Video production guide: style, frames and editing

How to read this

The ads come from five niches but one school: a dubbed or AI-voiced person talks to a phone in a real room about a kitchen ritual (oats, ginger, honey, lemon, syrup) that beats a named drug, with raw phone footage of the ritual and of real bodies cut in. The pages that follow describe what they share, not any one ad. Where ads disagree, the ads with the higher feel score (closer to a real post) win the argument, because the notes are consistent that polish costs trust.

Three terms used throughout. Feel score is the analyst's 1 to 10 rating from amateur phone post (1) to broadcast polish (10); the winners sit at 3 to 4 and the best at 6 to 7, so the target is low. Split is the 9:16 frame cut horizontally at the midline, presenter above, hands below. Chunk is one caption event of one to three words.

• Weight loss sets: 22-09 (15 ads, oats), Criativos OT (15, pancakes and oats), Slimjara (50, oats and ginger).
• Memory sets: MemoGuard (28, honey), 31-07 (15, honey, older presenters), 16-09 (15, honey jars), Top 1 to 5 (5, the client's own best performers).
• Men's health sets: revisados (15), CRIATIVOS (15), criativos-14-09 (15): ginger and honey, young women as presenters.
• Our ad: Kristian, 71, Norwegian, in his kitchen, golden juniper syrup over pancakes, aimed at Americans over 60. The closest models are the Top 1 to 5 ads, the 31-07 older presenters and the MemoGuard hallway testimonial, so those are cited most.

What the winners measure

Computed from the 188 notes by scripts/meta_video/ad_study_digest.py (medians, with the middle half and the full range). Treat them as the envelope, not the target: the targets are in the pacing section.

The recurring formats

Six formats cover 180 of the 188 ads. They differ in how the ritual B-roll meets the presenter: as a split, as a cold-open montage, as props in hand, as ping-pong cuts, as a jump-cut selfie, or as text instead of footage. Click any still for the full note.

1. Split-screen recipe hook into a kitchen talking head (about 60 ads, the dominant format)

For the first 4 to 13 seconds the frame is split at the midline: presenter chest-up in the top half talking down the lens, a locked top-down phone shot of hands doing one ritual step in the bottom half, the caption sitting on the seam. The split holds until the hook sentence ends, then hard-cuts to the presenter full frame. From there the body is talking head with 1 to 3 second recipe bursts and 2 to 9 second proof clips cut in full frame. Presenter always on top; the one inverted case reads wrong. The 16-09 set softens the seam into a 50/50 blend, face above and jar below, both layers half transparent through the middle, held 4 to 9 s; two of the three feel-6 memory ads open that way.

• Organic because: the bottom clip is a different phone, different white balance, hands only, clutter in frame; the top is a propped phone at chest height with window light.
• Controlled: the cut lands exactly on the last word of the hook sentence; the caption never leaves the seam; the bottom clip is already mid-action at frame 0.

2. Cold-open recipe montage, presenter arrives at 3 to 9 seconds (about 30 ads)

Frame 0 is hands and food full frame with the first spoken word already playing: oats poured into a glass bowl, Coke over raw ginger, egg yolks with powder, honey over lemon. Five to eight clips at about one second each, cut on the spoken beat, each carrying one caption chunk, then a hard cut to the presenter. The same clips come back for 3 to 5 seconds at the reveal beat.

3. The presenter holds the jar: props as B-roll (about 15 ads, including the client's number one)

One locked phone, the presenter head-to-waist at a kitchen counter, the finished recipe in a glass jar in hand for the whole ad, the hero ingredient (honey bear, syrup bottle) in the other hand. Zero to ten cuts; the variation is swapping the prop toward the lens every 5 to 8 seconds. Top 1 (49 seconds, no cuts) is the client's best performer; 22-09 AD 08 and Criativos OT_05 do the same with a Quaker box and a vinegar bottle.

4. Anchor talking head with B-roll ping-pong (about 25 ads, men's health)

A presenter on a tripod or lightly handheld phone speaks 3 to 6 seconds, then the edit hard-cuts to 1 to 4 seconds of scavenged phone footage (ginger chopped, honey poured, a reaction face), then back. 13 to 18 cuts a minute, average shot 3 to 4 seconds, no hold past 8 seconds after the hook. The pattern is what keeps a weak lip sync watchable.

