skills/ai-film-pipeline-master/references/phase-01-story-generation/formats-social.md

# Formats — social and short-form

Entry point into the database. Each kit is a ranked short list; open the `cat-*.md` named in the row for the full card.

A format kit answers one question: **given this format's constraints, which techniques earn their place, and why this one rather than the obvious one.** The ranking is the argument.

**Shelf life.** Every kit in this file carries platform convention as well as craft. The craft is durable; the numbers are not. Review dated 2027-09.

**Every length, ratio and size in these kits is a default or a reference, never an argument with the brief.** The brief's own length, aspect and size govern and are written as asked (root `SKILL.md` §0). Facebook and YouTube carry both vertical and horizontal video: a 16:9 feed video takes the same kit at 16:9.

---

<!-- GENERATED:toc — do not edit by hand. Generated in the source package -->

## Contents

- [Vertical Micro-Drama](#vertical-micro-drama)
- [Reels, Shorts and TikTok](#reels-shorts-and-tiktok)
- [Long-Form Video (YouTube)](#long-form-video-youtube)
- [Web Series and Serialised Short-Form](#web-series-and-serialised-short-form)
- [Carousel, Thread and Text Post](#carousel-thread-and-text-post)
- [Podcast Clip](#podcast-clip)
- [Livestream Script](#livestream-script)
- [Branded Serial](#branded-serial)
- [Creator Brand Partnership (Organic)](#creator-brand-partnership-organic)
- [Community Post](#community-post)
- [Instagram Reel](#instagram-reel)
- [TikTok](#tiktok)
- [YouTube Shorts](#youtube-shorts)
- [Coverage](#coverage)
- [Sources](#sources)

<!-- /GENERATED:toc -->

## Vertical Micro-Drama

**Constraints:** 60 to 180 seconds per episode, most near 90, shot and delivered 9:16 for a phone held in one hand. Series run 60 to 100 episodes with a paywall partway through, so the buttons before that wall carry the revenue. Episode length is dictated by **attention and cliffhanger rhythm, not by story content** — this is the inversion that breaks writers arriving from film. Sound is on but captions carry much of the load. Cast is small enough to hold while scrolling.

**Grammar:** Structure first, language second, always. The episode is drafted as a timestamp skeleton before a line of dialogue exists — `0:00` ignition, `0:32` first jolt, `1:18` spike, `1:50` button — and the dialogue is then written to carry those moments, not to create them. Every episode opens mid-situation; nothing is established. Emotional information runs at roughly three times feature density. Any line that neither advances the plot nor sets the button is cut. Exposition arrives as accusation, never as explanation. One value turn per episode — two reads as two episodes badly joined.

**Palette:** 4 beats per episode · 1 value turn · 1 button · 2–4 speaking characters · scenes of 20–45 seconds · no scene without a turn

**Rules:** Write the timestamp skeleton before the dialogue | One value turn per episode, no more | Cut on the question, not the answer, two seconds earlier than feels safe | The last line is never the protagonist's — the antagonist or the world takes it | Open mid-situation, never on arrival | Every scene turns or it is cut | Name the button type before writing: reveal, threat, revelation or decision | Front-load the strongest buttons before the paywall | Humiliation is the engine; reversal is the promise

**Examples:** the Chinese 短剧 catalogue and its US adaptations; ReelShort, DramaBox and Shortmax libraries

**Avoid:** Over-explaining the button — the most common failure, and it comes from not trusting the audience. Feature pacing in the first three episodes, which is where series die. A cliffhanger that stalls: an open loop the next episode does not close quickly is worse than no loop. Ensemble casts. Subtlety in the first ten seconds — this format has no patience budget to spend on restraint.

**The kit (ranked):**

1. **The Button** (Openings) `open.first_second_hook` → the close, not the open — see `tens.suspended_ending`
   The unresolved final beat is the entire product; everything before it is delivery mechanism.
2. **Suspended Ending** (Tension) `tens.suspended_ending`
   End on pressure, not on an answer. This is the mechanism that converts a viewer into a subscriber.
3. **Open Mid-Situation** (Openings) `open.open_mid_situation`
   No runtime exists for arrival. The first frame is already the argument.
4. **Enter on the Answer** (Openings) `open.enter_on_the_answer`
   Opening on a reply the audience did not hear buys the first second at zero cost.
5. **The Value Turn** (Scene) `scene.value_turn`
   One per episode. It is the only thing that makes 90 seconds feel like a story.
6. **Exposition as Accusation** (Dialogue) `dial.exposition_accusation`
   Backstory as a weapon in an argument. The format has no time for anyone to explain anything.
7. **Escalating Humiliation** (Tension) `tens.escalating_humiliation`
   The genre's engine: the protagonist is diminished in public and the audience buys the promise of reversal.
8. **Who Speaks Last** (Endings) `end.who_speaks_last`
   Hard convention here: give the last line to the force acting on the hero.
9. **The Cliffhanger Reveal** (Information) `info.cliffhanger_reveal`
   End on a reveal rather than a resolution, and make the reveal cost someone something.
10. **The Micro-Reveal Ladder** (Information) `info.micro_reveal_ladder`
    Every episode must pay something, or the audience learns that waiting is free.
11. **Late Entry** (Scene) `scene.late_entry`
    Applied to every scene, not just the episode open.
12. **Early Exit** (Scene) `scene.early_exit`
    Cut on the turn. The button is this technique at episode scale.
13. **The Delayed Reversal** (Information) `info.dramatic_irony`
    The audience knows the protagonist's real status while the cast does not. The wait for that gap to close is the series.
14. **Strategic Withholding** (Dialogue) `dial.strategic_withholding`
    The cheapest tension available: silence costs no seconds.
15. **Short-Form Compression** (Time) `time.shortform_compression`
    Weeks pass between episodes without a marker. The format permits it; use it.
16. **The Serial Clock** (Time) `time.serial_clock`
    One visible deadline across the series keeps 80 episodes from feeling like 80 incidents.
17. **The Season Ending** (Endings) `end.season_ending`
    Resolve, then open. A finale that only opens is a middle chapter.
18. **Shame** (Emotion) `emo.shame`
    The emotion this format runs on. It needs witnesses and a gap between claim and reality.
19. **Interruption** (Dialogue) `dial.interruption`
    Status readable in one line, which is all the establishing this format gets.
20. **The Public Scene** (Scene) `scene.public_scene`
    Private conflict staged where others can hear — subtext generated structurally, for free.

---

**The visual kit (ranked):**

1. **The Caption Bar** (Vertical) `vertical.caption_bar`
   A viewer arriving at episode forty of a hundred needs to know what they are in within a second. A fixed bar carrying the series name does it and costs nothing per episode.
2. **Stacked Two-Shot** (Vertical) `vertical.stacked_two_shot`
   The format's central problem is two people in a frame one hand wide. Stacking them vertically is the answer a horizontal two-shot cannot give.
3. **Fill The Frame** (Vertical) `vertical.fill_frame`
   A wide dies on a phone. In a micro-drama the performance is the entire budget, so it is framed at the size it reads.
4. **Eyes in the Upper Third** (Vertical) `vertical.eyes_upper_third`
   Where the viewer looks first, and where the interface is not. Get this wrong and every episode has its performance sitting under a caption stack.
5. **Cliffhanger Cut** (Vertical) `vertical.cliffhanger_cut`
   Sixty to a hundred episodes means the ending of each is the marketing for the next. The cut point is a structural decision, not an edit one.
6. **Medium Close-Up** (Shot size) `size.mcu`
   The working size of the form. Wide enough for a gesture, tight enough for a phone at arm's length in daylight.
7. **Flat Lighting / Beauty Lighting** (Lighting) `light.flat`
   Shot at volume on a schedule nothing else matches. Flat light is what lets sixty episodes look like one show without a relight.
8. **Colour Temperature as Mood** (Colour) `color.temperature`
   The cheapest way to separate two recurring locations when both are small rooms shot on the same day.
9. **Reaction Cut** (Cuts) `cut.reaction`
   Melodrama lives in the receiving face. This format cuts to the listener more often than to the speaker and that is correct.
10. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   Owns every vertical figure in the package, and they move - Reels grew its bottom band during 2026 and is now the tighter of the two. Compose to the tightest platform the piece ships on and take the number from the card.

## Reels, Shorts and TikTok

**Constraints:** 15 to 90 seconds, 9:16 by default (a 16:9 feed reel takes this kit unchanged), sound often off on the first pass, captions mandatory. The first one to three seconds decide retention. Platform ranking differs: TikTok rewards saves and shares, Reels rewards reach and profile follows, Shorts ranks on click-through and average view duration — **so the same script should be structured differently per platform.** No prior context; every viewer is a first-time viewer.

**Grammar:** Hook → Body → Payoff, or Hook → Problem → Solution → CTA. Roughly a 10/75/15 split between hook, build and payoff. One idea only. The promise made in the first second is the contract, and the whole piece exists to pay it. Open loops must close fast — a cliffhanger that stalls loses the viewer before the reveal.

**Palette:** 1 idea · 1 promise · 1 payoff · hook in second one · payoff before the last beat · loop-capable close

**Rules:** Decide which of the four promises you are making before writing a word — result, reveal, lesson or tension | Open on the strongest moment even if it comes from the end | One idea per piece | Close the loop you open, quickly | Write for sound-off first, then add sound as a bonus | Never promise a payoff you do not deliver — it costs more than a weak hook | Structure the piece so it can loop

**Examples:** educational explainer reels, process breakdowns, mistake lists, transformation reveals, single-joke sketches

**Avoid:** A preamble about what the video will contain — the viewer leaves before it starts. Two ideas in one piece. A hook that misrepresents the body, which trains the audience to distrust the account. An ending that just stops.

**The kit (ranked):**

1. **The First-Second Hook** (Openings) `open.first_second_hook`
   Decide the promise type first — result, reveal, lesson or tension — then write the hook to make it.
2. **The Stated Promise** (Openings) `open.stated_promise`
   In non-fiction short-form, saying it out loud is correct, not a failure.
3. **The Loop Opening** (Openings) `open.loop_opening`
   Write the open so it follows the close, and make the second viewing mean something different.
4. **First Five Seconds** (Tension) `tens.first_five_seconds`
   Establish the threat or the stake, not the world.
5. **Short-Form Shape** (Comedy) `comedy.short_form_shape`
   The comedy version of the same ratio, with the joke landing before the last beat.
6. **Enter on the Answer** (Openings) `open.enter_on_the_answer`
   The cheapest first second in the craft.
7. **Observational Truth** (Comedy) `comedy.observational_truth`
   The dominant native mode. Recognition is the share mechanism.
8. **The Specific Detail** (Emotion) `emo.specific_detail`
   The particular is what gets saved and sent; the general is what gets scrolled.
9. **Early Exit** (Scene) `scene.early_exit`
   Never wind down. The payoff is the last thing.
10. **Setup and Punchline** (Comedy) `comedy.setup_punchline`
    The whole form is one of these stretched to 40 seconds.
11. **The Rule of Three** (Comedy) `comedy.rule_of_three`
    One triple per piece. Two in a row reads as a crutch.
12. **Funny Word Last** (Comedy) `comedy.funny_word_last`
    Word order is the only timing tool available on a caption.
13. **Misdirection** (Comedy) `comedy.misdirection`
    The turn must arrive before the viewer's thumb does.
14. **Earn the Situation** (Emotion) `emo.earn_the_situation`
    The honest limit of the format: emotion cannot be brute-forced in 30 seconds, so borrow a situation the audience already carries.
15. **Write for the Ear** (Voice) `voice.write_for_the_ear`
    Then check it reads silently, because half the audience will never hear it.
16. **The Kids' Register** (Voice) `voice.kids_register`
    For any short-form aimed at children, where the constraints tighten again.
17. **Suspended Ending** (Tension) `tens.suspended_ending`
    Only for serialised pieces. On a standalone it reads as a broken file.

