pigmented mark when it should be invisible, transparent, or behave differently.

:::box APPLICATION DIRECTION BLOCK — paste verbatim
*Important application direction: When the creator [USES THE PRODUCT], it should [DESIRED VISUAL BEHAVIOR]. No visible [SPECIFIC FAILURE MODES THE AI DEFAULTS TO]. [WHAT IT SHOULD LOOK LIKE INSTEAD]. Her [TARGET AREA] should look exactly the same after [APPLICATION/USE] as it did before — [DESCRIPTION] — with only [REALISTIC RESIDUAL EFFECT] where the product touched.*
:::

*Principle: name the failure mode the AI defaults to and override it with positive description plus a clear visual rule.*

### 4.3 B-Roll Sequencing Block

Prevents the AI from showing the product before the voiceover earns it, which kills the hook by flipping the viewer into "this is an ad" mode too early.

:::box B-ROLL SEQUENCING BLOCK — paste verbatim
*Important b-roll sequencing: The product must not appear on screen until the voiceover specifically introduces it. During the hook and reframe beats, the camera stays on the creator. No environmental cutaways, no decorative shots of the apartment, no shots of furniture or art. The product is only revealed visually at the exact moment the voiceover names it.*
:::

*Principle: tie product visibility to a specific lyric in the voiceover, not vague timing. In chunked production, simply omit the @[product asset] reference from prompts where the product should not appear — this is a hard lock the AI cannot override.*

### 4.4 UGC Realism Direction Block

Prevents the AI from generating posed, frozen, commercial-feeling shots. Forces the natural body language of someone casually filming themselves.

:::box UGC REALISM DIRECTION BLOCK — paste verbatim
*Important UGC realism direction: This is a casually filmed video. The phone is propped somewhere off-camera, so both her hands are free throughout. Natural handheld jitter and small micro-movements give it a self-filmed feel. She gestures naturally with both hands as she talks — adjusting her hair, touching her face when relevant, doing small open-palm gestures, shifting weight between her feet. She is never frozen in a still pose with her hands in her pockets or behind her back. The energy is "I just want to tell you something" not "I am posing for a commercial."*
:::

*Principle: name the failure mode (frozen pose, hands in pockets) and replace it with specific natural behaviors.*

## 5. Voiceover Architecture

### 5.1 Full Stack Five-Beat Structure

| Beat | Timing | Function |
| --- | --- | --- |
| 1. Hook | 0:00–0:03 | Calls out exact person, confirms specific frustration. Two sentences. |
| 2. Problem Reframe | 0:03–0:10 | Reveals the hidden mechanism behind why the viewer's previous attempts failed. Reorients from "I am doing it wrong" to "I was given the wrong tools." |
| 3. Mechanism + Analogy | 0:10–0:20 | Introduces the product, explains the mechanism in plain language with a one-second tactile analogy. |
| 4. Payoff | 0:20–0:25 | Sensory and specific lived experience after the product. Visual and immediate, not abstract benefits. |
| 5. CTA | 0:25–0:30 | Soft suggestion-style close. |

### 5.2 Mid-Funnel Three-Beat Structure

| Beat | Timing | Function |
| --- | --- | --- |
| 1. Sharp Hook with Reframe Folded In | 0:00–0:06 | One sentence that calls out the audience and dismisses the wrong assumption in the same breath. |
| 2. Mechanism with Analogy | 0:06–0:16 | Compressed mechanism plus the tactile analogy. The analogy stays intact at all costs. |
| 3. Soft Payoff and Close | 0:16–0:21 | One sensory beat plus the suggestion-style CTA. |

### 5.3 Hook Frameworks

Pull from this list. Rotate across scripts to prevent fatigue.

- If you are trying to ___ this is how you finally do it without ___.
- Your ___ is ___, and ___ won't fix it.
- The biggest myth about ___ is...
- I tested ___ so you don't have to.
- I wish someone had told me this before I started ___.
- I found this out by accident while ___.
- What nobody warns you about with ___ is...
- Why ___ works when nothing else does.
- Why your ___ isn't working and how to fix it.
- Don't make this mistake with ___.
- One thing I'll never do again in ___.
- Three mistakes keeping you stuck in ___.
- This one thing changed everything for me in ___.
- What nobody admits about ___ is...
- How I stopped ___.
- The truth behind my ___.
- If I could go back and tell myself one thing about ___.

