mardab96/tiktok-ads-claude-skills

Hook Rate Scorer - TikTok Ads

Use this skill when

Use this when you want an attention read on how an ad opens before anyone writes new scripts. Two kinds of evidence are joined: export math (who watched past 2 and 6 seconds) and a read of the actual opening (what the viewer saw and heard).

Common requests:

• "What's my hook rate on these ads?"
• "Which of these openings is losing people?"
• "Score these three scripts before we shoot."
• "Write me a hook test plan for this winner."

When not to use

• Delivery is repeating and CTR is sliding over weeks -> creative-fatigue-detector-tiktok-ads.
• Hook is fine but clicks do not convert -> landing-page-match-tiktok-ads or cpa-spike-diagnosis-tiktok-ads.
• The user wants full new scripts -> winning-ad-remix-tiktok-ads.

Required input

Ad-level export, aggregated over the date range (daily rows are not required). Label the range in the filename. Use Ads Manager Reporting, ad level, and add the video-view columns; column labels vary, so keep them as labelled in your Ads Manager and map them in the first reply.

Field	Why it is needed
ad id, ad name	Row identity
campaign objective, ad group name	Comparable-set grouping
Audience type (prospecting or retargeting), as stated by the user	Retargeting viewers open differently
Format: Spark or brand-handle, video length, placement	Comparable-set grouping
impressions	Denominator for both view rates
2-second video views, 6-second video views	Hook and hold numerators
clicks, spend	Post-hook read

Also needed: the exact first spoken line (word for word), the first frame described, any on-screen text in seconds 0 to 6, and the product facts the opening claims. Optional: other video-view columns your export carries (watch-time or completion columns, as labelled), creative previews, comments.

A generic video views column is not a 2-second column. If only that exists, mark T02 needs_data and say so.

Metric definitions

All are calculated per ad from the same row, so they share one denominator. Label every one a proxy: TikTok does not publish a metric named "hook rate" in this pack's references.

Metric	Formula	Reads as
2s view rate	2-second views / impressions	Thumb-stop proxy: share of impressions that held for 2 seconds
6s view rate	6-second views / impressions	Early interest proxy
6s-to-2s ratio	6-second views / 2-second views	Diagnostic ratio of differently defined metrics, never viewer retention; it may exceed 1
Click per 6s view	clicks / 6-second views	Post-hook click intent proxy [heuristic]

Autoplay, sound state, and placement change how a "view" is counted by viewers, so compare only inside a comparable set.

Analysis workflow

1. Verify impressions and the exact exported metric definitions. TikTok 6-second views can include an early interaction or completion of a shorter video; 2-second views require two seconds of playback. A 6s count above 2s is therefore valid and must not stop analysis. See https://ads.tiktok.com/help/article/video-play?lang=en . Check missing denominators and incompatible report scopes instead.
2. Build the comparable set: same objective, same audience type, same format (T02). Drop ads under 3,000 impressions from the set (T02). State the set size. A median over two or three ads is fragile; cap the verdict at possible and say why.
3. Compute each ad's 2s rate, 6s rate, diagnostic 6s-to-2s ratio, and click per 6s view. Mark zero denominators undefined. Take the median 2s rate across the comparable set.
4. Apply T02: flag an ad when its 2s rate is at least 25% below the comparable median, after at least 3,000 impressions. Gap = (median - ad rate) / median.
5. Score the opening with the rubric below, from the transcript and first-frame description, not from the metrics. Score blind to the metrics where possible, then compare.
6. Cross the two reads using the quadrant table in Decision rules.
7. Check delivery differences before blaming the script: audience type, placement mix, ad age, learning window (T05), budget skew, sound-on share if known.
8. Write a controlled hook test: one ad, 3 openings, everything after second 2 identical.

Opening rubric (house heuristic, not a predictor)

Score each 0 absent, 1 partial, 2 clear. Maximum 10.

#	Criterion	Clear means
C1	Subject in first frame	Product, problem, or person visible at frame 1; no logo card, fade, or black frame
C2	Tension line	A spoken or on-screen line states a problem, question, claim, or surprise
C3	Speed	The tension line lands before second 2; count words (about 2 to 3 spoken words per second [heuristic])
C4	Visual change	A cut, motion, or hand action inside the first 2 seconds
C5	Sound-off read	On-screen text carries the point, sits inside the safe area, readable on a small screen

Bands: 0 to 4 weak, 5 to 7 workable, 8 to 10 strong [heuristic]. Separate claim check: every claim in the opening must trace to a supplied product fact, else hold on reusing that hook.

Scoring scripts that have not run yet

When no export exists (pre-shoot, or a brand-new ad), run the rubric alone. Each candidate needs its first spoken line, first frame, and any on-screen text from seconds 0 to 6. Tag the evidence creative_script.

• Score every candidate side by side, 0 to 10, and count the spoken words before second 2.
• The score orders ideas to test. It is not evidence of performance, so the verdict ceiling is possible.
• Recommend the top two plus one deliberately different opening, so the test spans a spread instead of three near-copies.

