Video Hook Lines Ad Creative

Video Hook Lines

Shared quality bar: ../references/output-standard.md. All numbers cited here live in ../references/thresholds.md. The hook is three things happening at once and they are judged together. A strong spoken line over a shot that shows nothing is a weak hook.

Use this skill when

A video ad needs its opening built or rebuilt: what is said, what is shown as text, and what is on screen inside limits.video_hook_seconds. Also use it when watch time collapses in the first seconds while the rest of the video holds. The angle and the audience temperature are given, not chosen here. Not for the primary text that sits above the video in the feed, which is paid-social-primary-text-ad-creative. Not for Google search assets, which is search-headline-set-ad-creative. Not for producing one angle across all three funnel stages at once, which is funnel-stage-variants-ad-creative.

Required input

• The angle as one sentence, and the audience temperature: cold, warm or retargeting.
• Placement and aspect ratio, against coverage.ratio_set.
• Sound assumption: does this play with sound on or off by default.
• What footage or assets actually exist. Shots that must be produced are flagged, not assumed.
• The claims available with their sources, and any brand or legal constraint.
• Existing hook performance if there is any: retention curve, thumbstop rate.

Analysis workflow

1. Confirm the angle and temperature in one sentence each. A cold hook and a retargeting hook for the same angle are different jobs and are not ranked against each other in the same list.
2. Name the moment of recognition: the specific thing the viewer sees or hears that tells them the ad is about their problem. Write it before writing any line.
3. Write spoken lines that can be delivered inside limits.video_hook_seconds at normal speaking pace, which is limits.hook_spoken_words for the spoken line. Count words and syllables, never characters.
4. Write the on-screen line for each hook so it stands alone with sound off, and check it does not simply transcribe the spoken line.
5. Specify the opening shot for each hook and mark whether the footage exists, needs_data where the asset library was not supplied.
6. Check the vertical crop. A hook whose on-screen text or subject falls out of frame at 9:16 fails at that placement, per coverage.ratio_set.
7. Rank the hooks and state the ranking basis explicitly as a judgement, not a measurement, labelled hypothesis per reporting.no_causal_read_from_export.
8. Apply claims.evidence_rule, claims.superlative_rule and claims.restricted_topics to spoken and on-screen lines alike, marking unsourced factual lines [claim: needs source].

Decision rules

• Spoken line cannot be delivered inside limits.video_hook_seconds: rewrite, and cut the qualifier, never the subject.
• On-screen line that only repeats the spoken line: rewrite. Two channels saying one thing wastes the sound-off viewer.
• Opening shot with no existing footage: keep the hook, label it approval_needed on a production cost, and rank it separately from hooks that can ship this week.
• Hook that only works for a viewer who already knows the brand, submitted as cold: reclassify it as warm rather than repairing it.
• Any line this skill produced that carries [claim: needs source] is approval_needed, whatever else its verdict says, and the marker is never stripped to make the set look finished. A marked line is not shippable copy: it waits for the named source, or it is rewritten without the claim. A marked hook is ranked outside the shippable set, never inside it.
• Existing hook underperforming, read on retention: judge the opening with early.hook_rate, which is 3-second plays over impressions, and the body with early.hold_rate, which divides video plays at its own pinned percentage over 3-second plays. Both are read against the median of the video's own early.duration_band and never against a median mixing lengths, and neither fires unless the band holds at least early.median_stability_floor creatives. Neither may fire until the creative clears early.min_exposure on both its halves, impressions and spend against one target cost per acquisition. Both are low by definition and neither is reported as a result.
• Under early.min_exposure: state the retention read is not usable and rank on judgement only. Do not reach for test.click_floor here. That floor is the CTR-judged fallback for a test being read on clicks, and a click count says nothing about whether the opening held a viewer; using it would answer a different question in the right-looking notation.
• early.hook_rate failing while early.hold_rate passes: rewrite the opening only. The reverse, hold failing while hook passes, is a body problem and this skill hands it back rather than rewriting an opening inside limits.video_hook_seconds that already worked.

Output format

No verdict line. A ranked table: rank, spoken line, on-screen line, opening shot, footage exists yes or no, sound-off readable yes or no, evidence label, decision label. Then one paragraph naming the ranking basis. Then What this could not see, Missing data, Approval gates.

This skill writes and ranks openings rather than reading a test or a time series, so it issues no high confidence label. The comparison family's high rests on test.min_runtime and test.volume_floor, conditions of a test that has not been run on these hooks. Where an existing hook's retention is read, that read sits in the early.* block and is low by definition, and the ranking itself is a judgement labelled hypothesis, which carries no confidence label at all.

Practical example

Illustrative and fictional. Advertiser "Placewise". Angle: "your ad budget is being spent on the wrong placements". Cold audience, 9:16 reels, sound off assumed. Six hooks requested. Existing footage: 4 clips of a screen recording.

Shippable, ranked 1 to 3. Each is spoken line, on-screen line, opening shot:

1. "Your budget went somewhere you never chose." / "You did not pick these placements" / the placement breakdown scrolling under the cursor. Footage exists.
2. "This ad worked. Look where it ran." / "It worked. Now look where." / split frame, the ad on the left, one placement row highlighted on the right. Footage exists.
3. "Nobody opens the placement report. Watch." / "The report nobody opens" / cursor clicking through to the breakdown tab. Footage exists.

All three sit inside limits.hook_spoken_words and inside limits.video_hook_seconds at normal pace, and each on-screen line stands alone with sound off rather than transcribing the spoken one.

Ranked outside the shippable set:

4. "We found four placements doing nothing." [claim: needs source]. The count came from a screenshot nobody attached, so it is approval_needed and waits for the export.
5. A hook opening on a founder face to camera. Kept, but there is no such footage, so it is approval_needed on a shoot.

Declined, not written in any form: "Most advertisers waste 60% of their budget." The figure has no source, claims.evidence_rule bites, and it is not softened into "most advertisers waste a big share" either. It would be written the day a named study or the account's own export is supplied.

Retention read: refused. The supplied data covers 640 impressions on the existing hook with spend under one target cost per acquisition, so early.min_exposure is not met and neither early.hook_rate nor early.hold_rate may fire. Video length was also not supplied, so under early.duration_band there is no band to take a median inside. The ranking above is therefore judgement, labelled hypothesis. The click count was offered as a substitute gate and is declined: test.click_floor answers a CTR question and this is a retention one.

Guardrails

• Never rank a hook you have not checked with sound off.
• Never write an opening that depends on footage nobody has confirmed exists.
• Never invent a statistic, a study, a customer or a comparison.
• Never present a judgement ranking as a measured result.
• Do not edit, upload or publish video assets.
• Report every hook declined and why, including hooks requested by name.