### 5.4 The Universal Reframe Pattern

Across niches, the strongest reframes follow this shape:

:::box UNIVERSAL REFRAME PATTERN
*"Most people think ___ is a [SURFACE] problem so they keep [SURFACE BEHAVIOR]. But it is actually a [STRUCTURAL/HIDDEN] problem, which is why [SURFACE SOLUTION] physically cannot reach it."*
:::

This works because it absolves the viewer of guilt for past failures and sets up the mechanism reveal. Adapt the structural-vs-surface framing to whatever the product addresses.

## 6. Variation Types

The same script structure supports multiple emotional registers. Rotate across the campaign.

| Type | Voice | Best For |
| --- | --- | --- |
| The Confessional | First-person, present tense, includes a personal admission | Emotional categories like beauty, wellness, parenting |
| The Tested It So You Don't Have To | Audience surrogate who tried every wrong solution first | Crowded categories where the audience has tried multiple competing products |
| The Myth Buster | Opens with a corrective reframe | Educational categories where the audience has absorbed wrong information |
| The Accidental Discovery | Creator stumbled onto the insight | Lowest sales pressure framing — positions creator as fellow explorer |
| The Animated Infomercial | No creator, pure product b-roll with voiceover | Scaling output without booking talent, or when the product is the hero |

## 7. Pain Point Selection

The first scripts in any campaign should target the surface complaint (visible symptom). Once those are running, expand into deeper pain points. Mass-desire pain points outperform niche ones.

### 7.1 Surface Complaints

The literal, visible symptom of the problem. Easy entry point but saturated.

### 7.2 Deep Pain Points

High-leverage angles that go beyond the surface symptom.

- **Public visibility pain —** being seen in unforgiving lighting (photos, video calls, daylight)
- **The mental tax —** daily labor of working around the problem
- **The aging or time pain —** carrying the problem for years/decades, the strange grief of unresolved issues
- **The "I have fixed everything else" pain —** the existential frustration of someone who solves problems and this is the one that has not budged
- **Intimacy and close-up pain —** how the problem feels in private moments, with people who look closely
- **The recognition pain —** not recognizing yourself, the gap between how you feel and how you appear

### 7.3 Mass-Desire Filter

The best pain points cut across age, gender, profession, and lifestyle. Test for universality. If the pain point only applies to office workers, parents, or one specific demographic, it narrows reach. If almost anyone with the problem has felt it, it is mass-desire.

## 8. Camera and B-Roll Direction Standards

### 8.1 Constant Motion Rule

Every shot has movement. No locked-off product photography. No still talking-head shots. Talking-to-camera footage uses natural handheld jitter. Hands-free shots use subtle drift or micro-push-ins. B-roll uses tracking, dolly pushes, orbits, reveals, or hand motion.

### 8.2 Visual-to-Voiceover Sync

Every visual matches what the narrator is saying at that exact moment. When the narrator says "fridge" show a fridge. When they describe the mechanism show the mechanism. Visuals follow the voiceover word for word.

### 8.3 Continuous Cuts — Cuts Are Motivated, Not Decorative

Cuts happen when:

- The creator physically moves to a new location to do something
- The voiceover names the product (cut to product reveal)
- A specific action happens that the script calls for

Cuts do NOT happen for:

- Decorative environmental shots of furniture, art, or interior design
- "Visual breaths" that have no relationship to what is being said
- Aesthetic moments that exist for cinematography reasons

### 8.4 Pacing Within Scenes

Scenes play at normal speed. No slow motion. No sped up footage. Within each scene, micro-cuts can be rapid for high-energy beats and slightly slower for emotional or explanatory beats.

### 8.5 Scene Count Discipline

- Mid-funnel format: 2 scenes default, 3 only if voiceover requires it
- Full stack format: 3 scenes default
- Animated infomercial: AI determines breaks based on voiceover beats

### 8.6 Setting Variation Across Campaign

Rotate environments and lighting setups across scripts so the audience does not see the same backdrop on repeat. Options: warm dim kitchen evening, bright sun-drenched morning, soft afternoon side window, cool clinical product studio, lived-in family kitchen, sunlit bedroom vanity, back patio outdoor light, evening warm bathroom. Tie each variation to the emotional register of that script.

### 8.7 Demographic Variation Across Campaign

Rotate creator demographics so the algorithm has signal on which audience segments respond. Standard spread: ages 19-23, 24-32, 32-38, 35-45, plus a male equivalent for at least one variation. Each demographic carries different emotional pressure points.