5. Selfie confession with jump cuts (about 15 ads, the highest feel scores)

One person, handheld or propped phone, chest-up, talking down the lens for two to three minutes with a jump cut every 3 seconds and no dead air. The hook is a 4 to 9 second recipe montage or a contrarian first line. OT_04 H1 (feel 7) and the Slimjara living-room creator (feel 6) are the two most native-looking ads in the study and both are this format.

6. Kinetic text on black (about 20 ads): seen, not recommended

A 3 to 17 second shock clip then 60 to 160 seconds of white and yellow words on a black or wireframe grid. Feel 2 to 4, the lowest in the study, and the notes flag it as the format most likely to read as an ad. The one use worth keeping is the 22-09 AD 09 tail: the last 30 seconds (authority, urgency, CTA) on a black card in the same caption style as the body.

The hook on video: the first ten seconds

The script hook is written to the four-element formula in D2 (superstructure, big idea, selection element, curiosity loop) and D3's rule that the first sentence carries 80 percent of the weight. On video the winners add a fifth element: a picture that is already moving at frame 0 and does not match the seriousness of the sentence.

Frame 0, ranked by how often the winners use it

1. Presenter in a kitchen with the ingredient physically in hand or on the counter, already mid-sentence (the most common first frame in every set).
2. Split: presenter above, hands mid-pour or mid-whisk below, caption on the seam.
3. Hands only, full frame, one ritual step in motion: pour, squeeze, grate, whisk.
4. A stranger's raw body shot in a bedroom or bathroom (weight loss only).
5. A strange prop action: honey poured over an anatomical model, a bottle shoved at the lens, a man in the rain pouring honey on ginger.

Seconds 0 to 3

• No logo, no title card, no fade. The first spoken word lands at 0.0 s and the first caption chunk is on screen at frame 0 with its key word already boxed.
• Something is moving: a sticky pour, a whisk, a hand reaching into a jar. Static presenter plus static caption is the weakest opener in the set.
• The presenter is looking down the lens, phone at chest height, slightly below the eyes, so the glasses and face fill the middle third.
• The caption carries the pain word or the drug name in the box within the first second (DIABETES, OZEMPIC, DEMENZ, PHARMAINDUSTRIE), so a muted viewer gets the topic.

Seconds 3 to 10

• Montage-first hooks: the clips keep coming at 1 s each to 5 to 9 s, then a hard cut to the full-frame presenter.
• Presenter-first hooks: the montage lands at 6 to 10 s (one second per shot, five to eight shots), then back to the presenter.
• Split hooks: the split holds until the hook sentence ends (4 to 13 s), the bottom clip changing once or twice, then collapses to full frame exactly on the last word.
• By 7 to 15 s the presenter says who they are or what they will show, and the first proof or pain clip is in by 15 to 25 percent of runtime.

Why they stop the scroll (the five reasons that recur in every set)

1. A cheap pantry item set against an expensive named drug in the same sentence while the item is physically in frame (oats versus Mounjaro, honey versus Aricept).
2. A serious or taboo sentence over a calm, oddly satisfying kitchen action; the brain has to reconcile the two.
3. A specific demographic and a specific number in the first words: after 45, over 60, 21 million views, 18 million euros, below 90.
4. A face the viewer's own age, badly lit, talking to a phone: it reads as a neighbour's post, not an ad.
5. A contrarian or forbidding command aimed at the viewer (don't watch this, no honey for breakfast, what are you doing) that qualifies and provokes at once.

The seven hook visuals worth copying, best first

The caption system

Captions are present in 85 percent of the ads and every uncaptioned ad scored 2 out of 10. The system is the same across all ten sets with only the highlight colour changing, so it is the one piece of polish that is allowed and expected.

• Avoid: italic captions at the bottom (MemoGuard AD 5, reads cheaper); full sentences on screen except in a held 8 s opener; machine-translation errors left in ('I AM STORM' for 'im Sturm' in 22-09 AD 06), which means every caption file is read before it ships.

Pacing by phase

The ads run 60 to 230 seconds with a median around 145; the best-feeling ones are 59 to 90 seconds or 160 seconds with jump cuts. Our short script is 2:10 at 2.3 words a second, which sits in the middle of the range. The table is the consensus rhythm of the twelve-element structure as it plays on screen.

