Short‑form doesn’t reward logos. It rewards the prediction you’ll drive action.

Looking for a short-form video ad framework that actually works across Reels, TikTok, and YouTube Shorts? HERO‑PROC turns Grab → Keep → Convert into a filmable sequence—clear hooks, one idea per shot, visible proof, a risk‑reversing CTA, and pacing that raises predicted action so campaigns scale.

What drives delivery (simplified auction value)

Bid / Cost Cap × Estimated Action Rate × Ad-/User-Quality

Who wins? The ad with the highest expected value for that specific person — to maximize user experience and revenue.

“Preference” is the model’s expectation: where your ad is most likely to drive the desired action versus competitors.

On Meta you’ll see proxies such as Conversion rate ranking, Quality ranking, and Engagement rate ranking. Your biggest lever remains creative: a clear first 2 seconds, one idea per shot, visible proof, and one risk‑reversing CTA.

Key terms (plain language)

  • Bid / Cost Cap: your max willingness to pay per result; sets the ceiling for auction value.
  • Estimated Action Rate (EAR): a platform’s predicted probability that a viewer will take your target action.
  • Ad-/User‑Quality: experience score (hype/clarity/speed/feedback). Proxies: Quality, Engagement, Conversion rate rankings.
  • Cold vs warm: cold = new to your brand; warm = engaged past visitors/buyers.
  • Link CTR (cold): link clicks ÷ impressions on cold audiences (not CTR All).
  • LP CVR (cold): landing‑page conversion rate on cold traffic (first session).
  • Hold ratio: Meta: ThruPlays ÷ 3‑sec plays; TikTok: Completions ÷ 2s views.
  • CAPI/Pixel: server + browser signals so the model learns who actually converts.

HERO‑PROC is the shot order that operationalizes Grab → Keep → Convert: Hook, Empathy, Reveal, Outcome demo, Proof, Risk reversal, One CTA, and Cadence. Think human psychology meets platform comfort.

The Attention Engine: Grab → Keep → Convert (and how HERO‑PROC implements it)

No attention → no data. No data → no delivery.

Grab — the thumb‑stop in the first 2–3 seconds.

Job: Stop the scroll with one on‑screen promise, readable on mute. Show the product in hand. (HERO‑PROC: Hook + first line of Empathy, supported by Cadence.)

Ask: Would a stranger ‘get it’ in two seconds, on mute?
Goal: One visible promise + product = early positive signal.

Keep — where belief forms.

Job: Let belief form with one idea per shot: Empathy → Reveal → Outcome → Proof. (Riding on Cadence.)

Ask: Does each shot carry only one idea?
Goal: Reduce cognitive load and show the result; a strong completion/hold ratio signals consistency.

Convert — where belief turns into one clear action.

Job: Remove the last fear (risk reversal) and make exactly one next step visible. (HERO‑PROC: Risk reversal → One CTA.)

Ask: Which promise would make a skeptic try today, and which single button do they press?
Goal: One low‑risk action that feels inevitable.

How platforms “think” (the useful version)

Every impression is a prediction. Early watch behavior (scroll stops, 3‑second views) upgrades your ‘predicted action’ score — Grab. Note: Estimated Action Rate (EAR) is Meta’s name for this prediction; TikTok and YouTube Shorts use analogous predicted‑action/ranking signals built from similar inputs. Mid‑watch (50% holds, completions) signals relevance — Keep. Clean post‑click signals (CAPI/Pixel) teach who actually converts — Convert. A parallel ‘experience tax’ penalizes clickbait tone, slow pages, or negative feedback with higher CPM.

Want the psychology behind the first 2 seconds and why still images sometimes win? See: Scroll psychology: image vs video.

Terms you’ll see: Predicted Action score (e.g., EAR), Ad/User Quality, FTIR (First‑Time Impression Ratio), Cadence (1.5–2.5s cuts, one idea per shot).

HERO‑PROC: the shot order that implements Grab → Keep → Convert

Hook → Empathy → Reveal → Outcome demo → Proof → Risk reversal → One CTA → Cadence

Think of it as choreography for the feed: human psychology meets algorithm comfort.