## 9. Common Failure Modes

These are recurring AI generation failures. Self-correct against them on every output.

:::fail RAMBLING HOOKS
"I scrubbed my fridge on Saturday on my hands and knees" is a failed hook. The first words call out the audience, not narrate the creator's life.
:::

:::fail CHOPPY PHRASING
Two-word fragments stacked together kill conversational flow.
:::

:::fail JARGON MECHANISMS
If a 12-year-old does not understand the mechanism word, replace it.
:::

:::fail VEERING CTAS
Use the locked formulas. Do not invent new structures unless testing has earned it.
:::

:::fail STATIC B-ROLL
Every shot moves.
:::

:::fail MISMATCHED VISUALS
If the narrator says one thing and the b-roll shows another, the script fails.
:::

:::fail DECORATIVE B-ROLL
Cuts to furniture, art, or interior design that have no relationship to the voiceover. UGC does not have cinematic establishing shots. Keep the camera on the creator unless there is a specific reason to cut away.
:::

:::fail OVER-FORMATTING THE OUTPUT
The deliverable is a clean prose prompt block. No headers, tables, or production notes attached unless using chunked production format.
:::

:::fail PHONE IN HAND WHEN NOT NEEDED
If the script calls for selfie-style direction, the phone in hand makes sense. Otherwise, leave the phone off-camera so both hands are free for natural gestures.
:::

:::fail FROZEN POSES
Hands in pockets, hands behind back, no movement. Real UGC creators gesture, shift weight, touch their face when relevant.
:::

:::fail PRODUCT VISIBILITY BEFORE VOICEOVER EARNS IT
Showing the product in the hook beat tells the viewer this is an ad before they have decided to keep watching.
:::

:::fail WRONG CONDITION ON THE PRESENT-TENSE CREATOR
Generating visible scars/blemishes/damage on the creator who is supposed to be showing the after.
:::

:::fail VISIBLE PRODUCT RESIDUE WHEN IT SHOULD BE INVISIBLE
Generating a matte stripe or pigmented mark when application should be transparent.
:::

## 10. Brand Input Variables

Before producing any script, collect these or ask for them. Do not invent values.

| Variable | Status | Notes |
| --- | --- | --- |
| Brand name and exact spelling | Required | |
| Brand pronunciation (dictionary phonetics) | Required if spoken | |
| Whether brand name is spoken in voiceover | Required | |
| Product name and exact spelling | Required | |
| Product pronunciation | Required if spoken | |
| Product category | Required | |
| Core mechanism in plain language | Required | Must be 12-year-old understandable |
| Tactile analogy for the mechanism | Required | Format: "Think of it less like ___ and more like ___" |
| Failed alternatives the audience has tried | Required | |
| Target avatar demographics | Required | |
| Specific pain points beyond surface complaint | Required for deep angles | |
| Setting and lighting preferences | Optional | Defaults to lived-in home environment |
| Reference image of the product | Required | Use as @Image 1 or @[product asset] |
| Words AI mispronounces in this category | Optional | Flag in advance to swap out |

## 11. Chunked Production Workflow

For projects where the same creator and setting are used across multiple generated clips that get stitched together in post.

### 11.1 Lock the Character Once

Write a detailed character description that gets pasted verbatim into every chunk. Include age range, energy, hair color and style, skin tone, eyes, accessories, specific outfit, and the apartment or setting they are in.

### 11.2 Lock the Universal Direction Blocks

Skin Direction, Application Direction, B-Roll Sequencing, and UGC Realism Direction all get pasted into every chunk verbatim.

### 11.3 Break the Voiceover Into Lines

Each line of voiceover becomes one chunk. Default chunk counts:

- **Mid-funnel (18-22 sec):** 3 chunks
- **Full stack (28-32 sec):** 5 chunks

Section 12 contains the full chunk breakdown templates for both formats.

### 11.4 Asset Tagging Locks Product Visibility

Use @[product asset name] (e.g., @stick, @bottle, @canister) in the prompts where the product should appear. Do NOT include the asset reference in chunks where the product should not appear. The AI cannot generate the product if the asset is not referenced — this is a hard lock that enforces the b-roll sequencing rule.