---

**The visual kit (ranked):**

1. **Face In The First Frame** (Vertical) `vertical.face_first_frame`
   One to three seconds decide retention, and a face is the fastest thing a human parses. Everything else in the kit is downstream of this.
2. **On-Screen Captions** (Vertical) `vertical.captions`
   Sound is off on the first pass, so the captions are the soundtrack. They are mandatory here in a way they are nowhere else.
3. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   Owns every vertical figure in the package and they move - Reels is now the tighter of the two at the bottom, which was the reverse until 2026. Compose to the tightest platform the piece ships on.
4. **Fill The Frame** (Vertical) `vertical.fill_frame`
   A wide shot dies on a phone. Shoot closer than feels right, and put movement in any wide you cannot avoid.
5. **Silent First Second Rule** (Vertical) `vertical.silent_first_second`
   The first second often plays before the audio loads. A piece whose hook is a line is a piece whose hook is missed.
6. **Pattern Interrupt** (Vertical) `vertical.pattern_interrupt`
   Retention is lost in the middle, not at the start. A change of frame, scale or texture every few seconds is what holds it.
7. **Loop Point** (Vertical) `vertical.loop_point`
   The last frame meeting the first turns one view into three. It is the only device here that multiplies the number rather than protecting it.
8. **Punch In** (Vertical) `vertical.punch_in`
   The cheapest energy in the format, and the one that survives being watched muted at arm's length.
9. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Judged on a phone, outdoors, in motion - which is where this is actually watched and not where it is ever edited.
10. **Jump Cut** (Cuts) `cut.jump`
   Ranked last deliberately. It is the format's default and the first thing an audience reads as a template; used without a reason it is dead air removed rather than energy added.

## Long-Form Video (YouTube)

**Constraints:** 8 to 25 minutes, 16:9, sound on, watched on a phone or a television. Title and thumbnail bring the viewer; the retention curve decides whether it travels, and the first thirty seconds determine roughly half the outcome. Usually self-produced with limited editing time, so any complex structure will not survive the weekly schedule.

**Grammar:** A promise made in the first thirty seconds and paid at the end. Sections marked by a change of place, pace or music so the viewer feels progress. The structure is built in the edit from far more material than is used, which means the script is a spine rather than a text. Retention is managed by re-opening a question every few minutes, not by raising volume.

**Palette:** 1 promise · 3–6 sections · a question open at all times · a payoff at the end that answers the title

**Rules:** Make the promise in the first thirty seconds and pay it at the end | Keep one question open at every moment | Change something every two or three minutes — place, pace, or register | Do not save the best material for the end; open with it | The title is the contract, and the video must answer it exactly | Cut anything that does not either advance the promise or entertain

**Examples:** video essays, process and build videos, investigation and explainer channels, documentary-style vlogs

**Avoid:** A long preamble about what the video will contain. Answering a different question from the title, which is the most damaging thing a channel can do repeatedly. One unchanging register for twenty minutes.

**The kit (ranked):**

1. **The Stated Promise** (Openings) `open.stated_promise`
   Thirty seconds, explicit, and paid exactly.
2. **Answer the Question That Was Asked** (Endings) `end.answer_the_question_asked`
   The title asked it. Answer that one, not a related one.
3. **The Scene Question** (Scene) `scene.scene_question`
   Applied to sections: one open question at all times, or retention falls.
4. **The Staged Reveal** (Information) `info.staged_reveal`
   Break the payoff into steps so each section pays a little.
5. **Theme in Non-Fiction** (Theme) `theme.theme_in_nonfiction`
   State the claim to yourself before the edit, or the edit will be decided by whichever footage is strongest.
6. **The Question Planted First** (Openings) `open.question_planted_first`
   One specific unexplained thing carries the viewer until the argument can.
7. **Exposition as Correction** (Dialogue) `dial.exposition_correction`
   Deliver facts by correcting a wrong version. Retention is far better than for a flat statement.
8. **The Expert Who Explains Badly** (Dialogue) `dial.expert_explains_badly`
   The interview version: let the first explanation fail, then ask again, simpler.
9. **The Confiding Voice** (Voice) `voice.confiding_voice`
   The register the format was built on. It is a relationship, and the relationship is the channel.
10. **Authority by Specificity** (Voice) `voice.authority_by_specificity`
    Numbers, names and dates buy more trust than any claim of expertise.
11. **The Voiceover Open** (Openings) `open.voiceover_open`
    The narration must know something the footage does not, or it is redundancy with music under it.
12. **Contrast Sets the Level** (Emotion) `emo.contrast_sets_level`
    Flatten the surroundings so the two big moments land.
13. **The Handoff Ending** (Endings) `end.handoff_ending`
    One nameable action, not a request to care.
14. **The Callback** (Endings) `end.callback`
    Return the opening line at the end. Cheap, reliable, and it signals the piece was designed.
15. **Register Creep** (Voice) `voice.register_creep`
    A named failure. Twenty minutes is long enough for the voice to drift without anyone noticing.

---

**The visual kit (ranked):**

1. **Cover Frame / Thumbnail Frame** (Vertical) `vertical.cover_frame`
   The thumbnail brings the viewer and nothing else does. Design the frame during the shoot rather than hunting for one in the edit.
2. **Depth Layers: Foreground, Midground, Background** (Composition) `comp.depth_layers`
   Eight to twenty-five minutes on one setup is where attention dies. Depth is what makes the same room hold a viewer past minute four.
3. **B-Roll Burst** (Vertical) `vertical.broll_burst`
   The retention tool of the form: a short run of cutaways at the point the talk is at its least visual.
4. **Medium Close-Up** (Shot size) `size.mcu`
   Watched on a phone and on a television in the same hour. This is the size that survives both.
5. **Motivated Lighting** (Lighting) `light.motivated`
   A channel is a recurring set. Light that comes from the room is what makes two hundred videos look like one channel.
6. **Lighting Ratio** (Lighting) `light.ratio`
   The single decision that separates a channel that looks made from one that looks recorded, and it costs one flag.
7. **Punch In** (Vertical) `vertical.punch_in`
   A second size on the same setup, cut on the point. It is the cheapest edit-side energy available to a talking format.
8. **Colour Temperature as Mood** (Colour) `color.temperature`
   Distinguishes the segments of a long video - the intro, the test, the verdict - without a caption on any of them.
9. **On-Screen Captions** (Vertical) `vertical.captions`
   Not mandatory here as it is in short form, but a large share of the audience watches muted anyway and the retention graph shows it.
10. **Symmetry** (Composition) `comp.symmetry`
   The frame a channel becomes recognisable by. Decided once, it survives every editor who touches the series afterwards.

## Web Series and Serialised Short-Form

**Constraints:** 5 to 15 minutes per episode, released on a schedule, with an audience that may arrive at any episode and may forget between them. Budgets are small, casts are small, locations repeat.

**Grammar:** Each episode is a complete unit with its own turn, plus one thread that does not close. Reorientation happens through action rather than recap. Locations repeat, so the same room must be made to feel different by what happens in it rather than by where it is.

**Palette:** 1 episode question closed · 1 series question left open · 2–4 locations reused · cold open under 40 seconds

**Rules:** Close the episode's question, open the series' question | Reorient through behaviour, never through recap | Give every episode one scene that could be someone's favourite | Repeat locations deliberately and change what they mean | Make the first episode work as a complete short film

**Examples:** independent serialised drama, branded serials, comedy shorts with a recurring cast

**Avoid:** A first episode that is entirely setup. Recap scenes. A series question so vague the audience cannot state it. Assuming the audience watched last week.

**The kit (ranked):**

1. **The Series Opening** (Openings) `open.series_opening`
   Close the episode's question and open the series' question in the same beat.
2. **The Episode Opening** (Openings) `open.episode_opening`
   Reorient through an action that only makes sense in light of last week.
3. **The Season Ending** (Endings) `end.season_ending`
   Resolve first, then open. Change the standing condition so the next run cannot repeat this one.
4. **The Cold Open Reveal** (Information) `info.cold_open_reveal`
   Open on a reveal rather than a recap.
5. **The Cold Open** (Openings) `open.cold_open`
   Complete in itself, with its own small turn.
6. **Thematic Consistency Across a Series** (Theme) `theme.series_consistency`
   One claim on one page, and every episode assignable to a position on it.
7. **Transplant the Scene** (Scene) `scene.transplant`
   The cheapest way to make a repeated location earn its return.
8. **The Ensemble Grid** (Character) `char.ensemble_grid`
   Small casts are remembered by difference, not by screen time.
9. **The Serial Clock** (Time) `time.serial_clock`
   One deadline across the run stops episodes reading as incidents.
10. **The Micro-Reveal Ladder** (Information) `info.micro_reveal_ladder`
    Pay something every episode.
11. **The Interrupted Scene** (Scene) `scene.interrupted_scene`
    The honest way to end an episode without the characters conveniently stopping.
12. **Suspended Ending** (Tension) `tens.suspended_ending`
    Used sparingly. Not every episode should end on pressure.

---

**The visual kit (ranked):**

1. **The Caption Bar** (Vertical) `vertical.caption_bar`
   An audience that may arrive at any episode and forgets between them. The bar is the cheapest continuous identity a serial has.
2. **Frame Within a Frame** (Composition) `comp.frame_within_frame`
   A recurring visual signature costs nothing to repeat and does the work a title sequence would cost five per cent of the runtime.
3. **Colour Temperature as Mood** (Colour) `color.temperature`
   Locked per recurring location in episode one, and every later episode inherits it without a decision.
4. **Medium Close-Up** (Shot size) `size.mcu`
   Five to fifteen minutes on a small budget: this is the size that holds a performance without asking for a set.
5. **Cliffhanger Cut** (Vertical) `vertical.cliffhanger_cut`
   Released on a schedule, so the last shot of each episode is doing the work of an advertisement for the next.
6. **Flat Lighting / Beauty Lighting** (Lighting) `light.flat`
   Shot in blocks across a weekend. Flat light is what lets episode two and episode nine be filmed an hour apart and still cut together.
7. **Depth Staging** (Blocking) `block.depth_staging`
   The same location has to play four times. Staging it from a different depth is the only variety available for free.
8. **Cut on Action** (Cuts) `cut.on_action`
   Holds the whole series to one cutting grammar regardless of who edits which episode.
9. **Lighting Ratio** (Lighting) `light.ratio`
   Declared once as the series' look. It is the property a viewer uses to recognise the show before the title appears.
10. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   If any part of the series ships vertically, the figure comes from this card, which owns them all and carries their date.

## Carousel, Thread and Text Post

**Constraints:** Static or near-static, read rather than watched, scrolled at speed. The first line and the first image do all the acquisition work. Length is elastic but attention is not. No sound, no motion, no performance — the writing is the entire product.