• Target 9 to 13 cuts a minute across the ad, with the first six cuts inside the first 6 seconds in a montage hook.
• Longest permitted hold: 13 to 15 s, and only in the mechanism beat.
• A recipe shot held past 3 s reads as padding; the 20 to 41 s stirring shots in Criativos OT_01 and OT_03 were marked draggy.
• Digital punch-ins on a static presenter (two framings from one take) count as cuts and are the cheapest cut there is.
• Hard cuts only. No transitions, no wipes, no zoom transitions; one flash cut at most.

B-roll: what to shoot and how to put it in

The ritual footage is the second asset after the presenter, and the notes are unanimous about what makes it work: hands only, a phone on the counter, one step per clip, nothing styled. The polished elements have a narrow allowed list.

The ritual shot ladder (what the winners actually show)

1. The ingredient raw and recognisable: the jar or bottle in frame with its label, the fruit cut on a board, the oats in the bag.
2. The pour: a continuous ribbon of the liquid from the jar or bottle into or over the food, shot top-down or at 45 degrees, bowl or plate filling 60 to 100 percent of the frame.
3. The second ingredient going in: a pinch, a squeeze, a spoon of powder.
4. The stir or the fork: motion in the bowl for 1 to 2 s with steam or drips.
5. The finished thing held up or carried toward the lens: the jar, the plate, the mug.
6. Optional: the first bite or sip, mouth closed, one second.

Three ways to insert it (and the one that is never used)

• Full-frame hard cut: 70 percent of inserts. 1 to 3 s each in runs of three to eight.
• Top-bottom split, seam at 50 percent, presenter always on top, B-roll edge to edge below, caption on the seam: the hook (first 3 to 13 s) and, in the Top ads, again every 10 to 15 s through the body.
• Semi-transparent superimposition: 0.3 to 0.5 s flashes of the ritual over the presenter's face in the first 8 s (31-07 AD05G HK1); a faint presenter laid over the B-roll in Top 5; and the 16-09 blend, face above and jar below with a soft transparent seam, held 4 to 9 s in the hook only.
• Never a small picture-in-picture box: it appears in four of 188 ads and none of the good ones.

The organic checklist (what the camera must get wrong)

• Phone propped on the counter or held in the other hand; slightly tilted horizon; auto exposure; no slider, no gimbal, no rack focus, no grade.
• Kitchen window light with blown highlights, or a yellow overhead bulb; shadows from the phone hand are fine.
• Clutter in frame: a tea towel, a measuring jug, a blue plastic bowl, cling film, the salt box, the butter dish.
• Glass bowls and a cheap metal sieve, not styled crockery; drips left on the jar; sleeves and un-manicured hands in shot.
• Mixed white balance between consecutive clips: the pour warm, the stir cool. Clips from two different kitchens were read as more real, not less.
• Slight focus hunt and motion blur on the pour. The one thing to add in the edit is the sound of the pour; the winners have none and one note asks for it.

Polished elements that are allowed

Proof beats

Visual proof is thinner than the scripts suggest. The winners speak their numbers (21 million views, 100,000 men, Harvard) and show candid phone footage of real people, never a produced testimonial.

• Weight loss: UGC body clips (waistband pulled away, oversized jeans, mirror selfies) at 7 to 9 s each in 22-09 and 1 to 3 s bursts in Slimjara; first burst at 2 to 25 percent of runtime, second at 55 to 65 percent.
• Memory and older audiences: a confused elderly person in a doorway or at a mirror (3 to 8 s), a family photo pasted for about 4 s, the presenter's own 'before' story spoken; placed at 40 to 50 percent and again before the first CTA.
• Men's health: no testimonials at all; reaction faces of women stand in for proof (2 to 5 s each).
• A pasted quote or screenshot overlay runs 2 to 4 s; an embedded testimonial clip 10 to 15 s is the longest proof element seen.