Illustrative candidates for a silicone food-lid set (fictional):

Candidate	First line	First frame	C1-C5	Score	Note
A	"Hi, we're Lumo Home"	Logo card	0,0,0,1,1	2	Greeting before the problem
B	"Silicone lids always pop off."	Bowl held upside down	2,2,2,2,1	9	Tension in 5 words
C	On-screen only: "Do these stay on?"	Hand pressing a lid	2,1,1,2,2	8	Works with sound off

Partial exports

What you have	What you can compute	Verdict ceiling
Impressions, 2-second views	2s rate, T02 gap	possible on cause, gap confirmed
Impressions, 2s and 6s views	Adds diagnostic 6s-to-2s ratio, not retention	Same
Only a generic video views column	Nothing for T02	needs_data
Impressions and clicks only	CTR, not a hook read	needs_data for the hook

Decision rules

T02 flags a gap in attention. It never supports a purchase conclusion.

Condition	Verdict	Next step
Gap at least 25% (T02), 3,000 or more impressions, rubric 0 to 4, set is like-for-like	Gap: confirmed for the supplied export. Opening as cause: likely	Hook test, then winning-ad-remix-tiktok-ads
Gap at least 25% (T02), rubric 8 to 10	possible delivery or audience difference	Check placement, audience, sound, learning (T05)
2s rate fine, diagnostic 6s-to-2s ratio below the set median	possible metric-definition, interaction or duration difference	Compare duration and interaction counts; inspect seconds 2 to 6 before proposing edits
2s and diagnostic ratio comparable, click per 6s view low	possible CTA or offer mismatch	landing-page-match-tiktok-ads
2s rate high, clicks and conversions weak	possible curiosity without payoff	Compare opening claim with page and offer
Under 3,000 impressions, or set too thin	needs_data or monitor	Wait or widen the set
2-second column missing	needs_data	Re-export with the video-view columns

Output format

### Comparable set
Objective: [..] | Audience: [..] | Format: [..] | Ads in set: [n] | Median 2s rate: [x%] | T02 line: [y%] [ads_export]

### Hook scorecard
| Ad | Impr | 2s rate | 6s rate | Diagnostic 6s-to-2s ratio | Gap vs median | First spoken line | First frame | Rubric C1-C5 | Verdict | Next hook |
|---|---|---|---|---|---|---|---|---|---|---|

### Hook test plan
Control: [ad, retained] | Variants: V1 [..], V2 [..], V3 [..] | Held constant: [body, CTA, audience, budget] | Read after: 3,000 impressions per variant (T02) | Metric: 2s rate; 6s-to-2s ratio as context, never retention

### Decision table
| Finding | Evidence | Verdict | Severity | Confidence | Business impact | Next step |
|---|---|---|---|---|---|---|

### Data gaps and approval needed
[missing item] -> [verdict prevented]. [live change] -> [named owner].

Practical example

Illustrative numbers only. Prospecting purchase ads, in-feed vertical video, six ads each past 3,000 impressions. 2s rates: A 31%, B 28%, C 33%, D 30%, E 26%, F 19%. Sorted median = (28% + 30%) / 2 = 29%; T02 line = 29% x 0.75 = 21.75%.

Ad F: 12,400 impressions, 2,356 2-second views (19.0%), 1,178 6-second views (9.5%), diagnostic 6s-to-2s ratio 0.50 against a comparable-set median near 0.61, 112 clicks. This ratio does not measure retention. Gap = (29 - 19) / 29 = 34.5%, so F clears T02. Opening: a "Hi, we're Lumo Home" logo card for 1.5 seconds, greeting line, product appears at second 3. Rubric: C1 0, C2 0, C3 0, C4 1, C5 1 = 2 of 10, weak. Claim check passes.

Verdict: the gap is confirmed for this export; the opening as the cause is likely, severity medium, impact conversion only as an attention input. Hook test: V1 problem-first line over the product in frame 1; V2 on-screen question with a hand demo; V3 result shown first, then the line. Body and CTA identical.

Common failure modes

• Comparing a retargeting ad with prospecting ads; warm audiences stop and watch for different reasons.
• Comparing a 9-second ad with a 45-second ad on 6s view rate.
• Reading a high 2s rate as a win when clicks and conversions are weak; the hook may promise something the offer does not.
• Mixing Spark and brand-handle ads in one set without saying so (spark-vs-brand-ad-split-tiktok-ads).
• Scoring a script from memory instead of from the first 2 seconds as filmed.
• Declaring a hook weak from an ad still in learning or under 3,000 impressions.
• Treating the rubric score as a forecast; it orders ideas to test, nothing more.

Guardrails

• Do not score a hook from fewer than 3,000 impressions or from a median built on two or three ads without capping the verdict at possible (T02).
• Do not substitute a generic video views column for 2-second views.
• Do not compare across objective, audience type, or format without labelling the mismatch.
• Do not write a claim in a new hook that is absent from the supplied product facts.
• Do not present a hook gap as a conversion finding or a sales prediction.
• Do not change a live ad or budget; stop at a test plan awaiting approval.
• Do not present rubric bands or the diagnostic 6s-to-2s ratio as platform benchmarks or viewer retention.