**Grammar:** One claim, evidenced. The first card or line states the promise; each subsequent unit advances it by exactly one step; the last unit pays it and asks for the one action. Line breaks are the only pacing instrument available.

**Palette:** 1 claim · 5–10 units · one step per unit · a payoff, then one ask

**Rules:** The first line is the whole acquisition — write it last, after you know the payoff | One step per card, never two | White space is the pacing | Never make the reader swipe to find out what the post is about | End on the payoff, then the ask, in that order and not reversed

**Examples:** explainer carousels, process threads, opinion threads, case-study posts

**Avoid:** Burying the claim. A first line that is a throat-clear. A carousel whose last card is a logo — the last card is the payoff, and the brand goes anywhere else.

**The kit (ranked):**

1. **The Stated Promise** (Openings) `open.stated_promise`
   This format is almost entirely this card.
2. **Sentence Rhythm** (Voice) `voice.sentence_rhythm`
   Line breaks are the only timing available, so rhythm is the whole delivery.
3. **The Specific Detail** (Emotion) `emo.specific_detail`
   One concrete particular outperforms any amount of framing.
4. **Authority by Specificity** (Voice) `voice.authority_by_specificity`
   A number and a date beat a claim of expertise.
5. **The Question Planted First** (Openings) `open.question_planted_first`
   For opinion and narrative threads, where the promise cannot be stated flatly.
6. **The Staged Reveal** (Information) `info.staged_reveal`
   One step per unit. This is the technique the format is built on.
7. **The Handoff Ending** (Endings) `end.handoff_ending`
   One nameable action, after the payoff and never before it.
8. **Understatement** (Comedy) `comedy.understatement`
   The register that travels furthest in text, because it survives being read in any tone.
9. **The Last Line** (Endings) `end.last_line`
   The most screenshotted unit in the format. Write it first and build backwards.
10. **Register Setting** (Voice) `voice.register_setting`
    Decide the register before the first line, because a thread cannot change it midway without reading as two people.

---

**The visual kit (ranked):**

1. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Read rather than watched, scrolled at speed, on a phone. Type judged on a monitor is type nobody finished.
2. **Negative Space** (Composition) `comp.negative_space`
   A carousel card holds one idea. The space around it is what tells a scrolling reader the idea has ended.
3. **Symmetry** (Composition) `comp.symmetry`
   Consistent placement across ten cards is what makes a carousel read as one object rather than ten images.
4. **Colour Contrast** (Composition) `comp.color_contrast`
   The first image does all the acquisition. Contrast is what stops the thumb.
5. **Never Colour Alone** (Access) `access.not_colour_alone`
   A carousel that separates its categories by hue alone has excluded roughly one reader in twelve, silently.
6. **Pick the Chart from the Question** (Data) `dataviz.chart_from_the_question`
   Threads carry numbers more than any other social format. Choose the shape from the sentence being proved, not from the spreadsheet.
7. **Build It, Do Not Generate It** (Data) `dataviz.build_do_not_generate`
   A chart asked for from an image generator returns plausible nonsense with numbers on it, and this is the format most likely to try.
8. **Rule of Thirds** (Composition) `comp.rule_of_thirds`
   The layout default that survives being cropped to square, portrait and landscape by three different platforms.
9. **4:5 Vertical** (Format) `look.ratio_4x5`
   The tallest ratio most static feeds will render without cropping, which makes it the most screen a still post can occupy.
10. **Text Safe Zone Planned in the Image** (Text) `textimg.text_safe_zone`
   Planned in the image rather than added over it, because a carousel is re-cropped by every platform it is reposted to.

## Podcast Clip

**Constraints:** 15 to 45 seconds is the working band, with 30 to 60 the wider short-form norm; 9:16 with burned captions. The clip is cut from material that was never written for it, which is the format's defining condition — you are not writing, you are finding. A 45-minute episode realistically yields eight to twelve usable clips. Talking-head video substantially outperforms the static audiogram on Instagram and LinkedIn, which is why the waveform card is now the fallback rather than the format. The viewer has no context, did not choose the show, and in the overwhelming majority of cases will never listen to the episode — so the clip has to be a complete object rather than an advertisement for one. **Guests agreed to a conversation, not to a decontextualised fifteen seconds; a clip that changes what a sentence meant is a misrepresentation of a real person, and clip approval is worth agreeing before the recording rather than after it.** *(Platform behaviour as at 2026-09.)*

**Grammar:** A clip is a single exchange with a turn in it, and the turn is almost always someone conceding, correcting or admitting something. That is why the useful search in a transcript is not for the best-phrased answer but for the moment a position moves. The clip opens on the answer — there is no time for the question, and the question usually reads as throat-clearing anyway — which means the first sentence has to carry its own setup. The last line is the whole share: it is the sentence that gets screenshotted, quoted and argued with, and choosing it is most of the editorial work. Pauses are content and survive at this length better than anywhere else, because a two-second hesitation before an admission is legible even to someone who arrived four seconds ago. Captions are the primary channel and the cleaning decision — whether to tidy the "um" out — changes what the speaker sounds like. And the clip should be cuttable to where it ends without the audience feeling a hole, which is a discipline of finding rather than of writing.

**Palette:** 15 to 45 seconds, 1 exchange, 1 turn, 1 chosen last line, captions always, 8 to 12 clips per long episode, 0 questions left in

**Rules:** Search the transcript for the moment a position moves, not for the best-phrased answer | Open on the answer; the question is almost never worth the seconds | Choose the last line first and cut backwards to it | Keep the pause before the admission | Write the captions as the primary track and decide the cleaning policy once | Never assemble a sentence the speaker did not say in that order | Where the brief raises clip rights or guest approval, note them with the brief | Make the clip complete; almost nobody will go to the episode | Cut the host's laugh unless it is the turn | Do not label a clip with a promise the clip does not keep

**Examples:** interview-show clip channels, business and health podcast clip accounts, comedy podcast moments, debate and disagreement clips, clip-first shows that record for the clip and edit the episode afterwards

**Avoid:** The clip that only makes sense if you heard the previous four minutes, which is most first-pass clips and all of the ones cut by someone who listened to the whole episode. A hook caption that promises a claim the clip does not contain, which buys a view and teaches the platform to stop serving the account. Cutting between two answers so a guest appears to say something they did not — the format's cardinal offence and the one that ends the relationship with the guest. Every "um" removed, which produces a speaker who sounds synthetic. And the clip that ends on a wind-down because that is where the natural pause fell.

**The kit (ranked):**

1. **Enter on the Answer** (Openings) `open.enter_on_the_answer`
   Literally how a clip begins. The card is about making a reply carry its own question, which in this format is the difference between a clip and a fragment.
2. **The Last Line** (Endings) `end.last_line`
   The sentence that gets screenshotted. Choose it from the whole transcript and cut backwards; everything before it exists to deliver it.
3. **The Interview Admission** (Information) `info.interview_admission`
   What you are actually searching the transcript for. A clip of someone stating a known position is information; a clip of someone conceding is a share.
4. **The First-Second Hook** (Openings) `open.first_second_hook`
   In a clip the hook is not written, it is located — the strongest sentence in the exchange, moved to the front if the edit allows it honestly.
5. **The Half-Truth** (Dialogue) `dial.half_truth`
   Recognising the partial answer tells you where the interesting cut is, and also where a clip would misrepresent if you took it.
6. **Grading the Gap** (Subtext) `sub.graded_pause`
   Not all pauses are worth their seconds at this length. This card is how you decide which hesitation to keep at full length in a twenty-second clip.
7. **Rhythm in Captions and Subtitles** (Rhythm) `rhythm.subtitle_beat`
   Line length, break point, hold. Most of this audience reads the clip rather than hearing it, and the caption's rhythm is the clip's rhythm.
8. **The Interview Transcript, Cleaned or Not** (Rhythm) `rhythm.transcript_clean`
   The single most consequential policy decision in this format. Decide it once for the show, and apply it to every clip.
9. **Exposition as Correction** (Dialogue) `dial.exposition_correction`
   A fact delivered by correcting a wrong version outperforms a flat statement, and clips built on a correction travel further than clips built on an explanation.
10. **The Specific Detail** (Emotion) `emo.specific_detail`
    "Eleven days in a hotel" beats "for a while". The particular is what gets saved and sent; the general is what gets scrolled.
11. **The Podcast Host Voice** (Voice) `voice.host_voice`
    The host's register is the show's brand and the reason a viewer follows after one clip. Knowing what it is tells you which clips represent the show.
12. **Behaviour That Contradicts the Line** (Subtext) `sub.behaviour_contradicts`
    In a video clip the hands and the eyes are visible at fifteen seconds. It is the cheapest subtext available and it survives muted playback.
13. **The Delayed Reaction** (Emotion) `emo.delayed_reaction`
    The feeling that arrives three seconds after the sentence. Holding it is why the clip is twenty-two seconds rather than seventeen.
14. **Funny Because True** (Comedy) `comedy.observational_truth`
    Recognition is the share mechanism in conversational clips exactly as it is in scripted short-form.
15. **Who Asks the Questions** (Dialogue) `dial.question_control`
    Read it upstream: control of the question in the room is what produces a clippable answer. The clip strategy starts in the recording, not the edit.
16. **The Cheat** (Information) `info.the_cheat`
    **This card names a failure, not a recommendation.** It is ranked here as the guard on the clip assembled from two answers, or trimmed so a qualifier disappears. Test every clip against it before it leaves the edit.

---

**The visual kit (ranked):**

1. **On-Screen Captions** (Vertical) `vertical.captions`
   The clip is the caption. Sound is off, the speaker is not famous, and the words are the only content.
2. **Stacked Two-Shot** (Vertical) `vertical.stacked_two_shot`
   Two people recorded on a horizontal setup, delivered vertically. Stacking is the only reframe that keeps both faces legible.
3. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   A clip is captioned to the edge of the frame more often than any other format, and the bands move. The figure is on this card.
4. **Punch In** (Vertical) `vertical.punch_in`
   Fifteen to forty-five seconds of two static cameras needs a second size, and the punch-in is the only one available in post.
5. **Freeze and Text** (Vertical) `vertical.freeze_text`
   The line the clip exists for, held. It is the format's one moment of design and most clips skip it.
6. **Cold Cut** (Vertical) `vertical.cold_cut`
   Start mid-sentence. A clip that begins with a greeting has spent its retention window on politeness.
7. **Reframe to 9:16** (Vertical) `vertical.reframe`
   The whole format is a reframe. Doing it as a decision rather than an auto-crop is the difference between a clip and a cropped clip.
8. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Burned captions at phone size in daylight, which is the only condition this is ever watched in.
9. **Flat Lighting / Beauty Lighting** (Lighting) `light.flat`
   The recording is already lit and cannot be relit. Flat is what a podcast set is, and the clip inherits it.
10. **Loop Point** (Vertical) `vertical.loop_point`
   A clip that ends where it began gets watched twice, and a clip watched twice is distributed.