CTA staging

• First CTA at 60 to 77 percent of runtime, then an open loop (one more secret, the three foods that make it worse, the free video gets deleted), then the final CTA in the last 5 to 15 seconds. Single-CTA ads place it at 80 to 93 percent.
• The presenter delivers it in the same framing as the body, leaning in and pointing down with the index finger on the word button; the caption boxes CLICK, BUTTON or NOW.
• Nothing after the last spoken word in most sets: the video ends on the presenter mid-frame. There is no end card in 150 of 188 ads.
• Arrows: Top 1 and Top 2 animate three stacked white chevrons under the caption on the CTA line (about 15 percent of frame width); MemoGuard AD 15 overlays three red arrows for the last 13 s. Both are the client's own high performers, so the chevrons are worth adding.
• The orange button appears only in the men's health sets and it is not a card: two pills overlaid on live footage for the last 8 to 20 s. Lower pill MEHR ERFAHREN (learn more) in black heavy caps on an amber-orange rounded rectangle about 75 percent of frame width by 7 percent of height, centred at 82 percent height, thin darker edge, soft shadow; upper pill HEUTE KOSTENLOS (free today) in the same style, 65 percent wide, at 15 percent height. Captions keep running between the pills. Optional for us; the Top ads' chevrons are the closer model.
• Wording the winners use: click the button below and watch it now; watch it while it is still online; before the video is deleted; tap learn more. Our hard rule keeps the button 'somewhere on your screen', never 'below', so the arrows carry the direction instead.

Voice, music, sound

• AI voice in 95 percent of the ads, dubbed over a real actor or an AI avatar. The handful of real reads (31-07 AD05G HK1, MemoGuard AD05G HK01, Slimjara AQPdht) score highest for trust in every note.
• Lip sync is the number one trust leak across all ten sets. The winners hide it with cuts every 3 to 6 seconds; a Kling render is only as good as its mouth, so the face crop QA stays and the UGC beats are cut short.
• Breaths, micro-pauses and a mispronunciation left in read as human; a continuous no-breath delivery is the most common AI tell. ElevenLabs settings should allow pauses, and the script's ellipses are breaths.
• Music: none in 78 percent. Where present it is a low bed under the hook only. Start with none.
• Sound effects: none in the study. The one addition the notes ask for is the natural sound of the pour and the fork on the plate under the ritual clips.
• One voice for the whole ad, first person, steady and serious, slightly fast; captions compensate for anything missed.

The older presenter: how Kristian is framed

Drawn from the 31-07, MemoGuard and Top sets, where the presenters are in their sixties and seventies and aimed at a 60-plus viewer.

The timeline template for a two-minute Edge Method ad

The production plan: what asset sits where, for how long, and which winner proves it. Times are for a 2:10 body at 2.3 words a second (our ad 01 short); scale the percentages for other lengths. Shot codes: UGC is the presenter render, B-roll is phone footage of the ritual or of people, Render is the polished insert, Card is text.

The asset list per ad

Everything that has to exist before the edit, with the tool and the spec. Approvals happen on the production sheet; nothing is generated until the storyboard is approved.

What breaks the feel

1. A single presenter take longer than 13 to 15 s anywhere, and especially the closing pitch (40 to 182 s uncut takes in every set were called the retention killer).
2. No captions, or captions in italics at the bottom of the frame.
3. Visible lip-sync drift, a looped base clip, or an avatar whose hands never move in 90 s.
4. Hook footage that contradicts the ritual (Coke over ginger as the hook for an oats recipe; coffee spoken, oats shown).
5. Split-screen B-roll unrelated to the topic (dancing schoolgirls under an endocrinologist claim).
6. Studio polish: blurred backdrop, even key light, styled props, a blazer in a magazine kitchen, scrubs as costume.
7. A recipe shot held past 3 s, or a stock end card held for minutes to inflate watch time.
8. AI composites held 35 to 57 s, AI body replacements, a jar pasted over the presenter's torso.
9. Machine-translation and TTS artefacts left in the captions or the read.
10. Left/right splits of two portrait clips.
11. Hidden-camera, caregiving or illness clips of real people used as shock openers (MemoGuard AD 4, 5, 16, 18): low feel scores and a trust cost.
12. A finished, measured recipe shown on screen: the winners tease the steps and never complete them; that is the VSL's job.

Study these twelve first

The ads the notes and the syntheses rate highest for what we are building. Each opens its full note.

Every ad studied

Click a still for the full note (timed shot list, hook anatomy, transcript) and all stills.

16-09-Google-Drive 15 ads

22-09 15 ads

31-07-Google-Drive 15 ads

ADS-WL---Slimjara-Google-Drive 50 ads

CRIATIVOS-Google-Drive 15 ads

Criativos-Google-Drive 15 ads

Melhores-Ads-Google-Drive 5 ads

MemoGuard---Ads-Google-Drive 28 ads

criativos-14-09-Google-Drive 15 ads

revisados-Google-Drive 15 ads