### 11.5 Bookend Structure

Make the first chunk and last chunk visually similar (same setting, same framing) so the stitched video feels intentional and any small inconsistencies between middle chunks read as natural cuts rather than continuity errors.

### 11.6 Identify Cut Points and Fallback Cuts

Note which chunks can be dropped from the edit if any single chunk fails to generate cleanly. Usually environmental or transitional chunks are the easiest to remove without breaking the script.

## 12. Chunk Breakdown Templates by Script Length

This section provides the exact chunk structure for each script format. Use these as the operational template when building scripts for chunked production. Each chunk is one separate AI generation that gets stitched together in post.

### 12.1 Per-Chunk Specification Template

Every chunk, regardless of script length, includes this structured set of fields:

| Field | Purpose |
| --- | --- |
| Chunk number and beat label | Identifies which voiceover beat this chunk covers (Hook, Reframe, Mechanism, Payoff, CTA) |
| Voiceover text | The exact line spoken in this chunk, copied verbatim from the full script |
| Estimated runtime (seconds) | How long the AI generation needs to be for this clip |
| @[product asset] inclusion | Whether the product asset reference appears in this chunk's prompt — this controls product visibility |
| Visual direction | Camera angle, gesture timing, what the creator is doing during the line |
| Continuity notes | What must match the previous chunk (outfit, hair, setting, lighting, position) |

### 12.2 Mid-Funnel Format — 3-Chunk Breakdown

**Total runtime target:** 18–22 seconds. **Word count:** 55–70 words. **Chunks:** 3.

This format compresses the five-beat structure into three operational chunks: the hook with reframe folded in, the mechanism reveal with product introduction, and the payoff with soft CTA.

:::box CHUNK 1 — HOOK + REFRAME (folded together)
**Voiceover content:** The opening line that calls out the audience and dismisses the wrong assumption in the same breath.
**Estimated runtime:** 5–7 seconds
**@[product asset] included:** **NO — product must not appear in this chunk**
**Visual direction:** Talking-to-camera shot in primary setting. Both hands free. Natural gestures on key words. Camera stays on the creator the entire time. No environmental cutaways. No product visible.
**Continuity notes:** Establishes baseline — outfit, hair, lighting, position. All subsequent chunks reference back to this.
:::

:::box CHUNK 2 — MECHANISM + PRODUCT REVEAL
**Voiceover content:** The compressed mechanism plus the tactile analogy. The product is named here for the first time.
**Estimated runtime:** 8–10 seconds
**@[product asset] included:** **YES — appears at the exact moment the voiceover names it**
**Visual direction:** Cut to a new angle motivated by physical movement (creator walks to a counter or surface). Phone propped, framing steadier. Quick application sequence: pick up product, twist open, glide across target area, close-up of result. Application leaves no visible residue per the application direction block.
**Continuity notes:** Same outfit, hair, lighting as Chunk 1. Location can shift if motivated by movement. Product matches reference image exactly.
:::

:::box CHUNK 3 — PAYOFF + SOFT CTA
**Voiceover content:** One sensory beat describing the lived experience after the product, plus the suggestion-style close.
**Estimated runtime:** 5–6 seconds
**@[product asset] included:** **NO — closing on the creator's face, not the product**
**Visual direction:** Return to talking-to-camera shot, ideally bookending the framing of Chunk 1. Soft genuine smile on the payoff line. Subtle downward gesture on the CTA. Both hands free.
**Continuity notes:** Match Chunk 1 framing as closely as possible to create the bookend. Same outfit, hair, lighting.
:::

**Stitching Notes — Mid-Funnel**

- Natural seam between Chunk 1 and Chunk 2 — that is where the no-product hook section transitions to the product reveal section.
- If a chunk fails to generate cleanly, Chunk 2 is the hardest to drop because it carries the product reveal. Chunks 1 and 3 are easier to regenerate or reshape because they are bookend talking-to-camera shots.

### 12.3 Full Stack Format — 5-Chunk Breakdown

**Total runtime target:** 28–32 seconds. **Word count:** 150–180 words. **Chunks:** 5.

This format gives each beat of the five-beat structure its own dedicated chunk, which is what the longer runtime supports. Each chunk handles one voiceover beat.