## Livestream Script

**Constraints:** Unrepeatable and unedited — you cannot recall a sentence, so compliance, legal exposure and factual accuracy are live rather than post. The audience is continuously replaced: retention drops sharply after about the five-minute mark unless something interactive happens, and the working convention is an engagement prompt every five to seven minutes and a restatement of the core pitch every five to ten. The document is a run of show or cue-to-cue rather than a script — segments, cues, talent, timings, interstitials — with scripted language only at the points that must be exact. Chat runs in parallel and is a second voice the host is answering in real time. Latency means the host is always responding to something that happened twenty seconds ago. And the VOD is a different product with a different audience, which usually needs a chaptered version rather than the live one. *(Platform retention behaviour as at 2026-09.)*

**Grammar:** The governing fact is that the audience at minute forty is not the audience at minute four, which inverts one of the strongest rules in the rest of this database: repetition here is not redundancy, it is the only way most viewers ever receive the information. So the structure is a loop of segments, each of which re-establishes enough to be entered cold, with an interactive prompt at the seam. Real time is the medium — the value is that this could go wrong, and a stream that has removed every possibility of that has removed its own reason to be live. The host's register has to be fixed early and held for hours, because drift over a long stream is the most common way a stream ends in a clip nobody wanted. Chat is a character, and treating it as one — quoting it, disagreeing with it, being interrupted by it — is what separates a stream from a webinar. And the hedge is functional: a host who is certain for three hours is a host who has stopped reacting to anything.

**Palette:** 45 to 120 minutes, segments of 8 to 12 minutes, 1 prompt every 5 to 7, the core message restated every 5 to 10, 1 held register, 1 chaptered VOD

**Rules:** Write a run of show, not a script; script only the lines that must be exact | Build every segment so it can be entered cold | Restate the core message every five to ten minutes without using the same words | Put an interactive prompt at every seam | Treat chat as a character and answer it by name | Fix the register in the first two minutes and hold it for the whole stream | Keep the hedges; certainty over three hours reads as a recording | Leave one thing genuinely unscripted, because that is why it is live | Say nothing you could not stand behind as a clip, because it will be one | Plan the VOD's chapters before you go live, not after

**Examples:** creator variety and just-chatting streams, live shopping and livestream commerce, product launches and developer keynotes streamed live, charity fundraising marathons, live Q&A and AMA formats, esports and watch-along broadcasts

**Avoid:** Going unscripted as a principle, which is the fastest way to lose the audience that arrived at minute twenty. A single unbroken forty-minute block with no seam anyone can enter at. The pitch delivered once, at minute eight, to four percent of the eventual audience. A host whose register has drifted by hour two into something the brand would not have approved in writing. And a VOD published as an unchaptered three-hour file, which is a recording rather than a product.

**The kit (ranked):**

1. **Information the Listener Already Has** (Dialogue) `dial.known_information`
   **Ranked deliberately, and in most formats the same move is the "as you know" failure** — a character explaining what everyone present already knows. The livestream is an exception for the same structural reason the soap is: the audience is continuously replaced, so telling the room what "everyone already knows" reaches people who genuinely do not. Read the card for why it reads as clumsy when the audience is fixed, then re-tell on purpose — and never in the same words twice.
2. **The Session Recap** (Interactive) `inter.session_recap`
   The "for anyone just joining" move, done as craft. The card is about re-entry that carries information without punishing the people who were already there.
3. **The Podcast Host Voice** (Voice) `voice.host_voice`
   The host is the format. Defining the register before the stream is the only way it survives three hours of improvisation intact.
4. **Talking to the Room** (Dialogue) `dial.talking_to_the_room`
   Direct address that makes a broadcast feel like a conversation. In a stream it is not a device, it is the default mode and everything else is the exception.
5. **The Implied Listener** (Voice) `voice.implied_listener`
   Who the host is imagining. Getting this wrong is why some streams feel like they are addressed to nobody in the chat.
6. **Real Time** (Time) `time.real_time`
   The medium's whole proposition. The card is about what real duration does to an audience, and why removing all risk removes the reason to watch live.
7. **The Real-Time Beat** (Beat) `beat.real_time_beat`
   Beats that cannot be compressed, because there is no edit. It changes what a segment can contain and how long a silence is allowed to be.
8. **What a Listener Cannot Re-Read** (Rhythm) `rhythm.listener_no_rewind`
   No scrubbing, no going back. Every sentence has to land on one hearing, which is a harder constraint than a podcast's and much harder than a page's.
9. **Entering a Sequence** (Sequence) `seq.run_entry_context`
   How a run of scenes establishes its own context at the start. In a stream every segment is one of these and most of them are written as though they were the first.
10. **The Parasocial Bond** (Relationship) `rel.parasocial_bond`
    The relationship that keeps people watching for three hours. Naming it honestly is also how a host avoids exploiting it.
11. **Conversational Persona Writing** (Interactive) `inter.conversational_persona`
    Writing a persona that has to respond rather than deliver. It is the card closest to what a live host is actually doing.
12. **The Hedge** (Voice) `voice.the_hedge`
    Hedging reads as a person thinking. Over hours it is what stops the stream sounding pre-recorded, which is the worst thing a stream can sound like.
13. **Rapid-Fire Exchange** (Dialogue) `dial.rapid_fire`
    For the chat-response passages, where a dozen short exchanges land in ninety seconds and the pace is the entertainment.
14. **Contrast Sets the Level** (Emotion) `emo.contrast_sets_level`
    Flatten most of the stream and the two moments that matter land. A host at maximum for two hours has no peak available.
15. **The Handoff Ending** (Endings) `end.handoff_ending`
    One nameable action at the close, and the same one at every restatement. A stream that ends with a menu of asks converts on none.
16. **Register Creep** (Voice) `voice.register_creep`
    **This card names a failure, not a recommendation.** It is ranked here because a three-hour unedited broadcast is the single most favourable environment for it in this database, and the drift is invisible to the person doing it. Read it before the stream, not after the clip.

---

**The visual kit (ranked):**

1. **Talk-To-Camera** (Vertical) `vertical.talk_to_camera`
   Unrepeatable and unedited: the lens is the only place to look and the address is the whole form.
2. **Flat Lighting / Beauty Lighting** (Lighting) `light.flat`
   Nothing can be relit mid-stream and nothing can be fixed after. Flat light is the only setup that survives an hour of unplanned movement.
3. **The Caption Bar** (Vertical) `vertical.caption_bar`
   Viewers arrive continuously and mid-thought. A fixed bar saying what this is answers the question nobody can stop to ask.
4. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   A live frame carries the platform's chat, its interface and its overlays at once, and the reserved bands are the tightest of any format here.
5. **Medium Close-Up** (Shot size) `size.mcu`
   The size that holds for an hour without becoming either a portrait or a room.
6. **Motivated Lighting** (Lighting) `light.motivated`
   A stream set is a permanent set. Light sourced from the room stays correct when the presenter turns, which they will.
7. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Lower thirds and on-screen text compete with a live chat column for the same attention and the same pixels.
8. **Eye Level** (Angle) `angle.eye_level`
   A camera above the presenter reads as a webcam and a camera below reads as an accident. Level is the register of address.
9. **Symmetry** (Composition) `comp.symmetry`
   A stream frame is looked at for an hour. Composed once and centred, it stops being noticed, which is the goal.
10. **Clear the Frame and Hold** (Blocking) `block.clear_frame_and_hold`
   The one blocking decision a live show can plan: where the presenter goes when they hand over, and what the frame shows while they do.

## Branded Serial

**Constraints:** Three to eight episodes of two to eight minutes, released on a schedule on the brand's own channels plus paid distribution — and the media budget, not the story, decides whether episode two is ever seen, so each episode has to work cold for a first-time viewer. Consumer preference has shifted measurably toward original series over one-off ads, and platform ad products have followed: TikTok's branded minidrama format arrived in 2026 as a native serialised placement. It is advertising and must be identifiable as such wherever it runs. The brand's legal function reviews a fiction script, which is a different and slower conversation than reviewing a claim. The award-circuit definition is the useful editorial one: the audience experience comes first and the brand plays a supporting but integral role. Casts and locations repeat because budgets are per-series rather than per-episode. *(Platform products and preference data as at 2026-09.)*

**Grammar:** The brand is the world, not the protagonist — this is the same rule as the brand film, but a serial tests it every episode because there are six more chances to break it. The reliable architecture is an episode question closed and a series question left open, with the brand present as a condition of the world rather than as the agent of the resolution. Each episode needs a cold entry, because a viewer who arrives at episode four from a paid placement is the normal case rather than the exception. The button matters more than in an independent serial, since there is no platform carousel automatically serving the next one; the viewer has to want it enough to go and find it. Small casts are remembered by difference, so the ensemble grid does structural work that runtime cannot. And the recurring object — the thing the brand actually makes — is how the product stays in the fiction without a character selling it.

**Palette:** 3 to 8 episodes, 2 to 8 minutes each, 1 series question, 1 episode question closed per episode, 2 to 4 locations reused, 1 recurring object, 1 identifiable disclosure

**Rules:** Close the episode's question and leave the series' question open | Write every episode to be entered cold | Make the brand the condition of the change, never its agent | Build the ensemble on difference, because there is no runtime for depth | Put the product in the world as an object, not in anyone's mouth | Design the button; nothing will autoplay the next episode for you | Repeat locations deliberately and change what they mean | Keep the disclosure visible and stop treating it as a compliance tax | Write episode two's opening before locking episode one | Resolve in the finale, then open — a last episode that only opens has spent the budget on a pilot

**Examples:** TikTok branded minidramas, creator-led brand series on YouTube and Instagram, long-running insurance and telecom character series, fashion-house short serials, in-app micro-drama series, restaurant and retail creator serials

**Avoid:** The six-episode advertisement, where every instalment resolves by someone using the product and the audience learns the shape in episode two. A character who exists to say the brand's line, which is visible instantly and retroactively spoils every scene they are in. Episode one that is entirely setup for a series most viewers will meet at episode three. A series question so vague no viewer could state it. And a finale that opens a second season nobody has funded, which is a middle chapter with a media plan.

**The kit (ranked):**

1. **The Controlling Idea** (Theme) `theme.controlling_idea`
   One sentence that is true about the world and provable across the run. Without it a branded serial becomes six unrelated films with a consistent logo.
2. **The Story Spine** (Structure) `struct.story_spine`
   Once / every day / until one day. The fastest test of whether a brand brief contains a series at all, applied per episode and across the run.
3. **The Series Opening** (Openings) `open.series_opening`
   Installs the series question, the world and the contract. On brand-owned channels it also has to establish that this is worth returning to, which nothing else will do for you.
4. **The Episode Opening** (Openings) `open.episode_opening`
   Every episode is somebody's first. Reorient through an action rather than a recap, because a recap in a four-minute episode is a quarter of it.
5. **Theme Carried by an Object** (Theme) `theme.object_carrier`
   The product as a thing in the world that means something. It is how the brand stays present without a character advocating for it.
6. **The Suspended Ending** (Tension) `tens.suspended_ending`
   The button. In a serial with no autoplay behind it, the button has to be strong enough to make someone go and look for the next one.
7. **The Ensemble Grid** (Character) `char.ensemble_grid`
   Small casts are remembered by difference, not by screen time. In four-minute episodes the grid does the work a first act would do elsewhere.
8. **The Micro-Reveal Ladder** (Information) `info.micro_reveal_ladder`
   Every episode pays something. Without it the audience learns that waiting costs nothing, which on a brand channel means they simply stop.
9. **Thematic Consistency Across a Series** (Theme) `theme.series_consistency`
   Brand series are written by rotating teams against a rotating approvals process. This is the card that stops the run drifting into three different arguments.
10. **Transplant the Scene** (Scene) `scene.transplant`
    Budgets force location reuse. This is the cheapest way to make the same room mean something different in episode five than it did in episode one.
11. **Emotion in Advertising and Brand Work** (Emotion) `emo.brand_emotion`
    What a brand may honestly ask an audience to feel across six episodes, and where the accumulation starts to read as a claim.
12. **POV in Advertising and Brand Voice** (POV) `pov.brand_pov`
    Whose eyes the series uses. A branded serial that uses the brand's eyes has already made the brand its protagonist.
13. **Serial Clock** (Time) `time.serial_clock`
    One visible deadline running under the run — a wedding, a deadline, a season — which stops six short episodes reading as six sketches.
14. **The Season Ending** (Endings) `end.season_ending`
    Resolve, then open, in that order. A branded finale that only opens has spent a media budget on a pilot for something nobody commissioned.
15. **The Mouthpiece** (Character) `char.mouthpiece`
    **This card names a failure, not a recommendation.** It is ranked here because the branded serial produces it structurally: somebody in the approvals chain wants a line said, and it gets given to a character. Read it as the test on every scene the client asked for.