Hook 0–2s · Grab
  • Job Stop the thumb with a single on‑screen promise, readable on mute.
  • Psych Pattern interrupt, curiosity gap, Von Restorff.
  • Examples “Install in 60 seconds. No tools.” / “Refill in seconds. Less plastic.”
  • Ask If it’s muted and I blink, do I still get it?
Empathy 2–5s · Keep
  • Job Name their annoyance in one line.
  • Psych Self‑reference, reactance reduction.
  • Example “I was done with flimsy gadgets and complicated routines.”
  • Ask Will they nod and think “that’s me”?
Reveal 5–8s · Keep
  • Job Category + one core promise.
  • Psych Uncertainty reduction, processing fluency.
  • Example “Refillable home cleaner — powerful, low‑waste.”
  • Ask One breath: product + one benefit?
Outcome demo 8–18s · Keep
  • Job Show, don’t tell.
  • Psych Vividness, mental simulation, dual coding.
  • Shots Unbox → snap refill → spray & wipe. On‑screen: “30s setup” • “Streak‑free mirror”.
Proof 18–24s · Keep → Convert
  • Job One number, one label, or one review snippet visible on screen.
  • Psych Social proof, authority, specificity.
  • Examples ★ 4.8/5 (12,431) • EU‑made • Ships in 24h • From €0.29 per use.
Risk reversal 24–28s · Convert
  • Job Remove the last fear.
  • Psych Loss aversion, zero‑risk bias.
  • Example “30‑day trial. Not a fit? Money back.”
One CTA 28–33s · Convert
  • Job Exactly one next step.
  • Psych Hick’s law, implementation intention.
  • Example “Shop now →”
Cadence throughout
  • Job Make it easy to follow; never ask the brain to hold two ideas at once.
  • Guardrails Cut every 1.5–2.5s; big captions; one claim per shot.
  • Ask Could I follow this on mute while stirring pasta?
  • 0–2s Hook on screen + product in hand.
  • 2–5s One‑line annoyance (Empathy) with big caption.
  • 5–8s Reveal: what it is + single promise.
  • 8–18s Outcome demo: show result; overlay one claim.
  • 18–33s Proof → risk reversal → one CTA on screen.

Building for scale (as a rhythm, not a checklist)

You earn attention honestly: a clear promise on screen, product in hand. You show the use and the result so belief can form. You let one number or label do the trust work. You make trying it risk‑free. You end with one next step. The platform sees stop → watch → act—without backlash—and routes you to more people like them.

When fatigue creeps in (it will), don’t rewrite the universe. Change the first two seconds and keep the spine of the story. Don’t confuse the model with daily micro‑edits—batch improvements so it can learn who this is for. Scaling isn’t shouting louder; it’s helping the system find more of the right people because your ad makes their feed better, not worse.

“Would a total stranger understand my promise in 2 seconds, on mute?”

If no, rewrite the first frame. Everything else is polishing a car with no engine.

Metrics that matter (and how to debug without guessing)

Performance doesn’t “mysteriously” tank; it degrades at specific points in Grab → Keep → Convert. Use this identity as a map:

CPA ≈ CPM / (1000 × CTR × CVR)

Benchmarks

Grab (HERO‑PROC: Hook + Empathy)
  • Attention Hook / 3s i % of impressions reaching 3s. Early signal you stopped the scroll fast. ≥ 30% 40%+ excellent
Keep (HERO‑PROC: Reveal → Outcome → Proof)
  • Hold Hold ratio i Meta: ThruPlays ÷ 3‑second video plays. TikTok: Completions ÷ 2s views. Higher = clearer pacing and relevance. ≥ 30% 40%+ excellent
Convert (HERO‑PROC: Risk reversal → One CTA)
  • Bridge Link CTR (cold) i Share of viewers who click to site. Bridge from belief to intent. ≥ 1.0–1.5% 2%+ excellent
  • Trust LP CVR (cold) i Landing page conversion rate for cold. Speed, clarity, low risk. ≥ 2–4% 5%+ excellent

Optional freshness check: monitor FTIR (First‑Time Impression Ratio) in cold sets. Low FTIR often signals recycled eyeballs → refresh hooks.

Where to find these in Ads Manager

  • Hook / 3s: Video metrics → 3‑second video plays ÷ impressions.
  • Hold ratio: Meta: ThruPlays ÷ 3‑sec plays; TikTok: Completions ÷ 2s views.
  • Link CTR (cold): Columns → Performance (or customize): Link clicks ÷ impressions.
  • LP CVR (cold): From analytics/checkout events (Pixel/CAPI) ÷ sessions from ads.
  • FTIR (cold): Breakdown → First‑time impressions share in cold sets.

Weekly debug flow (tell a short story, not a checklist)

  • CPM ↑, CTR & CVR steady → quality tax. Fix tone, add one proof number, mirror ad headline on LP.
  • CTR ↓ → first‑2‑seconds crisis. Keep the spine; reshoot only the opener.
  • CVR ↓ with steady CTR → page friction. Restate promise, show one proof, one low‑risk try.
  • Spend collapses with caps → caps too tight. Re‑establish delivery, then re‑introduce cost cap near 7‑day median CPA.

Common mistakes → exact remedies

  • Two claims in one shot → split across two shots.
  • Vague proof (“top quality!”) → one specific number or recognized label.
  • Sensational tone → dignified, factual language (esp. 50+).
  • Muddy audio → clip‑on mic or quiet room; normalize.
  • Daily micro‑edits → batch changes every 24–48h; let the model learn.

Quick self‑audit (before you publish)

Grab Is the promise readable at second two, on mute, with the product visible?
On‑screen promise by 2s; product in frame; readable on mute.
Keep One idea per shot; where’s the single proof?
One idea per shot; show outcome; one visible proof label/number.
Convert Which fear did I remove; what one step do I ask for?
State risk reversal; one CTA on screen; no competing buttons.
Cadence Could someone follow this while stirring pasta?
Cuts 1.5–2.5s; big captions; no competing ideas.