:::box CHUNK 1 — HOOK
**Voiceover content:** Two sentences that call out the exact person and confirm the specific frustration.
**Estimated runtime:** 5–6 seconds
**@[product asset] included:** **NO**
**Visual direction:** Talking-to-camera shot in primary setting. Both hands free, natural gestures. Steady eye contact. No cuts, no environmental shots. Background is just the lived-in setting behind her, not the focus.
**Continuity notes:** Establishes baseline for the entire video. Lock outfit, hair, lighting, position.
:::

:::box CHUNK 2 — REFRAME PART 1
**Voiceover content:** Opens the reframe by naming the wrong things the audience has tried. Sets up the structural-vs-surface distinction.
**Estimated runtime:** 5–6 seconds
**@[product asset] included:** **NO**
**Visual direction:** Same talking-to-camera shot as Chunk 1, with a slight angle shift or subtle camera drift to keep the visuals fresh. Dismissive hand gesture on the list of failed alternatives. Brief unconscious cheek-touch or self-reference if relevant to the product category.
**Continuity notes:** Same setting as Chunk 1. Same outfit, hair, lighting. Camera move should feel like a natural reframe within one take, not a hard cut to a new location.
:::

:::box CHUNK 3 — REFRAME PART 2 (the structural truth)
**Voiceover content:** The mechanism reveal: explains why surface attempts physically cannot work, setting up the product as the only viable option.
**Estimated runtime:** 5–6 seconds
**@[product asset] included:** **NO — product still has not appeared**
**Visual direction:** Continuous talking-to-camera. Optional brief gesture indicating the affected area on the creator's body. Dismissive flick on the failed alternatives. No environmental cutaways. No product visibility.
**Continuity notes:** Same setting, same outfit, same lighting. This is the final beat before the product is introduced — the hook tension peaks here.
:::

:::box CHUNK 4 — MECHANISM + PRODUCT REVEAL + APPLICATION
**Voiceover content:** The product is named, the mechanism is explained in plain language, and the tactile analogy lands here. The most important sentence in the entire script.
**Estimated runtime:** 8–10 seconds
**@[product asset] included:** **YES — first appearance, at the exact moment the voiceover names it**
**Visual direction:** Cut to a new angle motivated by physical movement (creator walks to a counter or surface). Phone propped on the counter, framing steadier with subtle drift. Quick application sequence: tight shot of product in hand, twist open on the word the product is named, macro shot of application across target area, push-in close-up showing the immediate result. Application leaves no visible residue per the application direction block.
**Continuity notes:** Location shift is motivated and reads as natural movement. Same outfit, hair, lighting. Product matches the reference image exactly.
:::

:::box CHUNK 5 — PAYOFF + SOFT CTA
**Voiceover content:** Sensory and specific description of life after the product, followed by the suggestion-style close.
**Estimated runtime:** 5–6 seconds
**@[product asset] included:** **NO — closing on the creator's face, not the product**
**Visual direction:** Return to talking-to-camera shot, ideally bookending the framing of Chunk 1. Soft genuine smile on the payoff line. Subtle downward gesture on the CTA. Both hands free, no posing.
**Continuity notes:** Match Chunk 1 framing as closely as possible to create the bookend. Same outfit, hair, lighting. The smile should feel earned, not staged.
:::

**Stitching Notes — Full Stack**

- Natural seam between Chunk 3 and Chunk 4 — that is where the no-product reframe section transitions to the product reveal section.
- Chunks 1, 2, 3, and 5 are all talking-to-camera in the primary setting. If any of these generate slightly off, the bookend structure (Chunks 1 and 5 matching) masks small inconsistencies because viewers register the visual symmetry as intentional.
- Chunk 3 is the easiest to drop or compress if any single chunk fails to generate cleanly. The voiceover line in Chunk 3 can be folded onto the end of Chunk 2 with the visual just being a continued talking-to-camera shot. Keep that as your fallback.
- Chunk 4 is the hardest to regenerate because it contains the product reveal and application. Plan extra generation time for this chunk.

### 12.4 Animated Infomercial Format — Variable Chunks

The animated infomercial uses no creator and is pure product cinematography with voiceover. Because there is no continuity requirement on a human creator across chunks, the chunk count flexes based on script length:

- **Mid-funnel runtime:** 3–4 chunks aligned to voiceover beats
- **Full stack runtime:** 5–6 chunks aligned to voiceover beats

Each chunk in this format covers a distinct visual concept rather than a continuity-locked scene: orbit shot of the product, animated cross-section showing the mechanism, macro application footage, hero shot at the close, etc. Because there is no creator continuity to maintain, the chunks have more visual freedom and the AI can generate each one with different lighting and angle without breaking the video.