---

**The visual kit (ranked):**

1. **Palette Lock** (Identity) `ident.palette_lock`
   A brand serial is judged on consistency before it is judged on anything else, and the palette is the property that drifts fastest across episodes.
2. **Frame Within a Frame** (Composition) `comp.frame_within_frame`
   The recurring signature that lets three to eight episodes read as one campaign without a logo on every frame.
3. **Lighting Ratio** (Lighting) `light.ratio`
   Declared at episode one as the campaign's look. It is what a brand actually means when it asks for premium.
4. **Motivated Lighting** (Lighting) `light.motivated`
   Brand work defaults to unmotivated beauty light, which is why so much of it looks like an advertisement rather than a story.
5. **Colour Temperature as Mood** (Colour) `color.temperature`
   Maps the episodes against each other, so a viewer who sees episode four knows it belongs to the one they saw last week.
6. **Depth Layers: Foreground, Midground, Background** (Composition) `comp.depth_layers`
   Lets the product live in the frame without being the subject of it, which is the whole problem of branded storytelling.
7. **Medium Close-Up** (Shot size) `size.mcu`
   Two to eight minutes on a paid-distribution budget. This is the size that carries a performance and still reads in a feed.
8. **Cliffhanger Cut** (Vertical) `vertical.cliffhanger_cut`
   Released on a schedule, so each episode ends with the reason to come back rather than with the brand's name.
9. **Clean Plate Clause** (Text) `textimg.no_watermark_clause`
   Brand deliverables get versioned, localised and re-titled. A frame with type burned into it is a frame that cannot be reused.
10. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   Paid distribution means vertical cutdowns, and the figures are on this card with their date.

## Creator Brand Partnership (Organic)

**Constraints:** This is the paid UGC ad's sibling and a different object: the post lives on the creator's own feed, in front of the creator's own audience, with no media spend behind it, so it has to perform as a normal post for that account or it does not perform at all. The material connection must be disclosed clearly and conspicuously — a platform's paid-partnership label plus a visible in-content disclosure, not a hashtag buried in the caption. The brand gets approval rights but not authorship, and the negotiation over that line is the job. Usage rights and whitelisting for paid amplification are priced separately and frequently forgotten until the brand wants to run it as an ad. The post stays on the creator's grid indefinitely and will be read against everything they post afterwards. Audience trust is the asset being rented, and it does not renew automatically. *(Disclosure practice as at 2026-09; the rules and the platform labels both change.)*

**Grammar:** Everything here follows from one asymmetry: the brand is buying access to a relationship it does not own and cannot operate. So the creator's voice wins every conflict with the brand's voice — not as a courtesy but because the audience is there for the former and will detect the latter in one sentence. The post must be a post that creator would plausibly have made anyway, with the product inside it rather than as its subject. Disclosure is written as content, in the creator's own cadence, early, because a disclosure delivered in a different register is the single most reliable tell that the rest was scripted elsewhere. The claim the brand wants is carried by a specific experience rather than a stated benefit. And the durability test is unusual for advertising: this thing will still be on the grid in a year, so a post that only works inside a campaign window is a post the creator will quietly delete.

**Palette:** 1 creator, 1 product inside 1 ordinary situation, disclosure in the first line and on the label, 1 experience rather than 1 claim, 0 brand-written sentences, 1 post that survives a year on the grid

**Rules:** Give the creator the angle, never the words | Put the disclosure in the creator's own cadence, early, and on the platform label as well | Write it so it would still be a reasonable post if the product were removed | Carry the claim with one specific experience, not with a benefit statement | Let the creator keep their hedges, their tics and their sentence shapes | Where the brief plans paid amplification, note usage rights and whitelisting with the brief | Cut anything the creator would not say to their own audience unpaid | One product point, and let the caption carry the second | Check the post still makes sense a year from now | Never ask for a claim the creator cannot verify from their own use

**Examples:** paid-partnership posts on Instagram and TikTok, long-running creator ambassadorships, "things I actually use" integrations, creator-hosted product drops, newsletter and podcast host-read integrations in their video form

**Avoid:** Brand-written copy handed over to be read, which is audible in one sentence and destroys the only asset in the deal. A disclosure in a different voice from the rest of the post, which announces that everything after it was negotiated. Three product points where the creator's own posts never make more than one. A post that could be moved to any other creator's account without changing, which means the brand paid for reach it could have bought cheaper. And the ambassadorship that continues past the point the creator plainly stopped using the thing.

**The kit (ranked):**

1. **The Narrating Persona** (Voice) `voice.narrating_persona`
   The creator's on-camera self is constructed, consistent and load-bearing. Naming it stops a brief from fighting the thing it is paying for.
2. **First Person Central** (POV) `pov.first_person_central`
   The creator is the subject and the product is inside their life. A post that inverts this is a commercial hosted on a personal account.
3. **Writing to an Existing Brand Voice** (Voice) `voice.brand_voice_intake`
   Read inverted here. The intake still matters, but in this format the creator's voice wins every conflict and the brand voice is a set of boundaries rather than a style.
4. **Setting the Register** (Voice) `voice.register_setting`
   The register is already fixed by two years of the creator's own posts. The job is matching it, which is a listening task before it is a writing one.
5. **The Two-Audience Layer** (Subtext) `sub.dual_audience`
   The brand and the audience read the same post differently. Writing for the second while satisfying the first is the format's whole negotiation.
6. **The Parasocial Bond** (Relationship) `rel.parasocial_bond`
   What is actually being rented. The card is also the honest account of why a bad partnership costs the creator more than it costs the brand.
7. **The Specific Detail** (Emotion) `emo.specific_detail`
   "Three weeks and it still hasn't chipped" beats every benefit line in the brief, and it is the sentence the audience repeats.
8. **Funny Because True** (Comedy) `comedy.observational_truth`
   Recognition is how a partnership post earns its place in a feed rather than being tolerated in it.
9. **The Hedge** (Voice) `voice.the_hedge`
   "It's not for everyone, but." The qualification is what makes the recommendation worth something, and it is the first thing a brand review removes.
10. **Talking to the Room** (Dialogue) `dial.talking_to_the_room`
    Direct address to an audience the creator actually knows. The card's mechanics are different when the speaker can name the people watching.
11. **Real People and the Composite** (Character) `char.real_person_composite`
    A creator testimonial is a claim about a real person's real experience. The line between edited and fabricated is where the enforcement risk sits.
12. **The Ad's Implicit Claim** (Theme) `theme.implicit_ad_claim`
    Name what the post argues before it goes out, because in this format the implicit claim is usually stronger than anything in the caption.
13. **Rhythm in Captions and Subtitles** (Rhythm) `rhythm.subtitle_beat`
    The caption is the primary channel and also where the disclosure lives. Write both together rather than adding one to the other.
14. **The AI-Default Register** (Voice) `voice.ai_default_register`
    **This card names a failure, not a recommendation.** It is ranked here because brand-supplied creator copy lands in exactly this flattened, generic, agreeable register, and audiences now recognise it instantly. Read any brand-written line against it before it is filmed.

---

**The visual kit (ranked):**

1. **Handheld Phone Look** (Vertical) `vertical.handheld_phone`
   The post lives on the creator's feed in front of their audience. A frame that looks produced reads as an advertisement and performs like one.
2. **Talk-To-Camera** (Vertical) `vertical.talk_to_camera`
   The creator's own register, which is the thing the brand actually bought. Directing it into something else spends the asset.
3. **Face In The First Frame** (Vertical) `vertical.face_first_frame`
   The audience follows a person, not a product, and the first frame decides whether they read this as that person.
4. **On-Screen Captions** (Vertical) `vertical.captions`
   In the creator's own caption style rather than the brand's. A brand-styled caption on a creator's feed is the tell.
5. **Available Light / Natural Light** (Lighting) `light.available`
   Deliberately, not by default. Lit work signals a paid post before a disclosure does.
6. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   The creator's own layout habits usually already respect these; the brand's supplied assets usually do not.
7. **Product In Hand** (Promo) `promo.ugc_product_in_hand`
   Held, used, at the creator's own distance. A product presented rather than used is the shot the audience scrolls past.
8. **Fill The Frame** (Vertical) `vertical.fill_frame`
   Close, because it is a phone, and because the alternative is the brand's wide product shot arriving in the middle of a personal feed.
9. **Jump Cut** (Cuts) `cut.jump`
   Here it is native rather than lazy: it is how the creator's own uncommissioned posts are cut, and matching it is the point.
10. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Any legal or disclosure text is subject to this as much as the captions are, and it is where brand posts most often fail.

## Community Post

**Constraints:** Text, or text with one image, in a feed and in its own comments. It is the cheapest format in this file to make and the most expensive to get wrong, because the reply is part of the product: replying to comments measurably lifts engagement, with reported 2026 lifts around 42% on Threads, 30% on LinkedIn, 21% on Instagram and 8% on X — **figures that should be re-checked, since they are platform behaviour rather than craft.** A specific question outperforms an open one by a wide margin: "which feature do you wish we had" gets answers where "what do you think" gets none. The post is a moderation liability as much as an editorial object, since the brand is responsible for the conversation it invites. A post with five hundred comments arguing is worth more than one with five thousand likes and no replies, because the comments are the thing the ranking and the community both respond to. *(Platform figures as at 2026-09.)*

**Grammar:** A community post is an opening move in a conversation, which makes it the only format in this file whose success is measured in what other people write. So it is built backwards from the reply you want: the question is designed so that an ordinary member of this audience can answer it in one line without expertise, without risk and without having to agree with you. Specificity is the mechanism — a narrow question is easy to answer and a broad one is work. The post has to leave something undone; a complete, polished statement invites admiration rather than participation, and admiration is the metric that means nothing here. Register is set in the first line and cannot shift, because a thread that changes voice reads as two people. And the brand has to be willing to be answered, including badly, which is a writing decision as much as a moderation policy.

**Palette:** 1 claim or 1 question, 3 to 8 lines, 1 answerable ask, 1 thing left undone, 1 held register, replies as part of the draft

**Rules:** Write the reply you want first, then write the post that produces it | Ask a narrow question; a broad one is work and nobody does unpaid work in a feed | Leave something undone so there is room to answer | Say one specific thing rather than three general ones | Set the register in the first line and hold it to the last | Use line breaks as the only pacing instrument you have | Keep a hedge in — certainty closes a conversation | Plan the replies before you post, including the hostile one | Never ask a question whose answer you would refuse to publish | Cut the closing sentence that summarises what the post already said

**Examples:** brand question posts on Threads and LinkedIn, product-feedback prompts, "what did we get wrong" posts, community shout-out and member-spotlight posts, changelog posts that invite comment, founder posts on an owned channel

**Avoid:** "What do you think?" under a polished statement, which reads as an invitation to applaud. The post that is complete, correct and finished, leaving a reader nothing to add. A question the brand plainly does not want answered honestly, which the audience detects instantly and answers honestly anyway. A register that shifts from institutional to casual within four lines. And the post published with no plan for the first hostile reply, which is when the actual writing starts.

**The kit (ranked):**

1. **The Opening Question** (Openings) `open.opening_question_line`
   The format is often just this card, executed well. The card's whole subject is what makes a question answerable rather than rhetorical.
2. **Turn the Statement Into a Question** (Exposition) `expo.statement_into_question`
   The conversion the format runs on. Most failed community posts are a good statement that was never turned.
3. **The Implied Listener** (Voice) `voice.implied_listener`
   Who is imagined on the other end. A post written to "the community" in the abstract gets replies from nobody in particular.
4. **Who Asks the Questions** (Dialogue) `dial.question_control`
   Control of the question is the power structure of any exchange. Handing it to the audience deliberately is what makes a community post feel unlike marketing.
5. **Setting the Register** (Voice) `voice.register_setting`
   Fixed in line one. A thread that changes register mid-way reads as two people and the audience says so in the replies.
6. **The Specific Detail** (Emotion) `emo.specific_detail`
   One concrete particular gives readers something to agree or disagree with. Abstractions produce likes, which is the metric this format is trying to escape.
7. **The Parasocial Bond** (Relationship) `rel.parasocial_bond`
   What makes a member answer a brand at all. The card is also where the limits are, and this format finds them quickly.
8. **The One-Line Paragraph** (Rhythm) `rhythm.one_line_paragraph`
   White space is the only pacing instrument available. The one-line paragraph is this format's beat.
9. **The Fragment** (Rhythm) `rhythm.fragment`
   Speech is fragmentary and a feed reads as speech. Complete sentences throughout are how a post announces it came from a content calendar.
10. **The Confiding Voice** (Voice) `voice.confiding_voice`
    One person to one person, even from an account with a logo on it. It is the register that gets answered rather than filed.
11. **Talking to the Room** (Dialogue) `dial.talking_to_the_room`
    Direct address that assumes a room rather than an audience. The difference is audible in the first six words.
12. **The Hedge** (Voice) `voice.the_hedge`
    "I might be wrong about this." Uncertainty is an invitation; certainty is a closing statement, and a closing statement gets no replies.
13. **Descriptive, Not Prescriptive** (Theme) `theme.descriptive_not_prescriptive`
    Describe what you saw and let the community draw the rule. The moment a community post prescribes, it has stopped being a conversation.
14. **The Two-Audience Layer** (Subtext) `sub.dual_audience`
    Members and the market read the same post. Knowing the second audience exists keeps the first one from being performed at.
15. **Comedy That Requires the Audience to Already Agree** (Comedy) `comedy.requires_agreement`
    **This card names a failure, not a recommendation.** It is ranked here because the in-group post that only lands for people who already agree is this format's characteristic mistake — it feels like community and functions as a closed door. Test every insider line against it.

---

**The visual kit (ranked):**

1. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   The cheapest format in the file to make and the easiest to make unreadable. Type at phone size in a feed is the entire craft here.
2. **Negative Space** (Composition) `comp.negative_space`
   One image, one idea. The space is what makes a scrolling reader register that there is an idea at all.
3. **Colour Contrast** (Composition) `comp.color_contrast`
   A single post competes with a feed. Contrast is the only acquisition tool a static post has.
4. **Never Colour Alone** (Access) `access.not_colour_alone`
   A poll, a list or a comparison encoded in colour alone excludes part of the audience it is asking to reply.
5. **Rule of Thirds** (Composition) `comp.rule_of_thirds`
   Survives the platform's crop, which is the only composition constraint that actually binds here.
6. **4:5 Vertical** (Format) `look.ratio_4x5`
   The most feed real estate a still gets without being cropped, and it costs nothing to choose.
7. **Cover Frame / Thumbnail Frame** (Vertical) `vertical.cover_frame`
   The post is its own thumbnail. There is no second chance at a different frame.
8. **Text Safe Zone Planned in the Image** (Text) `textimg.text_safe_zone`
   Planned in the image, because a community post is screenshotted and reposted more than any other format here.
9. **Symmetry** (Composition) `comp.symmetry`
   Recognisable at a glance across a series of posts, which is how a channel builds an identity out of the cheapest thing it makes.
10. **Clean Plate Clause** (Text) `textimg.no_watermark_clause`
   Posted, reposted, translated and re-dated. Burned type is type that cannot be corrected after the first mistake.

## Instagram Reel

**Constraints:** 9:16 at 1080×1920, H.264, up to a 1 GB file. Length is where the platform's own rules bite: Reels can run to three minutes, but Instagram has said Reels over three minutes will not be recommended to new audiences, and a Reel must be under 90 seconds to be boosted from Instagram — so the practical ceiling for reach is shorter than the technical one. Roughly 85% of Meta video is watched muted, making captions the primary audio track. Ranking rewards reach, sends and profile visits, which means a Reel's job is frequently to convert a viewer into a follower rather than to be watched to the end. The cover frame sits in the profile grid permanently and is a design decision about the account, not just about the post. Trending audio is a significant boost rather than a requirement. Remix and collaborator posting are distribution mechanisms that have to be designed for rather than discovered. **All of this is platform behaviour as at 2026-09 and it is revised often — re-check the current recommendation and boost thresholds before you plan a series.**

**Grammar:** Because the reward is a profile visit, an Instagram Reel is written to make a viewer curious about the person who made it, not only about the thing in it. That is a different target from a piece optimised for completion, and it changes the ending: the close should leave a reason to look at the account rather than a reason to feel finished. The piece is seen muted first and captioned always, so it is written as a visual object with the sound as a bonus layer. The 90-second boost threshold is a real structural constraint rather than a guideline, because it decides whether paid distribution is available at all. The loop is the format's cheapest retention device and Instagram's audience is habituated to it. And the cover frame must work as a grid tile, which means the strongest image has to be one you are willing to have on the profile forever.

**Palette:** under 90 seconds for boostability, 1 idea, 1 promise, hook in second one, captions always, 1 cover frame chosen for the grid, 1 loop point

**Rules:** Cut under 90 seconds unless you are certain you will never boost it | Write for muted and add sound as a second layer | Choose the cover frame as a grid decision, not as a thumbnail decision | End on a reason to look at the account, not on a summary | Close the loop so the second pass reads differently | Make the first frame legible before anyone reads a word | Use trending audio where it fits and never build the idea on it | Design the remix or collaborator version before you post the original | One idea; the second one is next week's Reel | Never promise a payoff the piece does not contain

**Examples:** process and craft reels, transformation reveals, single-joke sketches with an audio, carousel-adjacent explainer reels, collaborator posts between two accounts, saved-and-sent educational reels

**Avoid:** A 2:40 Reel with no paid plan, which quietly removes the distribution route the brief assumed. A piece whose entire meaning is in audio that 85% of viewers never hear. A cover frame chosen in the upload screen at the last second and then living on the grid for two years. An ending that wraps up neatly and gives the viewer no reason to tap the name. And an idea built on a trending sound, which expires with the sound.

**The kit (ranked):**

1. **The First-Second Hook** (Openings) `open.first_second_hook`
   Decide the promise type before the first word, then write the hook to make it. Everything else in the piece is delivery.
2. **The Loop Opening** (Openings) `open.loop_opening`
   Write the opening so it follows the close. A loop is the cheapest additional watch time on a platform that counts it.
3. **The Loop Close** (Endings) `end.loop_close`
   The other half. The join has to be invisible, and the second pass should mean something the first did not.
4. **The Title-Thumbnail Contract** (Openings) `open.title_thumbnail_contract`
   Read it for the cover frame. On Instagram that frame is a permanent grid tile as well as an entry point, which is a harder brief than a thumbnail.
5. **The Visual Promise** (Openings) `open.visual_promise`
   This is an image-first feed. The promise has to be visible before it is audible or written, because the muted first pass is the only one most viewers give.
6. **Rhythm in Captions and Subtitles** (Rhythm) `rhythm.subtitle_beat`
   With most of the audience watching silent, caption timing is the piece's timing. Write the captions first and the voice-over second.
7. **The Stated Promise** (Openings) `open.stated_promise`
   In non-fiction short-form, saying what you will show is correct rather than a failure of nerve. The contract is the retention mechanism.
8. **Short-Form Comic Shape** (Comedy) `comedy.short_form_shape`
   The ratio that puts the payoff before the last beat. On Instagram it also has to survive being read rather than heard.
9. **Funny Because True** (Comedy) `comedy.observational_truth`
   Recognition is what gets a Reel sent to one person rather than merely watched, and sends are what the ranking rewards.
10. **The Specific Detail** (Emotion) `emo.specific_detail`
    The particular is what gets saved. On a platform that counts sends and profile visits, saveability is closer to the goal than completion is.
11. **Early Exit** (Scene) `scene.early_exit`
    Never wind down. The payoff is the last thing, and anything after it is a reason to leave rather than to tap the name.
12. **Setting the Register** (Voice) `voice.register_setting`
    A profile visit is a decision about a person. The register is what the visitor is deciding about, so it has to be the same one the grid promises.
13. **The Handoff Ending** (Endings) `end.handoff_ending`
    One action. On this surface the highest-value one is usually "look at the account", which is a different close from "buy this".
14. **The Opening That Promises a Different Piece** (Openings) `open.wrong_promise`
    **This card names a failure, not a recommendation.** It is ranked here because a hook that misrepresents the body buys a Reel one strong first second and teaches the ranking to stop serving the account. Test the hook against the payoff, not against the brief.

---

**The visual kit (ranked):**

1. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   Ranked first here because Reels is the platform this card was corrected for: it reserves more of the bottom than TikTok, which was the reverse until 2026, and four files had copied the wrong reading.
2. **Face In The First Frame** (Vertical) `vertical.face_first_frame`
   The ranking runs on early retention and nothing parses faster than a face.
3. **On-Screen Captions** (Vertical) `vertical.captions`
   Muted autoplay is the default state of this surface. The captions are the audio.
4. **Cover Frame / Thumbnail Frame** (Vertical) `vertical.cover_frame`
   A Reel lives twice: in the feed and on the profile grid. The cover frame is the one that decides the second life.
5. **Fill The Frame** (Vertical) `vertical.fill_frame`
   Shot at arm's length in daylight on a screen one hand wide. Every wide is a wasted second.
6. **Transition On The Beat** (Vertical) `vertical.transition_on_beat`
   The surface is built around audio, and a transition off the beat reads as a mistake even to a viewer with the sound off.
7. **Loop Point** (Vertical) `vertical.loop_point`
   A loop turns one play into several, and plays are what the ranking counts.
8. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   Judged on the phone, outdoors. Not on the timeline.
9. **Punch In** (Vertical) `vertical.punch_in`
   The cheapest mid-clip change of state, and the one that survives the platform's own compression.
10. **End Call to Action** (Vertical) `vertical.end_cta`
   Placed inside the safe area rather than at the bottom of the frame, which is where it is covered by the audio attribution bar.