**Animated Infomercial Per-Chunk Field Adjustments**

- Replace "Visual direction" with "Cinematography direction" — orbit, dolly, macro, animated cutaway, hero shot.
- Replace "Continuity notes" with "Aesthetic continuity" — what visual style stays consistent (lighting temperature, color grading, surface materials).
- @[product asset] inclusion is YES on every chunk because the product is the hero of every shot.

### 12.5 Quick Reference — Chunk Count by Format

| Format | Runtime | Chunks | Product Asset Pattern |
| --- | --- | --- | --- |
| Mid-Funnel Punchy | 18–22 sec | 3 chunks | NO, YES, NO |
| Full Stack | 28–32 sec | 5 chunks | NO, NO, NO, YES, NO |
| Animated Infomercial (mid) | 18–22 sec | 3–4 chunks | YES on all chunks |
| Animated Infomercial (full) | 28–32 sec | 5–6 chunks | YES on all chunks |

### 12.6 Workflow Summary

To produce a chunked script from scratch:

- **Step 1.** Decide format (mid-funnel or full stack) based on funnel stage.
- **Step 2.** Write the full voiceover script using the beat structure from Section 5.
- **Step 3.** Lock the character description and the four universal direction blocks.
- **Step 4.** Break the voiceover into the chunk count for that format (3 or 5).
- **Step 5.** Fill in the per-chunk specification template for each chunk.
- **Step 6.** Generate each chunk separately, regenerating any that fail to land cleanly.
- **Step 7.** Stitch chunks in post, prioritizing the bookend match between Chunk 1 and the final chunk.

## 13. Captions for Organic Reach

Captions on the video post matter as much as the script for reach. Native-feeling captions outperform branded copy.

### 13.1 Caption Principles

- Lowercase, no over-punctuation, conversational tone
- One soft emoji maximum, optional
- No hashtags unless testing shows they help on the platform
- 1–2 short sentences
- Does not directly sell — alludes to the topic of the video and lets the video do the work

### 13.2 Caption Templates by Variation

| Variation Type | Template |
| --- | --- |
| Confessional / Accidental Discovery | "nobody told me ___. wish I had figured this out years ago" |
| Tested It | "spent way too much money on ___ before I figured out why none of it worked" |
| Myth Buster | "if you've been ___ thinking it would ___, read this" |
| Universal | "genuinely did not know this until recently. if you ___ this might explain why nothing's been working for you" |

### 13.3 First Comment

Drop the actual link in the first comment of the post. Keep the main caption clean. The first comment can be slightly more direct: "linking what I'm using in my bio if anyone wants to check it out."

## 14. Testing and Scaling Logic

### 14.1 Initial Test Set

Build out a test set of 6–10 scripts before scaling spend. Cover at minimum:

- All five variation types
- 2–3 different demographic profiles
- 2–3 different hook frameworks
- Both formats (mid-funnel and full stack)
- At least one mass-desire deep pain point angle

### 14.2 What to Watch

- Watch time and completion rate, not just CTR
- Cost per click against creative angle
- Comment sentiment — **specifically: are viewers saying the analogy back to each other? That is the signal that the script is working as intended**
- Demographic breakdown of who the algorithm is serving each script to

### 14.3 When to Cut and When to Scale

:::box KILL CRITERIA
• Watch time drops in the first 3 seconds — the hook is not working
• Viewers watch but do not click — the payoff or CTA is weak
:::

:::box SCALE CRITERIA
• Comments quote the analogy or the payoff line back — that is product-language fit
• Once a winner emerges, write hook variations on it before fatigue sets in
:::

### 14.4 Demographic and Setting Variation as Fatigue Defense

Rotate demographics and settings across the campaign. Same script, different creator and setting, can extend a winning angle by weeks before the audience tires of it.

## 15. How to Use This Document

This document deploys in three ways:

- **As a system prompt.** Paste sections 1–13 into the system prompt of any LLM. Provide brand variables from section 10 and the desired hook framework, format, variation type, and pain point. The model produces structured output.
- **As a training reference.** Use as a style guide for human copywriters or contractors. The structure rules apply equally to human-written and AI-written scripts.
- **As an audit checklist.** Use the failure modes in section 9 and the principles in section 2 to grade any script before approving it for production.

*End of Document — Universal UGC Script Writing System v2*