## TikTok

**Constraints:** 9:16 at 1080×1920; in-app recording up to 10 minutes and uploads up to 60 on some accounts, with file limits that split by route — roughly 287.6 MB from a phone against up to 4 GB through the desktop studio. Ranking rewards saves, shares and watch-through, and the comment section drives re-serving, so a piece that provokes a reply outperforms a piece that merely completes. Trending audio is closer to essential here than on the other surfaces. Sound is on more often than on Meta, but captions remain standard. Stitches, duets and reply-videos mean somebody else writes the second draft in public, and the format rewards work that leaves room for that. The comment section is part of the text, and the creator's replies are a second publication. **Platform behaviour as at 2026-09; the ranking signals and the upload limits are both revised regularly.**

**Grammar:** TikTok is the surface where the piece is a turn in a conversation rather than a broadcast, which is why the strongest structure is one that ends slightly unresolved in a way that invites an answer. That is not a cliffhanger — it is a deliberate gap in the argument, an unstated caveat, an obvious objection left unaddressed, and the comment section closes it. The opening is nearly always mid-reply: the piece enters on an answer to a question nobody heard, which costs nothing and is instantly legible. "Wait for it" is a native and effective move here, which makes this one of the few places in the database where announcing your own payoff is correct. The voice is close, fast and specific, and the specific detail is the share mechanism. And the loop matters as much as it does on Reels, because watch-through is counted and a seamless loop doubles it for free.

**Palette:** 15 to 45 seconds typically, 1 idea, 1 unresolved gap for the comments, hook in second one, 1 loop point, trending audio where it fits

**Rules:** Enter on the answer; the question is the slowest possible opening | Leave one obvious objection unanswered and let the comments raise it | Announce the payoff when the payoff is worth waiting for, and only then | Write for sound-on but caption it anyway | Build the piece so a stitch or a duet has something to do | Make the last second loop into the first | One specific detail nobody else would have said | Reply in the comments as part of the publication, not after it | Never use a trending sound the idea does not need | Cut two seconds before it feels safe

**Examples:** explainer and "actually" correction videos, single-joke sketches, stitch-and-respond chains, POV and skit formats, process and craft demonstrations, comment-reply video series

**Avoid:** A piece with nothing to say back to, which completes and disappears. A preamble explaining what the video will contain, which on this surface is the single most expensive three seconds available. "Wait for it" attached to a payoff that is not worth waiting for, which is the fastest way to train an audience to scroll past your account. A trending sound carrying an idea that had nothing to do with it. And an ending that just stops, when a loop was available for free.

**The kit (ranked):**

1. **Enter on the Answer** (Openings) `open.enter_on_the_answer`
   The cheapest legitimate first second in the craft, and the native opening of this surface. The piece starts mid-reply to a question nobody heard.
2. **The First-Second Hook** (Openings) `open.first_second_hook`
   Under two seconds, spoken, specific. Choose the promise type — result, reveal, lesson or tension — before the first word.
3. **Telegraphing** (Information) `info.telegraphing`
   **Ranked deliberately, and it is a defect nearly everywhere else in this database** — announcing your own reveal normally spends it. TikTok is an exception: "wait for it" works because the platform's currency is watch-through rather than surprise, so the announcement buys the seconds it costs. Read the card for the mechanism, then use it only where the payoff justifies the wait; the same move on a weak payoff is what kills accounts.
4. **The Deliberately Unanswered Question** (Information) `info.deliberate_unanswered_question`
   The gap the comment section fills. On a surface where comments drive re-serving, a piece that answers everything has closed its own distribution.
5. **Funny Because True** (Comedy) `comedy.observational_truth`
   Recognition is the share mechanism and the reply mechanism at once. It is the native mode here more than on any other surface.
6. **Tension in the First Five Seconds** (Tension) `tens.first_five_seconds`
   Establish the stake, not the world. There is no world-building budget and the audience does not want one.
7. **Talking to the Room** (Dialogue) `dial.talking_to_the_room`
   Direct address to a specific person rather than an audience. On TikTok the second-person address is the default register, not a device.
8. **The Confiding Voice** (Voice) `voice.confiding_voice`
   One person to one person. It is why a phone-camera piece outperforms a produced one on this surface with the same content in it.
9. **The Specific Detail** (Emotion) `emo.specific_detail`
   The particular nobody else would have said. It is what gets saved, and saves are weighted here more heavily than anywhere else.
10. **Misdirection and the Reveal** (Comedy) `comedy.misdirection`
    The turn has to arrive before the thumb does. In a 20-second piece the misdirection is usually two seconds long and the reveal is one.
11. **The Funny Word Goes Last** (Comedy) `comedy.funny_word_last`
    Word order is the only timing instrument that survives a caption. It is also what makes a line quotable in a comment.
12. **Early Exit** (Scene) `scene.early_exit`
    Cut on the turn, two seconds earlier than feels safe. The wind-down is where watch-through goes to die.
13. **The Loop Close** (Endings) `end.loop_close`
    A seamless join doubles watch-through for free on a surface that counts it. Design it; do not discover it.
14. **Rhythm in Captions and Subtitles** (Rhythm) `rhythm.subtitle_beat`
    Caption timing is the piece's timing even where sound is on, because the caption is what a rewatcher reads.
15. **Conversational Persona Writing** (Interactive) `inter.conversational_persona`
    The account is a persona that responds. Comment replies are publication, and this card is the nearest thing to a brief for them.

---

**The visual kit (ranked):**

1. **Start Mid-Action** (Vertical) `vertical.start_mid_action`
   The surface punishes a setup harder than any other. The piece begins after the beginning.
2. **On-Screen Captions** (Vertical) `vertical.captions`
   Native to the platform's own grammar, not an accessibility addition. A TikTok without them reads as imported.
3. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   TikTok reserves less at the bottom than Reels and more on the right rail. The figures and their date are on this card.
4. **Face In The First Frame** (Vertical) `vertical.face_first_frame`
   Still the fastest parse, and still the first frame's only reliable job.
5. **Pattern Interrupt** (Vertical) `vertical.pattern_interrupt`
   Retention is lost in the middle. A change of scale or texture every few seconds is the whole retention craft of the format.
6. **Handheld Phone Look** (Vertical) `vertical.handheld_phone`
   The register the surface rewards. Produced-looking work underperforms here in a way it does not on Reels.
7. **Sound Sync** (Vertical) `vertical.sound_sync`
   The platform is a sound platform that shows pictures. Every visual event carries its own hit.
8. **Stitch** (Vertical) `vertical.stitch`
   A distribution mechanism disguised as an edit. Designing the clip to be stitched is designing for reach.
9. **Loop Bait Ending** (Vertical) `vertical.loop_bait_ending`
   The ending that sends the viewer back to the start, which the ranking reads as a completed watch twice over.
10. **Fill The Frame** (Vertical) `vertical.fill_frame`
   Closer than feels right, always, for the same reason as every other vertical format.

## YouTube Shorts

**Constraints:** 9:16 at 1080×1920, up to three minutes, through the same upload pipeline as long-form. Ranking runs on click-through and average view duration rather than on sends or saves, and Shorts has the strongest search discoverability of the three surfaces — which makes the words unusually load-bearing, in the title, in the spoken script and in what the automatic transcript contains. A Short sits inside a channel rather than standing alone, so it is measured against the channel's long-form retention and the subscriber is the asset. The viewer may arrive months later through search, which is a materially different audience from a feed scroll. The same piece frequently exists to lead somewhere: a long-form video, a playlist, a product. **Platform behaviour as at 2026-09; the length ceiling, the ranking signals and the Shorts-to-long-form surfacing all change.**

**Grammar:** Shorts is the surface where the piece behaves most like a small YouTube video, which means the contract rather than the hook is the governing object: a promise made explicitly at the top and paid exactly at the end, with the title as the contract's written form. Because search is real, the words have to be the words people type — a Short that answers a question in language nobody uses is a Short nobody finds. Average view duration is the metric, so the piece is structured to stay rather than to provoke, and the staged reveal is worth more here than a gap left for the comments. The close is a handoff rather than a summary, and on this surface the highest-value handoff is usually to a longer video the viewer is now qualified to want. And the register has to match the channel, because a Short is an audition for the subscription and the subscription is the point.

**Palette:** 30 to 60 seconds typically, up to 3 minutes available, 1 question in the words people search, 1 promise paid exactly, 2 to 3 steps, 1 handoff to the channel

**Rules:** Write the title as the contract and answer that exact question | Say the search words aloud; the transcript is part of the discovery | Make the promise in the first three seconds and pay it precisely | Build in two or three steps so each part of the runtime pays | Structure for duration, not for provocation — the metric here is how long people stay | Match the channel's register; the Short is an audition for the subscription | Hand off to the long-form video rather than summarising | Never answer a question adjacent to the one the title asked | Caption it even though sound is more often on | Cut the wind-down; the payoff is the last thing

**Examples:** channel-clip Shorts cut from long-form, standalone how-to and answer Shorts, question-format explainer Shorts, process and build Shorts, search-driven evergreen answer Shorts

**Avoid:** A Short that answers a related question rather than the one in the title, which is the most damaging thing a channel can do repeatedly on a surface where search brings the viewer. A piece phrased in the channel's internal vocabulary rather than in the words people actually type. A tone that has nothing to do with the channel's long-form, so the subscription the Short earns immediately churns. A cliffhanger with no next episode behind it. And a summary ending that spends the last five seconds repeating the piece instead of handing the viewer somewhere.

**The kit (ranked):**

1. **The Stated Promise** (Openings) `open.stated_promise`
   Explicit, in the first three seconds, and paid exactly. On a surface ranked by duration, the contract is the retention mechanism.
2. **The Title-Thumbnail Contract** (Openings) `open.title_thumbnail_contract`
   The title is what the search returns and what the click is made on. It is the piece's most-read sentence and usually the least-written one.
3. **Answer the Question That Was Asked** (Endings) `end.answer_the_question_asked`
   The title asked it; answer that one. Answering the adjacent question is what teaches a search audience not to trust the channel.
4. **The Question Planted First** (Openings) `open.question_planted_first`
   One specific unresolved thing carries the viewer until the answer can. It is what makes an explainer Short hold for fifty seconds rather than nine.
5. **Short-Form Front-Loading** (Exposition) `expo.short_form_frontload`
   Whatever is needed to read the answer goes in the first three seconds. There is no room to explain anything twice.
6. **The Staged Reveal** (Information) `info.staged_reveal`
   Break the payoff into two or three steps so every part of the runtime pays. On a duration-ranked surface this is worth more than a single big reveal.
7. **Authority by Specificity** (Voice) `voice.authority_by_specificity`
   The number, the name, the date. It is also the language a search viewer used to arrive, so precision serves discovery and credibility at once.
8. **Theme in Non-Fiction** (Theme) `theme.theme_in_nonfiction`
   State the claim to yourself before the cut, or the strongest available footage will decide what the Short argues.
9. **The Confiding Voice** (Voice) `voice.confiding_voice`
   The register the platform was built on. A Short is an audition for a relationship, and this is what the audition is for.
10. **Writing for the Ear** (Voice) `voice.write_for_the_ear`
    Sound is on more often here, and the transcript is part of discovery. Read it aloud once; anything you stumble on is also a line the transcript garbles.
11. **Rhythm in Captions and Subtitles** (Rhythm) `rhythm.subtitle_beat`
    Still mandatory. Caption line length is also what a viewer scanning the first two seconds reads before deciding.
12. **Early Exit** (Scene) `scene.early_exit`
    The payoff is the last thing. A wind-down on a duration-ranked surface costs exactly what it looks like it costs.
13. **The Handoff Ending** (Endings) `end.handoff_ending`
    One action, and here it is usually the long-form video the viewer is now qualified to want. A summary in its place wastes the one move this surface rewards.
14. **The Callback** (Endings) `end.callback`
    Returning the opening line at the close signals the piece was designed, which on a channel is a claim about everything else you make.
15. **The Answer to a Question Never Asked** (Information) `info.answer_to_unasked_question`
    **This card names a failure, not a recommendation.** It is ranked here because Shorts is search-ranked, which makes it the one surface where you can verify the failure empirically: if nobody types the question, the Short is an answer to nothing. Check the phrasing before you shoot.

---

**The visual kit (ranked):**

1. **Cover Frame / Thumbnail Frame** (Vertical) `vertical.cover_frame`
   Ranked first because Shorts runs on click-through in a way TikTok does not - it shares a pipeline with long-form and inherits its thumbnail logic.
2. **Face In The First Frame** (Vertical) `vertical.face_first_frame`
   Retention decides the rest, and the face is what buys the first three seconds.
3. **On-Screen Captions** (Vertical) `vertical.captions`
   Muted first pass, same as every vertical surface.
4. **UI Safe Zones** (Vertical) `vertical.safe_zones`
   Shorts carries a different interface again, and the figures belong to the card that owns them and dates them.
5. **Title Card at Zero** (Vertical) `vertical.title_card_zero`
   Shorts is searched as well as scrolled, which makes an explicit statement of subject in frame one worth more here than elsewhere.
6. **Fill The Frame** (Vertical) `vertical.fill_frame`
   Up to three minutes, still one hand wide. The length does not widen the screen.
7. **Pattern Interrupt** (Vertical) `vertical.pattern_interrupt`
   A three-minute Short has a middle, which is more than most vertical formats have to defend.
8. **Punch In** (Vertical) `vertical.punch_in`
   The second size on a one-camera shoot, and the only one available to a Short cut from long-form material.
9. **Loop Point** (Vertical) `vertical.loop_point`
   Works here as everywhere, and matters more because the surface counts the replay.
10. **Legible at the Size It Is Actually Watched** (Access) `access.legible_at_size`
   A Short is also watched on a television through the same app, which is the one vertical surface where that happens.

## Coverage

13 format kits: Vertical Micro-Drama, Reels/Shorts/TikTok (shared craft), Long-Form Video (YouTube), Web Series and Serialised Short-Form, Carousel/Thread/Text Post, Podcast Clip, Livestream Script, Branded Serial, Creator Brand Partnership (Organic), Community Post, Instagram Reel, TikTok, YouTube Shorts.

Three adjustments to the target list, all made because the honest count mattered more than the listed one.

**"UGC ad" was already written, in another file.** The paid creator ad is covered in full as **UGC and Creator Ad** in [`symptoms-commissioned-and-applied-work.md`](symptoms-commissioned-and-applied-work.md), and duplicating it here would have produced two kits drifting apart on the same subject. What this file was missing is its sibling: the *organic* paid-partnership post, which has no media spend behind it, lives on the creator's own feed in front of the creator's own audience, and therefore has a different failure mode entirely. That is written here as **Creator Brand Partnership (Organic)**, and it is deliberately ranked against the promo kit rather than duplicating it.

**"Branded serial" overlaps with Web Series and Serialised Short-Form and is still its own kit.** They share the episode-closes / series-opens architecture, and nothing else: a branded serial's distribution is bought rather than earned, its script is reviewed by a legal function that is reading fiction, and its characteristic failure is a line given to a character because somebody in the approvals chain wanted it said.

Remaining social and short-form targets: the newsletter and owned-audience post, the LinkedIn long post, the live-shopping script, the Discord and forum announcement, the comment-reply video series as a format in its own right, and the platform-native ad unit that is neither a Reel nor a UGC ad.

**Card-shaped gaps found while writing these kits** — techniques these formats need that no card yet covers, named in prose above rather than linked:

- **Writing the reply as part of the post.** In a community post and a comment-reply series the published object includes the author's own replies. Nothing in the database treats the response as a drafted unit.
- **Finding rather than writing.** The podcast-clip kit is built on searching a transcript for a moment that already exists. `rhythm.transcript_clean` covers what to do with the words; no card covers the search itself.
- **The disclosure written as content.** Also recorded as a gap in the commissioned-advert material, and it bites harder here: in an organic partnership post the disclosure has to sound like the creator or the whole post is exposed.
- **The platform contract that expires.** A card-shaped need for the discipline of writing so that a piece survives a ranking change — separating the part of a format that is craft from the part that is a current spec.
- **Designing for the stitch.** Leaving structural room for someone else to reply in public, which is a real TikTok writing move with no card behind it.

**Shelf life:** dated — review 2027-09. The craft cards these kits rank are durable; the constraint blocks and the ranking order rest on 2026 platform behaviour, and the three platform-variant kits are the fastest-decaying entries in this file.

**Shelf life:** dated — review 2027-09. The craft cards these kits rank are durable; the constraint blocks and the ranking order rest on 2026 platform behaviour.

*Build notes from when this file was drafted — targets, running totals and what remained to be written — are kept verbatim in the [coverage history](../tools/history/coverage-history.md). They were true of the draft, not of the file, so they no longer sit here as if they were measurements.*

## Sources

Verified 2026-09-17.

- [How to Write a Vertical Drama Script in 2026 — Filmustage](https://filmustage.com/blog/how-to-write-a-vertical-drama-script/)
- [Vertical Short Drama Script Handbook — Maosika](https://www.maosika.com/blog/vertical-short-drama-script-handbook-2)
- [Script Structure for Vertical Dramas — Axis AI Studios](https://www.axisaistudios.com/blog/script-structure-for-vertical-dramas-episode-by-episode-guide)
- [Short Drama Script Format — ScreenWeaver](https://www.screenweaver.ai/blog/short-drama-script-format)
- [Short-Form Video Structure: Hook, Body, Payoff — Socialync](https://www.socialync.io/blog/short-form-video-structure-guide-2026)
- [Video Hooks That Stop Scrolling, 7 Patterns — Green Frog Labs](https://greenfroglabs.com/blog/video-hooks-scroll-stopping-2026)
- [How to Optimize Short-Form Video in 2026 — Jetfuel](https://jetfuel.agency/how-to-optimize-short-form-video-content-for-success/)
- [Writing Hooks and Cliffhangers for AI Micro-Drama — InVideo](https://invideo.io/blog/ai-micro-drama-hooks-cliffhangers/)

Added 2026-09-17 for the podcast-clip, livestream, branded-serial, creator-partnership, community-post and platform-variant kits.

- [How to Make Podcast Clips That Get Shared (2026 Guide) — Fame](https://www.fame.so/post/ultimate-podcast-clip-guide)
- [Audiogram Best Practices for Social Media (2026 Guide) — Recast Studio](https://recast.studio/blog/audiogram-best-practices-for-social-media)
- [How to Repurpose a Podcast into Short-Form Clips (2026 Workflow) — Choppity](https://www.choppity.com/blog/how-to-repurpose-podcast-into-shorts/)
- [Audiogram Clips: Turning Podcast Interviews Into Social Assets — Influencers Time](https://www.influencers-time.com/audiogram-clips-turning-podcast-interviews-into-social-asset/)
- [Livestream Script and Hosting Playbook: Proven Templates That Drive Sales (2026) — Confetti](https://www.confettisg.com/insights/livestream-script-hosting-playbook)
- [Run of Show Template and Best Practices for Live Streaming — Wave.video](https://wave.video/blog/run-of-show-template/)
- [Live Stream Script Template (With Examples) — Be.Live](https://blog.be.live/live-stream-script-template-with-examples/)
- [Branded Entertainment Will Just Be Entertainment in 2026 — Fast Company](https://www.fastcompany.com/91462980/branded-entertainment-will-just-be-entertainment-in-2026)
- [Branded Entertainment Category — The One Show 2026](https://oneshow.org/categories/branded-entertainment/)
- [TikTok Branded Minidramas 2026: Retail Media Storytelling Checklist — Crescitaly](https://blog.crescitaly.com/tiktok-branded-minidramas-2026-retail-media-storytelling-checklist/)
- [The Rise of Episodic Content: Should Your Brand Launch a Social Media Content Series? — Sprout Social](https://sproutsocial.com/insights/social-media-content-series/)
- [Episodic Creator Series: Brand Briefs, Rights and Measurement — Influencers Time](https://www.influencers-time.com/episodic-creator-series-brand-briefs-rights-and-measurement/)
- [UGC Creator Brief Template (2026): The Exact Doc to Get Scroll-Stopping Content — Cinerads](https://www.cinerads.com/blog/ugc-creator-brief-template)
- [What a Winning UGC Brief Actually Looks Like — Insense](https://insense.pro/blog/ugc-brief)
- [UGC Brief Template: Eight Sections That Actually Work — Tik-Joy](https://tik-joy.com/blog/ugc-brief-template/)
- [Social Media Community Management: The Definitive Guide for 2026 — Planable](https://planable.io/blog/social-media-community-management/)
- [Community Management in 2026: How to Build a Loyal and Engaged Audience — Ampfluence](https://www.ampfluence.com/community-management-2026-how-to-build-and-nurture-loyal-followers-online/)
- [How to Master Community Engagement on Social Media — Socialinsider](https://www.socialinsider.io/blog/community-engagement-social-media/)
- [YouTube Shorts vs TikTok vs Instagram Reels (2026): Full Comparison — ShortSync](https://www.shortsync.app/resources/youtube-shorts-vs-tiktok-vs-reels-2026)
- [Best Video Length for Shorts, Reels and TikTok (2026) — Reap](https://reap.video/blog/best-video-length-youtube-shorts-reels-tiktok)
- [How Long Can an Instagram Reel Be in 2026? — Ordinal](https://www.tryordinal.com/blog/how-long-can-an-instagram-reel-be)
- [TikTok Video Size and Dimensions 2026 (Resolution and Specs) — ScreenSnap](https://www.screensnap.pro/blog/tiktok-video-size-guide)
- [YouTube Shorts Specs 2026: Dimensions, Aspect Ratio and Length — Xroad Studio](https://xroadstudio.com/platform-specs/youtube)
- [YouTube Shorts vs Instagram Reels (2026): Which Wins for Brands? — Hammerhead](https://www.hammerhead.global/blogs/youtube-shorts-vs-instagram-reels-what-you-must-know)